| Licence or credential required | None. No licence, board exam or registration gates technical writing in the US, UK, EU, Canada or Australia, and nobody can stop you calling yourself a technical writer or taking paid work tomorrow. A bachelor's degree appears as preferred on most corporate postings and as a hard filter on some government and defence requisitions, but the subject rarely matters: working technical writers come out of engineering, support, QA, teaching, nursing, linguistics, journalism, physics and the trades at least as often as out of technical communication programmes. The Society for Technical Communication offers the Certified Professional Technical Communicator credential, and it is honest to say it is rarely decisive: it occasionally satisfies a contractor or public sector requisition that wants a named certification, and almost never changes a hiring manager's mind. Free and cheap training that does earn its time: Google's technical writing courses, the Diataxis framework documentation, the Write the Docs community material, and the Google, Microsoft and Red Hat style guides, which are the three style guides most software employers actually work from. Two real gates exist in specific sectors, and neither is a writing credential. For US defence and a lot of aerospace work the gate is a security clearance plus export control eligibility under ITAR and EAR, which in practice means US person status and an employer willing to sponsor the clearance. For medical device and pharmaceutical work the gate is demonstrable familiarity with the quality system you would be writing inside, typically ISO 13485, ISO 14971 for risk, IEC 62366 for usability and the labelling and instructions-for-use requirements of the market the device is sold into. |
|---|---|
| What actually gates the job | The writing samples, then the exercise. Samples are the screen: a technical writing application with no samples, or with samples that are marketing prose when the job is procedures and reference, is the most common silent rejection in this field. Hiring managers read two or three pages of your actual output and decide from that, which is why the candidate who cannot produce shareable work because everything they wrote sits behind a customer login loses to a weaker writer with three merged pull requests on an open-source documentation set. The exercise is the decision: most often an edit of a deliberately poor page, a short interview with a subject matter expert followed by a write-up, or a reference page written from an OpenAPI spec or a changelog entry. The third gate is quieter and decides seniority rather than pass or fail: whether you can run the procedure you documented, in the actual product, and say what you found when the documented steps did not work. |
| How long it takes to become hireable | From a standing start with no samples, three to nine months of deliberate work is realistic, and the variable is whether you are producing documentation for somebody else's product or reading about documentation. Developer documentation is the fastest route to shareable evidence, because open-source projects accept documentation contributions from strangers and a merged pull request is public, attributable and dated: three to six months of weekly contributions plus one self-directed sample, such as a working quickstart for a public API that currently has a bad one, is enough to start getting interviews. Structured authoring for hardware, aerospace and defence takes longer to enter cold because the tools are expensive and the content is controlled, so the usual way in is a staffing agency contract at a tier-one supplier, and people who already know the equipment get hired fastest. Medical device and quality documentation is usually a lateral move from inside a regulated company, often from QA, manufacturing, clinical or regulatory affairs, and that path can be weeks rather than months because the employer already trusts you with the quality system. Software help centre and knowledge base work is the easiest lateral move from technical support, and support agents who have written the top fifty articles in their own company's knowledge base are already doing the job and should say so on the resume. |
| Titles this job is actually posted under | Searching only for "technical writer" hides most of the market and most of the better-paid half of it. Real postings for this work are titled technical writer, senior technical writer, staff technical writer, lead technical writer, documentation specialist, documentation engineer, docs engineer, developer documentation engineer, developer educator, technical content developer, information developer, information architect, content developer, API documentation specialist, knowledge base specialist, knowledge management specialist, technical publications specialist, publications engineer, technical illustrator, technical editor, documentation manager, content operations manager, SOP writer, procedure writer, policy and procedure analyst, labelling specialist, regulatory labelling associate, medical device technical writer, quality systems documentation specialist, validation documentation specialist, instructional designer, training content developer, curriculum developer, proposal writer, bid writer, systems documentation analyst, business process analyst, UX writer, content designer and technical marketing writer. Set alerts on the sector and the artefact (API reference, service manual, instructions for use, standard operating procedure, knowledge base) rather than on the job title. |
| Where the volume actually is | Aerospace and defence, including primes and the tier-one and tier-two supply chain, which is the single steadiest employer of technical writers in the US and UK and the one most software-focused applicants never look at. Medical devices, diagnostics and life science instruments. Semiconductors, test and measurement, and industrial hardware. Industrial automation, robotics, energy, grid and nuclear. Enterprise infrastructure and developer platforms: cloud, data, security, payments, observability, developer tooling and the API-first companies whose documentation is effectively their product surface. Automotive, including service information and the driver-facing documentation that comes with assisted driving features. Telecommunications and network equipment. Financial services and insurance, where the work is procedures, controls, policy and model documentation rather than manuals. Government agencies and the contractors who serve them. Pharmaceutical and biotechnology, although the writing roles there are usually titled medical writer or regulatory writer and are a related but separate profession. AI companies themselves, which need API documentation, prompting and integration guides, model and system documentation and evaluation write-ups, and have been hiring writers rather than replacing them. The thin ends are generalist content roles at small software companies, where documentation has often been folded into a product or support job, and basic how-to libraries for simple consumer software. |
| How you are paid, and where to check | Point at the sources rather than at a number from a forum thread. In the United States the authoritative wage series is the Bureau of Labor Statistics Occupational Employment and Wage Statistics data for SOC 27-3042, Technical Writers, which has its own code and its own national, state and metropolitan wage distributions, so read your own metro area first-hand rather than a national median. The neighbouring codes worth reading are 27-3043 (Writers and Authors), 27-3041 (Editors) and, for documentation engineer roles that sit inside engineering, the software developer codes, because a documentation engineer at a developer platform company is often paid on an engineering band and reported there. O*NET's entry for 27-3042 is also worth reading because many employers build their job descriptions from it. In the UK use the Office for National Statistics Annual Survey of Hours and Earnings under the writing and editorial occupation codes, and read the Institution of Engineering and Technology and recruiter salary surveys as directional only. Canada's National Occupational Classification has a dedicated technical writer code and Job Bank publishes provincial wage ranges against it; Australia's ANZSCO likewise lists technical writer as its own occupation. Then read postings: pay transparency rules in a growing list of US states and cities mean a meaningful share of ads now carry a real band, so read fifty for your city, level and sector rather than one. Three things move the number more than seniority does. An active security clearance commands a visible premium in defence work and the premium rises with the level of clearance. Contract work is often quoted hourly through a staffing firm, where you must compare the hourly rate net of unpaid time and benefits against a salary rather than multiplying by 2080. And the structured authoring and regulated specialisms pay above the generalist market because the supply of writers who can work in DITA, S1000D or a validated quality system is genuinely thin. |
| Tools employers actually name | Which tools appear tells you which market a posting belongs to, so read the tool list before the responsibilities. Docs-as-code: Git, GitHub or GitLab, Markdown and MDX, AsciiDoc, reStructuredText, Docusaurus, MkDocs with Material, Sphinx, Antora, Hugo, Mintlify, ReadMe, Redocly, GitBook, plus Vale and markdownlint for automated style and link checking in continuous integration. API documentation: OpenAPI, AsyncAPI, JSON Schema, GraphQL schemas, protobuf, Postman, Swagger UI, Redoc, Scalar. Help centre and knowledge base: Zendesk Guide, Salesforce Knowledge, Intercom Articles, Document360, ServiceNow knowledge, Confluence, Notion, Guru. Help authoring and structured authoring: MadCap Flare, Adobe FrameMaker, Oxygen XML Editor, Arbortext, RWS Tridion Docs, Heretto, Paligo, Ixiasoft, the DITA Open Toolkit, and in aerospace an S1000D common source database. Standards you will be asked about by name: DITA, S1000D, ASD-STE100 Simplified Technical English, the ATA specifications in aviation, the MIL-STD publication specifications in defence, the ISO/IEC/IEEE 265xx documentation standards, IEC 82079 for instructions for use, WCAG and Section 508 for accessibility of documentation. Graphics and media: Snagit, Camtasia, Figma, Illustrator, Visio, Lucidchart, Mermaid and PlantUML for diagrams in source control, and for hardware work SolidWorks Composer, Creo Illustrate or Arbortext IsoDraw for exploded assembly views. Localisation: Trados, memoQ, Phrase, Smartling, Crowdin, Lokalise, and machine translation post-editing as a named task. Analytics and feedback: GA4, search analytics inside your own docs site, Pendo, and support ticket data from Zendesk, Jira Service Management or Salesforce. Delivery: Jira, Azure DevOps, Linear. Assistants and AI docs infrastructure: ChatGPT, Claude, Gemini, GitHub Copilot, Cursor, docs answer engines such as Kapa.ai, Inkeep or Algolia Ask AI, Model Context Protocol servers that expose documentation to agents, and the llms.txt and plain-Markdown endpoint conventions. Name the ones you have used and say at which step, because a list of thirty tools reads as having used none of them deeply. |
| Typical hiring timeline and what the test is | Software and developer platform employers: two to six weeks across a samples screen, a recruiter call, a hiring manager conversation, a timed or capped writing exercise and a panel that usually includes an engineer, a product manager and sometimes a support lead. Aerospace, defence and industrial contractors: three weeks to four months, and the long tail is clearance processing and export control checks rather than interviews, with a shorter interview loop than software and a practical test on a real manual extract. Medical device and regulated manufacturing: four to ten weeks, more reference checking, a document control and quality system conversation, and frequently a redlining exercise on a procedure. Contract roles through a staffing firm: often days, with a recruiter screen, a thirty-minute call with the publications or documentation lead and a short writing test, and these move fast enough that a slow applicant simply misses them. The exercise itself is almost always one of five things: edit a deliberately bad page and list what you changed and why; interview an engineer for twenty to thirty minutes and write the procedure from it; write one reference page from a specification, a changelog or a pull request; restructure a sprawling existing page into task, reference and concept; or review a machine-generated draft and mark what you would not publish. |
Five technical writing markets, and the tool list tells you which one a posting is
Technical writer is one job title covering at least five hiring markets that assess different evidence, use incompatible toolchains and sit in different parts of a company. The applicant who does not know this spends months sending a Docusaurus site to an aerospace publications manager and a service manual extract to a developer platform team, and gets silence from both. Decide which market you are entering for the next two years, build the evidence that market reads, and keep a second sample set for the market you want next.
Developer documentation is the market most people picture and the most competitive one. You write quickstarts, conceptual guides, API and SDK reference, CLI documentation, tutorials, migration guides, release notes and error message catalogues, usually in Markdown inside the same repository as the code, shipped through pull requests and a continuous integration pipeline. You are assessed on whether a developer can get to a working call using only your page, on whether you understand the product well enough to say which path is the recommended one, and on whether you can read a specification, a diff or a pull request without needing an engineer to translate it. The toolchain is the giveaway: if a posting names Git, OpenAPI, Docusaurus, MkDocs or Vale, this is the market.
Software end-user documentation and knowledge base work is the largest market by headcount and the least discussed. You write help centre articles, in-product help, onboarding content, admin guides, troubleshooting trees and release notes, usually in Zendesk, Salesforce Knowledge, Document360 or a help authoring tool rather than in Git. You sit near support rather than near engineering, your success is measured in tickets that did not get raised, and the single most valuable thing you can show is that you know which fifty articles carry most of the volume and why. Technical support agents are the natural recruits here, and the move is often internal.
Structured authoring for hardware, aerospace, defence and heavy industry is the market with the steadiest demand and the thinnest supply of qualified writers. You write operation and maintenance manuals, installation instructions, wiring and schematic documentation, illustrated parts catalogues, service bulletins and interactive electronic technical manuals, in DITA or S1000D, inside a content management system with a formal review and approval cycle, often in Simplified Technical English so that a non-native-speaking technician in another country reads it the same way you wrote it. The content is reused across variants and configurations rather than rewritten, which means you are really doing information architecture with words attached. These jobs pay above the generalist market, they are frequently clearance-gated, and a large share of them are filled through staffing agencies.
Regulated and quality documentation is a related corner with a different centre of gravity. Medical device instructions for use, labelling, user interface text that forms part of a usability engineering file, design history file documents, standard operating procedures, work instructions, validation protocols and reports, batch records, policy and controls documentation inside a bank. The content is an accountability artefact: it is audited, it is signed, it is version controlled in a document management system, and a wrong step is a safety and legal problem rather than a bad reading experience. People get hired here on familiarity with the quality system rather than on writing flourish, which is why writers who learn one regulated domain tend to stay employed in it for a decade.
The fifth market is the adjacent band that technical writers move into and out of, and it is worth naming so you do not mistake it for the same job. Instructional design and training content development. Technical marketing: whitepapers, solution briefs, reference architectures. Proposal and bid writing, which is compliance and deadline work with its own certification ladder. Content design and UX writing, which is interface text inside a design tool. Medical and regulatory writing for pharmaceutical submissions, which is a separate profession with its own credential culture and usually wants a life sciences background. Each one hires on different evidence from documentation work, and applying to all five with one resume is how a good writer ends up with no interviews.
- Read the tool list first. Git and OpenAPI mean developer docs; Zendesk means support content; DITA or S1000D means structured authoring; a quality management system means regulated work.
- Developer docs: assessed on whether a developer reaches a working result from your page alone.
- Help centre: assessed on whether you know which articles carry the ticket volume.
- Structured authoring: assessed on reuse, standards fluency and whether you can survive a formal review cycle.
- Regulated: assessed on the quality system, document control and traceability, not on prose style.
- Pick one for now. A second sample set exists for the market you want next, and you send it only in the right room.
What shrank, what grew, and what just got renamed
The honest picture of this market in 2026 is neither collapse nor boom, and candidates lose interviews by arriving with the wrong one. Documentation headcount at small and mid-size software companies has genuinely thinned, because the cheap half of the work (restating the interface, writing a how-to for every button, keeping a help centre superficially current) is the half a product manager with an assistant can now do badly enough to defer hiring. At the same time, demand at the top of the field went up, because documentation stopped being a cost centre at the end of the release and became the surface that both customers and machines read a product through. Fewer, more senior writers who own a system, and more writers in regulated and hardware industries where nothing about the obligation changed.
The clearest growth is in documentation that is treated as infrastructure. Companies whose product is an API now staff documentation like they staff a service: an owner, a build pipeline, automated link and style checks, a review gate in the definition of done, an analytics loop, and increasingly a retrieval layer that answers questions from the docs. The title that carries this work is often documentation engineer or docs engineer rather than technical writer, and it is a real distinction rather than inflation: the posting expects you to work in the repository, fix the build, write a script, read the specification and own the information architecture, and it usually pays on an engineering band. If you can do that, apply under that title, because fewer writers do and the postings get less traffic.
The second area of growth is the oldest part of the field. Aerospace, defence, medical devices, energy, rail and industrial equipment have not reduced their documentation obligations by a single page, and in several of those sectors the obligation grew. A technician still needs a maintenance procedure that is correct at the torque value. A device still ships with instructions for use that are part of its regulatory file. A plant still runs on standard operating procedures that an auditor will read. This work is less visible than software documentation because it is not published on the open internet, it is advertised through staffing firms as often as through careers pages, and it is where a writer who wants a decade of steady employment should be looking.
What got renamed is worth tracking because it changes how you search. A lot of end-user documentation work is now posted as knowledge management, content operations, support content or customer education. A lot of interface text work is posted as content design or UX writing. A lot of internal documentation work is posted as enablement, developer experience or platform documentation. And a growing number of postings fold documentation into a hybrid role: developer advocate, solutions engineer, support engineer, product operations. Those hybrids are real jobs and often the way in, but read the split honestly before accepting, because a job that is twenty percent documentation and eighty percent something else will not build the portfolio you need next.
One structural feature of this field surprises people arriving from software: a large share of good technical writing jobs are contracts, and that is not a signal of a weak employer. Aerospace, defence, medical device and industrial employers staff documentation against programme milestones, so they hire six, twelve and eighteen month contracts through staffing firms, and experienced writers in those sectors spend much of their career on them, often converting to permanent at one employer and then contracting again. The practical implication is that you should be registered with the staffing firms that actually serve your sector, because a meaningful number of these roles never appear on a public job board at all.
- Shrank: generalist documentation headcount at small software companies, and basic how-to libraries.
- Grew: documentation treated as infrastructure at API-first companies, and everything in regulated and hardware industries.
- Renamed: documentation engineer, knowledge management, customer education, content design, developer experience, platform documentation.
- Contract work is normal in this field, especially in aerospace, defence and medical devices. Register with the staffing firms that serve your sector.
- If you can work in a repository and fix a build, apply under documentation engineer. Fewer writers do, and the band is higher.
Who hires steadily, and how to find the postings that are not on a job board
The searcher question that matters most for a long career is which industries still hire technical writers reliably through a soft market, and the answer is consistent: the ones where documentation is required by somebody other than the marketing department. Where a regulator, a standard, a safety case, a maintenance programme, a contract clause or a certification audit requires the document, the writing job survives downturns, because the document is a deliverable rather than a nice-to-have. Where documentation exists to help customers feel supported, the job is discretionary and it moves with the budget.
Aerospace and defence is first on that list. The primes hire directly and continuously, and the larger market is the tier-one and tier-two supply chain: avionics, engines, landing gear, actuation, radar, sensors, munitions, ground support equipment, simulation and training. Work is organised around publication specifications and configuration control, and entry is usually either a staffing contract or a junior technical publications role at a supplier. The two gates are export control eligibility and clearance. If you are a US person without a clearance, say so clearly and apply anyway, because sponsorship for a secret clearance is routine at this level and the posting language often reads as if it is not.
Medical devices, diagnostics and life science instruments hire steadily and churn less. The work is instructions for use, labelling, user interface text that is part of the usability file, service manuals for field engineers, training material and the procedures that run the quality system itself. Employers screen on whether you have worked inside a controlled document process, so if you have ever had a document routed for electronic signature in a system such as Veeva, MasterControl, Windchill or an equivalent, name the system. Expect translation to be central, because a device sold across the European Union ships documentation in many languages and that drives everything about how you write.
Enterprise infrastructure and developer platforms are the healthiest part of the software side: cloud providers, databases, data platforms, security, identity, payments, observability, developer tooling and API-first companies. Documentation there is competitive surface area, and the budget holder is often engineering rather than marketing, which is why the roles survive marketing cuts. Semiconductors, test and measurement, industrial automation, robotics, energy and grid, nuclear, rail, automotive service information, telecommunications and network equipment all hire continuously and quietly, usually against programme need. Financial services and insurance hire procedure, policy, controls and model documentation writers, often inside risk or operations rather than inside a content team. Government agencies and their contractors hire against contract vehicles and are worth monitoring directly.
Finding these postings takes a different search habit from software job hunting. Set alerts on the artefact rather than the title: search for S1000D, DITA, instructions for use, service manual, standard operating procedure, technical publications, OpenAPI and knowledge base as keywords, with and without the word writer. Read the careers pages of the ten largest employers in your metropolitan area regardless of industry, because a writer job at a pump manufacturer is invisible on a tech job board. Register with the staffing firms that actually place in your sector, which in US engineering and life sciences means firms such as Actalent, Aerotek, Kelly, Yoh, Belcan, Insight Global, TEKsystems, Aquent and Robert Half, and tell the recruiter precisely which document types you can produce. Join the communities where these jobs get mentioned before they are posted: Write the Docs, the Society for Technical Communication chapters, the DITA and S1000D user groups, and the developer community of whatever product you want to document.
One more route that works and is underused: document the product before they hire you. Pick an employer you want, find the page in their public documentation that is visibly wrong, incomplete or three releases out of date, rewrite it properly, and send it to the person who owns documentation with one paragraph saying what was wrong and what you changed. This converts at a rate that no amount of applying does, it is legitimate rather than a gimmick, and in open-source-adjacent companies you can simply open the pull request.
- The reliable rule: if a regulator, standard, contract or audit requires the document, the writing job survives downturns.
- Aerospace and defence: apply even without a clearance if you are export-control eligible. Sponsorship at secret level is routine.
- Medical devices: name the controlled document system you have worked in, by name.
- Search for artefacts and standards, not titles: S1000D, DITA, instructions for use, service manual, SOP, OpenAPI, knowledge base.
- Register with sector staffing firms and tell them the document types you can produce.
- Rewrite a broken page in a target employer's public docs and send it to whoever owns docs. It converts.
How hiring actually works, who screens, and what each stage is really testing
Technical writing is screened by fewer people than a software job and by different people than most applicants expect. At a software company the first human read is usually a recruiter who checks for the toolchain keywords and the presence of samples, then the hiring manager, who may be a documentation lead but at smaller companies is a head of product, engineering, support or developer relations with no writing background. At an aerospace or industrial employer it is HR against a requisition that was written from a contract requirement, then a technical publications manager who has been doing this for twenty years and will test you on the specification. At a medical device company the decisive voice is often quality or regulatory affairs rather than a writing manager. At a staffing firm it is a recruiter who has a live requirement and will submit you within hours if your resume matches the document types.
Stage one is the samples screen and it is where most applications die. Two or three pages get read, quickly, for a small number of things: is this the kind of document the job requires, is it correct, is it organised so a reader can find one thing without reading all of it, and did this person write it. A beautiful conceptual essay loses to a plain procedure when the job is procedures. A link that is broken, gated, or made of screenshots from a product the reviewer cannot see is a rejection without a reply.
Stage two is a recruiter or hiring manager call, usually thirty minutes, and the real content is fit to the market rather than your writing. Expect to be asked what you have documented, who read it, where it lived, how it got reviewed, how you got information out of engineers, and how much of the product you used yourself. Have a two-sentence answer ready for the question that trips up experienced writers: what was the hardest thing you had to understand, and how did you work it out. A candidate who answers that with a real mechanism, in the product's own vocabulary, has effectively passed.
Stage three is the exercise and it decides the offer. The five common forms are worth practising in advance because they are predictable. First, the edit test: a page deliberately full of passive constructions, undefined terms, buried steps, missing prerequisites and inconsistent naming, with an instruction to improve it and list your changes. The list is what is being graded, not the prose. Second, the subject matter expert interview: twenty to thirty minutes with an engineer, then a write-up, which tests whether you ask about failure modes, prerequisites and the parts users get wrong rather than taking dictation. Third, write one reference page from a specification, a changelog or a pull request, which tests whether you can read source material. Fourth, restructure a sprawling page into task, reference and concept, which tests information architecture. Fifth, and increasingly common, review a machine-generated draft and mark what you would not publish, which tests judgment about correctness. Capped take-home exercises in this field are usually two to four hours; if one is open-ended and large, ask what the cap is, because a good employer has one.
Stage four is a panel, which exists in software hiring and rarely elsewhere. The engineer on the panel is checking whether talking to you will cost them time, so answer technical questions with the mechanism rather than the metaphor, and say clearly when you do not know something and what you would do to find out. The product manager is checking whether you will fight the right battles and ship on the release date. A support lead, if present, is checking whether you will fix the pages their tickets come from. In aerospace, defence and industrial work the equivalent stage is a practical review with the publications manager, often with a real manual extract in front of you, and the test is specification compliance and reuse rather than opinion about style.
Two practical notes on the process. Contract roles move in days rather than weeks, so if you want that part of the market, answer recruiters the same day and have your resume tailored per document type ready to send. And clearance processing in defence work can add months after the offer, during which you may be put on unclassified work; ask early what you will be doing while the clearance is in process, because the answer varies enormously between employers.
- The samples screen is the gate. Send the document type the job requires, not your best-written piece.
- Prepare the question nobody prepares for: the hardest thing you had to understand, and how you worked it out.
- For an edit test, the list of changes with reasons is what is graded. Write it even if they do not ask.
- For a subject matter expert interview, ask about failure modes, prerequisites and what users get wrong.
- Ask for the time cap on a take-home. A good employer has one, usually two to four hours.
- Contract roles close in days. Permanent defence roles can take months after the offer because of clearance.
Writing samples: what to send, and what to do when everything you wrote is behind a login
Almost every technical writer hits the same wall: the work that proves you can do this job belongs to an employer, sits behind a customer login or a firewall, and is covered by a confidentiality agreement. That wall is not an excuse that hiring managers accept, because other candidates solved it, so solve it deliberately rather than apologising for it in a cover letter.
Send three to five samples, not twenty, and choose for coverage rather than for polish. The set that passes in most markets is: one procedure with prerequisites, numbered steps and a verification step at the end; one reference page, ideally API, parameter, error code or parts reference, where the test is completeness and consistency; one conceptual or explanatory piece that shows you can describe a system's model rather than its buttons; one before-and-after edit with margin notes on what you changed and why; and one piece that shows you structured content for retrieval, which in 2026 is the sample almost nobody sends and the one that gets remembered. Add a single page in front of the set with one line per sample: audience, what the document had to achieve, what constraint you worked inside, and what you wrote versus what you edited. Reviewers read that page.
If your work is public, link it and be precise about authorship, because reviewers assume the worst about unattributed documentation. Say "I wrote the installation and upgrade sections of this guide and rewrote the troubleshooting tree; the API reference is generated from source" rather than linking a documentation set and hoping. If a page has a Git history with your commits in it, say so and link the history, because provenance is now a real asset.
If your work is not public, there are five legitimate routes and all of them beat a sample you cannot show. First, open-source documentation contributions: find a project you actually use, fix the documentation issue nobody wants, get it merged, and you have a public, dated, attributable artefact with a review trail. This is the single best investment for a developer documentation candidate, and projects with a documentation label on their issue tracker are abundant. Second, write a complete quickstart for a public API that has a bad one, end to end, and include the working code and the error you hit on the way. Third, sanitise: rebuild a document you wrote with a fictional product name, scrubbed numbers and no customer identifiers, and label it clearly as a sanitised rewrite, which is accepted practice when it is declared. Fourth, document something physical: write the maintenance procedure for a bicycle hub, an espresso machine or a 3D printer, with real torque values and real part numbers from the real service documentation, because that sample proves more about hardware writing than an essay does. Fifth, if you are moving from support, ask your employer for permission to use two knowledge base articles you wrote; many say yes for customer-facing content that is already public.
Do not send: anything confidential, anything you cannot discuss in detail, a resume disguised as a sample, marketing copy when the job wants procedures, a sample that is mostly screenshots of a product the reviewer cannot see, a document with somebody else's template and no indication of what you contributed, or a PDF that has been exported from a help authoring tool and lost its structure. And do not send a twenty-page document when a reviewer will read two pages: send the two best pages with a note saying where they sit in the whole.
One format decision matters more than it looks. If the job is docs-as-code, your samples should look like they came from a repository: Markdown or AsciiDoc rendered on a real site, a visible Git history, a build. If the job is structured authoring at an industrial employer, a clean PDF extract that follows a specification is exactly right and a GitHub link reads as irrelevant. Sending the wrong format is the most common self-inflicted rejection in this field, and it is the easiest to avoid.
- Three to five samples: procedure, reference, concept, before-and-after edit, and one that shows retrieval-aware structure.
- Put a one-page index in front: audience, purpose, constraint, and what you wrote versus edited, per sample.
- State authorship precisely. Unattributed documentation is assumed to be someone else's or generated.
- No public work: merged open-source documentation pull requests, a real quickstart for a public API, a declared sanitised rewrite, or a hardware maintenance procedure.
- Match the format to the market: a repository and a rendered site for docs-as-code, a specification-compliant extract for industrial work.
- Never send anything you cannot discuss line by line in the interview.
The resume, the cover letter and the application: what gets read and what gets skipped
A technical writing resume is read for evidence of four things, in this order: what you documented, who read it, where it lived and how it shipped, and what changed because of it. Everything else is skipped. The adjectives at the top (excellent communicator, detail-oriented, passionate about clarity) are the first thing a documentation lead's eye slides past, partly because every applicant has them and partly because a writer who describes themselves as an excellent communicator in a resume summary has already failed a small test.
Write the products, not the duties. "Documented a payments API: 40 endpoints, three SDKs, quickstart plus migration guide, in Markdown in the service repository, shipped through pull request review in the same release train as the code" tells a hiring manager more than a paragraph of responsibilities. For hardware, the equivalent is "authored and maintained operation and maintenance manuals for a line of industrial pumps in DITA, 1,200 topics, reused across nine configurations, translated into eleven languages". Scale, document type, toolchain and delivery mechanism are the four facts a reviewer is hunting for, and most resumes omit three of them.
Put numbers where you genuinely have them and describe the shape where you do not. Real and checkable numbers in this field: topic or page counts, number of releases shipped into, number of languages, number of products or configurations, support ticket volume on the area you rewrote before and after, search success rate inside the docs site, time for a new developer to reach a first successful API call, reduction in translation cost after a terminology or reuse project, audit findings closed. If you do not have the number, say the shape honestly: "rewrote the top twenty troubleshooting articles, which were the source of most password-reset tickets" is credible; an invented percentage is not, and a documentation lead who has run the same analytics will spot it.
Name the toolchain specifically and keep it honest, because this is where the keyword filter actually bites. List the authoring format, the publishing system, the version control, the ticketing system, the standards and the localisation tooling. Distinguish what you have used in production from what you have tried, because a writer who claims S1000D and cannot explain a data module code will be found out in ten minutes. If the posting names a tool you do not know, say in one line what you do know that is adjacent: Flare if you know RoboHelp, DITA if you know AsciiDoc and content reuse, Oxygen if you know XML editing anywhere.
Formatting rules that matter in this field more than most: one page per decade of work, plain structure, real section headings, no two-column layout that a parser scrambles, no skills matrix with star ratings, no graphics that carry information. A technical writer whose resume is hard to skim has made an argument against themselves. Link your samples near the top rather than at the bottom, and make the link work without a password, because a reviewer will not email you for access.
The cover letter is worth writing in this field, which is not true everywhere, because documentation hiring managers read for judgment and a short letter can demonstrate it. Four sentences: which market you are in and why this posting is it, one specific observation about their documentation that proves you read it, the single most relevant thing you have done, and what you would look at first. The observation is the whole letter. "Your quickstart assumes an API key is already provisioned, and the provisioning steps are two pages away in the admin guide" gets a reply at a rate nothing else does.
Two application mechanics worth knowing. Government and defence applications are often routed through systems that require the resume to restate the requisition's language explicitly, and omitting a stated requirement means an automatic screen-out even if you clearly have it, so mirror the requirement wording in your own words. And in contract markets, the same role is frequently represented by multiple staffing firms, so track who submitted you where, because duplicate submissions to the same client get both withdrawn.
- Lead with products documented, audience, toolchain and delivery, not with adjectives.
- Real numbers in this field: topics, releases, languages, configurations, ticket volume on the area you rewrote, time to first successful API call, translation cost, audit findings closed.
- Separate production experience from exposure in the tool list. Overclaiming a standard gets caught in minutes.
- Plain single-column formatting, headings, working sample links near the top.
- Cover letter: one specific, accurate observation about their documentation beats three paragraphs about passion.
- In defence and government applications, mirror the stated requirement language or be screened out for something you have.
Pay: where to look instead of guessing, and what actually moves the number
Technical writing pay varies more by industry and specialism than by years of experience, which is unusual and exploitable. Two writers with eight years each can be a long way apart because one documents a consumer app and the other documents a flight control computer under clearance, and the gap is not about writing ability. Before you negotiate anything, read the authoritative series for your country rather than a forum consensus.
In the United States, start with the Bureau of Labor Statistics Occupational Employment and Wage Statistics data for SOC 27-3042, Technical Writers. It is one of the few writing occupations with its own code, which means the data is unusually clean, and it is published by state and metropolitan area with a full wage distribution rather than only a median. Read your own metro rather than the national figure, because the spread between metros in this occupation is large. Then read O*NET's 27-3042 entry, because a surprising number of employer job descriptions are assembled from it and knowing the task list tells you what the requisition means. If you are targeting documentation engineer roles inside engineering organisations, also read the software developer codes, because those roles are often reported and paid there instead.
Outside the US: the Office for National Statistics Annual Survey of Hours and Earnings covers the UK under the relevant writing and editorial occupation codes, Canada's National Occupational Classification has a dedicated technical writer code with provincial wage ranges published on Job Bank, and Australia's ANZSCO lists technical writer as its own occupation with pay data available through government labour market sources. In all of these, government data lags the market by a year or more, so use it to establish the shape and then calibrate against live postings.
Live postings are now a real source rather than a rumour. Pay transparency laws in a growing list of US states and cities require a band in the advertisement, and many UK and EU employers publish one voluntarily, so read fifty postings for your city, level, industry and document type and look at the distribution rather than the top of it. Recruiter salary guides from staffing firms are free, directional and published by the people who profit from the number, which is worth remembering but does not make them useless for relative comparisons between specialisms.
Four things move the number materially. An active security clearance in defence work commands a visible premium, and it rises with level, which is why writers in that sector guard clearance continuity carefully. Regulated and structured authoring specialisms (DITA, S1000D, medical device labelling, validation documentation) pay above the generalist market because the supply of writers who can do them is thin. Engineering-adjacent documentation engineering roles pay on engineering bands. And industry pays more than function: the same writing skill is worth more at a semiconductor company than at a nonprofit, and moving industry is usually a faster raise than moving title.
On contract rates, do the arithmetic properly. An hourly rate through a staffing firm covers no unpaid holiday, often thinner benefits, and gaps between contracts, so compare it to a salary by estimating billable hours you will actually work in a year and adding the cost of the benefits you are replacing, not by multiplying by 2080. Ask whether the contract is W2 through the firm or incorporated, whether there is a conversion clause and at what point, and whether overtime on a programme crunch is paid. Experienced contract writers in aerospace and medical devices negotiate those terms as a matter of routine, and new entrants do not know they are negotiable.
- US: BLS OES SOC 27-3042, Technical Writers, read by metro area with the full distribution. Then O*NET 27-3042.
- UK, Canada, Australia: ONS ASHE, the NOC technical writer code with Job Bank wage ranges, and ANZSCO.
- Read fifty live postings with transparency bands for your city, level and industry. Look at the distribution.
- Premiums that are real: clearance level, structured authoring and regulated specialisms, engineering-band documentation roles.
- Changing industry usually pays better than changing title.
- On contract rates, ask about W2 versus incorporated, conversion terms and paid overtime, and never compare by multiplying the hourly rate by 2080.
The interview, the exercise, and your first ninety days
The interview for a technical writing job tests five things and only one of them is writing. Can you learn a complex thing fast and explain it at the altitude the audience needs. Can you get information out of a busy engineer, technician or clinician who does not want a meeting. Do you have judgment about structure, which means knowing when a reader needs a task, a reference, a concept or a tutorial, and not blending all four into one page. Will you actually verify what you wrote by doing it. And can you take and give editorial feedback without it becoming personal, because documentation review is relentless and a writer who flinches costs the team more than a slower writer who does not.
Prepare specific answers to the questions that come up in almost every loop. Walk me through how you documented something you did not understand at the start. How do you get time from an engineer who ignores you. Tell me about a document you were wrong about and what happened. How do you decide what not to document. What do you do when the product is bad and the documentation is being asked to cover for it. How do you keep documentation current through a fast release cycle. What would you look at in your first month here. Each of those wants a mechanism and an example, not a principle.
There is one answer that distinguishes senior candidates immediately, and it is about verification. Say that you run the procedure. Describe a time the documented steps did not work, what you found, and what you did about it, including filing the bug rather than writing around it. Writers who document from a specification without touching the product produce documentation that is plausible and wrong, which is the exact failure mode that machine generation also produces, and every experienced documentation manager is now screening for it harder than they used to.
Ask questions that reveal whether the job is doable, because a documentation role in a company that does not value documentation will exhaust you in a year. Who owns documentation, and who signs off. Is documentation in the definition of done, or does it arrive after release. Where does the content live and who can edit it. How do writers get access to the product, and will I have a licence, a test environment or a device. What happened to the last person in this role. How is documentation measured here, and what happens if the number moves. How much of the backlog is new content versus fixing existing content. The answers to those seven questions predict your experience better than the salary does.
For the first ninety days, resist the instinct to rewrite everything. The pattern that earns trust fastest is narrow and measurable. In the first two weeks, use the product as a new user would and keep a log of every place you got stuck, which is the most valuable document you will ever write at that company and has a shelf life of about a month. In weeks two to four, find the evidence: the ten pages with the most traffic, the ten questions that generate the most support tickets, the search queries inside the docs that return nothing, and if there is an assistant answering from your documentation, the questions it answers wrongly. In weeks four to eight, fix the top ten, properly, and show the before and after. In weeks eight to twelve, fix one piece of the system rather than the content: a style check in the build, a template, a terminology list, a review step that engineers will actually use, or a release note process that does not depend on you chasing people.
Two habits to establish immediately because they are hard to add later. Keep a decision log of terminology and structure choices, with the reason, because you will be asked to re-litigate every one of them and the log ends the argument in thirty seconds. And build one relationship with an engineer or technician who will answer a question in a direct message, because in every documentation job the single biggest determinant of your output is how quickly you can get an answer from someone who knows.
- Five things tested: learning speed, extracting information from experts, structural judgment, verification, and handling review.
- Say that you run the procedure, and tell the story of the time the documented steps did not work.
- Ask the seven questions: ownership, definition of done, where content lives, product access, what happened to the last person, how documentation is measured, and the new-versus-fix split in the backlog.
- First two weeks: use the product as a new user and log every place you got stuck.
- Weeks four to twelve: fix the top ten pages by evidence, then fix one thing about the system rather than more content.
- Start a terminology and structure decision log on day one.
What a technical writer has to know about AI in 2026-27
This is the role where the AI question has a real answer rather than a slogan, and it is not mainly about drafting. The structural change is that your documentation acquired a second audience that reads everything, remembers nothing between questions, and cannot ask a follow-up. Product documentation is now the corpus that an in-product assistant, a support copilot, a coding agent and the general-purpose assistants answer from, which means a user's experience of your documentation is increasingly an answer synthesised from three paragraphs of it rather than a page they read. A technical writer who understands that and can design for it is worth noticeably more in 2026 than one who does not, and this is the single most useful thing to be able to talk about in an interview.
Concretely, designing for retrieval changes how you write in ways you can demonstrate. Sections have to be self-contained, because a retrieval system will hand a reader one chunk without its surroundings: that kills "as described above", "in the previous section" and a procedure whose prerequisites live on a different page. Headings have to be literal and question-shaped rather than clever, because they are what gets matched. One task per page, with the task named in the title. Terminology has to be consistent to the word, because a concept split across three synonyms is a concept that retrieval fragments and answers badly. Nothing load-bearing can exist only inside a screenshot or a diagram, so every image needs the fact stated in text as well. Tables need headers that survive being flattened. Error codes, parameters, limits and defaults belong in complete, explicit tables rather than in prose, because that is the content assistants get right and humans search for anyway. Most of these were already good practice for human readers and for search; what changed is that the cost of ignoring them went up sharply.
The machine-readable layer is now part of the job in developer documentation specifically, and knowing the vocabulary puts you ahead of most applicants. Plain Markdown versions of pages served alongside the rendered HTML, so an agent can fetch the source rather than scraping a React page. A "copy page as Markdown" affordance. The llms.txt convention, which is a proposed community convention rather than a ratified standard, with uneven adoption, and worth describing accurately as such rather than as a requirement. A Model Context Protocol server that exposes documentation and reference material so an agent can read it directly inside a developer's tooling, which a growing number of developer platform companies now ship and which documentation teams are often asked to own. And the quality of the OpenAPI specification itself, because that single file is what an agent builds against, which makes descriptions, examples and error enumerations in the spec a writing deliverable rather than an engineering afterthought.
Measurement changed with it, and this is where a candidate can sound current in one sentence. Documentation pageviews have fallen at many companies while the product got no easier to use, because the question that used to be a page visit is now a question to an assistant. A writer who reports pageviews as their success metric in 2026 sounds out of date. The metrics that have replaced them: whether the assistant answers a defined set of real user questions correctly from your documentation, support ticket volume on the area you rewrote, search queries inside the docs that return nothing, time for a new user to reach a first successful result, and deflection where it is measured honestly rather than as a proxy for cost cutting. The specific artefact worth building, and worth mentioning in an interview, is an evaluation set: a list of fifty real questions with the correct answer, run against whatever assistant answers from your docs, scored, and used to find the documentation gap rather than to tune a prompt. The insight that lands is that when an assistant answers wrongly, the fix usually belongs in the documentation.
On drafting, be honest and specific about what you automate. First drafts of release notes from merged pull requests and changelogs. Converting an engineer's rambling explanation or a recorded call into a structured outline. Generating the first pass of a reference table from a specification. Finding inconsistent terminology across a thousand topics. Drafting the tedious variants of a procedure across configurations. Suggesting edits for reading level, sentence length or Simplified Technical English compliance. Translating and then post-editing, which is now standard practice rather than a shortcut. Summarising a long specification so you know what to ask about. None of that is controversial, and saying it plainly is better than pretending you write every word from scratch.
What has not moved, and it is the centre of your value: nothing in this toolchain can be accountable for a procedure being correct. A model will produce a confident maintenance step with a wrong torque value, a quickstart that calls an endpoint that was removed two releases ago, an instruction for use that omits a contraindication, and a troubleshooting step that bricks the device, and in each case the failure is invisible because the prose looks finished. Several employers tried replacing documentation with generation, got documentation that was plausible and wrong, and discovered the cost in support tickets, field failures and in regulated industries in audit findings. The writer's job absorbed the verification: you run the procedure, you check the value against the source, you ask the engineer the question the draft glossed over, and you are the person who signs. Say that in an interview and mean it.
Where AI has changed this job less than the hype suggests is worth saying plainly, because the hype is loudest exactly where it is least true. In aerospace, defence, medical device and other structured authoring environments, the constraints are publication specifications, configuration control, validated tool chains, translation memory and a signed review cycle, and none of those yield to a better draft. Assistants help at the edges there: terminology checking, Simplified Technical English compliance, finding reuse candidates across a topic library, drafting a change summary. The core work of deciding what a technician needs at a maintenance step, in a format that a specification dictates, inside a system that records who approved it, is substantially the job it was five years ago. If you interview for that market, do not arrive claiming transformation. Arrive able to say exactly which steps you have automated and which you have not, and why.
Two cautions that come up in real hiring conversations. On confidentiality: most employers of technical writers now have a written AI policy, and unreleased product detail, customer data, controlled technical data under export regulations and anything inside a validated quality system are exactly the material that must not be pasted into a consumer tool. Asking which tools are approved marks you as someone who has worked in a real company, and in defence work getting this wrong is a reportable incident rather than an embarrassment. On detection: automated AI detectors are unreliable in both directions and some employers use them anyway, so your protection in a writing exercise is provenance rather than protest. Work somewhere with version history, keep your notes, and be ready to open the document and show the mess.
Structuring documentation so a retrieval system can answer from it
Your pages are now read in fragments by systems that cannot ask a clarifying question. A section that depends on its neighbours, a procedure whose prerequisites live elsewhere, a concept named three different ways, or a critical value that exists only inside a screenshot will produce a confidently wrong answer to a real user, and the user will blame the product rather than the documentation. This is the most concrete, least hand-wavy AI skill in technical writing and it is the one hiring managers at developer platform and support-heavy companies are actively looking for.
Show it: Bring one sample that is explicitly built this way and say so: self-contained sections, literal question-shaped headings, one task per page, prerequisites restated rather than cross-referenced, every fact in a screenshot also stated in text, complete error and parameter tables rather than prose. In the interview, take their actual documentation and name two things you would change and the failure each one causes. A candidate who says "your rate limit is only stated in a diagram, so an assistant answering about rate limits will guess" has demonstrated the whole skill in one sentence.
Owning terminology as a system, not a style preference
Consistent naming used to be a quality nicety; it is now load-bearing. Retrieval splits a concept that has three names, translation memory and machine translation both degrade on inconsistent terms, and a support assistant trained on documentation that calls the same object a workspace, a project and an environment will answer about the wrong one. Terminology discipline is also the cheapest large improvement available in most documentation sets, which makes it a strong first-ninety-days project.
Show it: Bring a terminology list or glossary you actually maintained, with the chosen term, the rejected synonyms and the reason. Describe how you enforced it: a Vale or lint rule in the build, a review checklist, a glossary in the documentation itself. If you have a number, use it: terms consolidated, translation cost change, inconsistencies found across a topic library. If you have not done this at work, do it for a public documentation set and show the audit.
Building and running an evaluation set against your own documentation
If an assistant answers from your documentation, then the documentation has a measurable correctness property and almost nobody is measuring it. A writer who shows up with the habit of collecting real user questions, running them against the assistant, scoring the answers and treating a wrong answer as a documentation bug is offering something a documentation team usually does not have and increasingly knows it needs. It also reframes your work as engineering rather than opinion, which changes how your proposals are received internally.
Show it: Describe the artefact concretely: fifty real questions pulled from support tickets, community forums and docs search queries that returned nothing, each with the correct answer, run against the assistant, with the failures triaged into missing content, ambiguous content and content the retrieval could not reach. Say what you fixed and what happened. Even a small version done on a public documentation set is a strong interview exhibit, because it proves the instinct rather than the scale.
Making documentation machine-readable: specs, Markdown endpoints and docs servers
In developer documentation the agent has become a first-class consumer, and the quality of the OpenAPI specification, the availability of plain-Markdown source, and whether there is a Model Context Protocol server exposing the reference are now part of whether a developer can build against the product at all. These are documentation deliverables that sit in the gap between writing and engineering, which is exactly where a documentation engineer earns an engineering band. Most applicants cannot discuss any of it.
Show it: Name the pieces accurately and with the right level of confidence: descriptions, examples and complete error enumerations in the OpenAPI spec as a writing responsibility; Markdown served alongside rendered pages and a copy-as-Markdown affordance; llms.txt as a community convention with uneven adoption rather than a standard; a documentation MCP server as something some developer platform companies now ship. If you have shipped any of it, lead with that. If you have not, say which one you would propose first for their documentation and why.
Reviewing generated drafts, and verifying by doing
The failure mode of generated documentation is not bad grammar, it is plausible wrongness: a parameter that does not exist, a removed endpoint, a step in the wrong order, a value that is right for a different model of the device. That is harder to catch than bad writing because the prose looks finished, and it is the specific risk employers are hiring a human to carry. Verification is also the clearest single line between a writer who will cost a team money and one who will save it.
Show it: Describe your review pass as a sequence rather than a feeling: check every value, flag, parameter and limit against the source of truth; run the procedure in the product or on the device; confirm prerequisites and permissions; check that the stated outcome is what actually happens; delete anything you could not verify rather than softening it. Then tell the story of a time the documented steps failed, what you found and that you filed the bug instead of writing around it. Include a before-and-after sample where the before is a generated draft and your margin notes say what you would not publish.
Writing for translation and for Simplified Technical English, with machine translation in the loop
Machine translation with human post-editing is now the default for most multilingual documentation, which makes the source text the control variable: short sentences, one idea each, consistent terms, no idiom, no ambiguous pronouns, explicit subjects. Hardware, medical device and industrial employers ship documentation in ten or more languages and the cost and risk of that pipeline is decided by how you wrote the English. Writers who understand this are valuable and comparatively rare.
Show it: Say which standard or style you write to, name the translation management system and the terminology base you worked with, and give a concrete example of a sentence you rewrote because it would translate badly. If you have worked in Simplified Technical English, say so plainly, because it is a hard requirement on a lot of aerospace postings and a filter most candidates cannot pass. If you have post-edited machine translation output, describe what you look for first.
Automating the mechanical parts of documentation delivery
The difference between a writer and a documentation engineer is often a handful of automations: style and terminology checks in continuous integration, link checking, drafting release notes from merged pull requests, generating reference tables from a specification, flagging pages that reference a deprecated parameter, finding reuse candidates across a topic library. Each is small, each removes work you would otherwise do forever, and together they are the strongest argument that you will make the documentation set better rather than just bigger.
Show it: Name specific automations you have shipped and what they caught: a Vale rule set and the terms it enforces, a broken-link check in the build, a script that diffs the API specification between releases and lists what needs documenting. If you are not technical, the honest version is still strong: describe a manual process you turned into a checklist, a template or a recurring report, and say which part you would want help automating.
What a screen is looking for
These are the terms that a resume screen, human or automated, is matching against for this role. Use the ones that are true of you, in the words the posting uses.
- Technical writer
- Technical writing
- Senior technical writer
- Documentation specialist
- Documentation engineer
- Developer documentation
- API documentation
- OpenAPI
- Swagger
- REST API reference
- SDK documentation
- Docs as code
- Git
- GitHub
- GitLab
- Markdown
- MDX
- AsciiDoc
- reStructuredText
- Docusaurus
- MkDocs
- Sphinx
- Antora
- Mintlify
- ReadMe
- Redocly
- GitBook
- Vale
- markdownlint
- Static site generator
- DITA
- DITA XML
- S1000D
- Structured authoring
- Content reuse
- Information architecture
- Topic-based authoring
- MadCap Flare
- Adobe FrameMaker
- Oxygen XML Editor
- Arbortext
- RWS Tridion Docs
- Heretto
- Paligo
- Ixiasoft
- DITA Open Toolkit
- Simplified Technical English
- ASD-STE100
- Technical publications
- Illustrated parts catalogue
- Interactive electronic technical manual
- Service manual
- Operation and maintenance manual
- Installation guide
- Instructions for use
- IFU
- Labelling
- ISO 13485
- ISO 14971
- IEC 62366
- IEC 82079
- EU MDR
- Design history file
- Standard operating procedure
- SOP writing
- Work instruction
- Validation documentation
- Document control
- Quality management system
- Veeva
- MasterControl
- Windchill
- Change control
- Redlining
- Zendesk Guide
- Salesforce Knowledge
- Document360
- Confluence
- Knowledge base
- Knowledge management
- Help centre content
- Release notes
- Troubleshooting guide
- Admin guide
- User guide
- Quickstart
- Style guide
- Google developer documentation style guide
- Microsoft Writing Style Guide
- Chicago Manual of Style
- Diataxis
- Task-based writing
- Plain language
- WCAG
- Section 508
- Accessibility
- Subject matter expert interview
- SME interviews
- Technical editing
- Peer review
- Editorial review
- Content strategy
- Content operations
- Taxonomy
- Metadata
- Terminology management
- Glossary
- Controlled vocabulary
- Localization
- Localisation
- Trados
- memoQ
- Phrase
- Smartling
- Crowdin
- Machine translation post-editing
- Snagit
- Camtasia
- Figma
- Visio
- Lucidchart
- Mermaid
- PlantUML
- Technical illustration
- SolidWorks Composer
- Jira
- Azure DevOps
- Agile
- Scrum
- Sprint
- Release train
- Continuous integration
- Retrieval augmented generation
- RAG
- llms.txt
- Model Context Protocol
- MCP
- Docs for AI assistants
- AI-assisted documentation
- Prompt engineering
- Evaluation set
- Documentation analytics
- Support ticket deflection
- Security clearance
- ITAR
- Export control
- Aerospace documentation
- Defence documentation
- Medical device documentation
Mistakes that cost people this job
Applying to every technical writing job with one resume and one sample set, as if developer documentation, help centre content, S1000D manuals and medical device labelling were the same job. They screen on different evidence and incompatible toolchains.
Pick the market, build the evidence it reads, and tailor per document type. Keep a second sample set for the market you want next and send it only in the right room.
Sending the wrong format. A GitHub repository of Markdown to an aerospace publications manager who works in a common source database, or a specification-compliant PDF extract to a developer platform team that lives in pull requests.
Read the tool list in the posting and match it. Rendered site plus Git history for docs-as-code, a clean specification-compliant extract for industrial and regulated work.
Applying with no samples because everything you wrote is behind a customer login or under a confidentiality agreement, and explaining that in the cover letter.
Solve it before you apply: merged documentation pull requests on an open-source project, a complete working quickstart for a public API, a clearly declared sanitised rewrite, or a real maintenance procedure for a physical object.
Documenting from the specification without ever running the procedure or touching the product. This produces documentation that is plausible and wrong, which is exactly the failure mode employers are now screening hardest for.
Run it. Then make verification an explicit part of how you describe your process, and tell the story of the time the documented steps failed, what you found, and that you filed the bug rather than writing around it.
Describing yourself as a storyteller who is passionate about clarity, at the top of a resume for a job that produces error code tables and torque values.
Lead with what you documented, who read it, where it lived, how it shipped and what changed. The adjectives are the first thing a documentation lead's eye skips.
Listing thirty tools and five standards with no indication of depth. A claim to S1000D that collapses when you are asked to explain a data module code ends the interview.
Separate production experience from exposure, explicitly. For a tool you do not know, name the adjacent one you do and say what transfers.
Reporting documentation pageviews as the success metric. At many companies pageviews fell because the question became a question to an assistant, so this reads as being two years out of date.
Talk about support ticket volume on the area you rewrote, docs search queries returning nothing, time to first successful result, and whether an assistant answers a defined question set correctly from your documentation.
Claiming you do not use AI at all, or the opposite, leading with how much documentation you can now produce. The first reads as incurious, the second tells a hiring manager you will flood them with plausible drafts they have to verify.
Name the steps you automate (release note first drafts from pull requests, reference tables from a specification, terminology audits, translation post-editing), the step you keep by hand and why, and the fact that you verify by doing.
Writing documentation that only works as a whole page: cross-references instead of restated prerequisites, "as described above", three synonyms for one concept, and the rate limit stated only inside a diagram.
Write self-contained sections with literal headings, one task per page, consistent terminology, and every load-bearing fact in text as well as in an image. A reader and a retrieval system both get the answer.
Only searching the title "technical writer", which hides most of the market including the better-paid half of it.
Search artefacts and standards: S1000D, DITA, instructions for use, service manual, SOP, OpenAPI, knowledge base. Search documentation engineer. Register with the staffing firms that place in your sector.
Ruling yourself out of defence and aerospace work because the posting asks for a clearance you do not hold.
If you are export-control eligible, apply and say so plainly. Sponsorship at secret level is routine at tier-one suppliers, and the clearance is the thing that later commands a premium.
Spending the first ninety days rewriting the whole documentation set, which produces a lot of effort and no visible result before your first review.
Log every place you got stuck in week one, find the top ten pages by traffic and ticket volume, fix those properly with a visible before and after, then fix one thing about the system: a style check in the build, a template, a terminology list, a release note process.
Taking an interview without reading the employer's own documentation.
Read it and bring two specific, accurate observations and the failure each one causes. This is twenty minutes of preparation and it changes the conversation more than anything else you can do.
Accepting a hybrid role without checking the split, then discovering the job is eighty percent support or product work and twenty percent documentation.
Ask what share of the week is documentation, who owns it, and whether documentation is in the definition of done. Hybrids are a legitimate way in, but only if you know what you are buying.
Questions people ask
Is technical writing still a good career in 2026, or is AI replacing it?
Technical writing is still a viable career in 2026, and the honest shape of the change is a redistribution rather than a collapse. The cheap half of the work thinned out: generic how-to articles for simple software, restating the interface, keeping a shallow help centre superficially current, and that is the half entry-level writers used to break in through. Demand went up at the other end, because documentation became the corpus that in-product assistants, support copilots and coding agents answer from, which made documentation structure and correctness a product concern rather than a tidiness concern. The industries that employ most technical writers, aerospace and defence, medical devices, semiconductors and industrial hardware, energy, enterprise infrastructure and government contracting, did not reduce their documentation obligations at all. The practical consequence for a technical writer entering now is that the generalist junior role is scarcer and the ways in are sideways: support, QA, open-source documentation contributions, a staffing contract in a regulated industry, or a documentation engineer role that expects you to work in the repository.
Do I need a degree or a certification to become a technical writer?
No. Nothing licenses a technical writer in the US, UK, EU, Canada or Australia: there is no board, no registration and no mandatory certificate, and a technical writer can take paid work tomorrow without one. A bachelor's degree appears as preferred on most corporate postings and as a hard filter on some government and defence requisitions, but the subject rarely matters, and working technical writers come out of engineering, support, QA, nursing, teaching, physics, linguistics and the trades at least as often as out of technical communication programmes. The Society for Technical Communication's Certified Professional Technical Communicator credential exists and is rarely decisive: it occasionally satisfies a contractor requisition and almost never changes a hiring manager's mind. The free material that genuinely helps is Google's technical writing courses, the Diataxis framework, and the Google, Microsoft and Red Hat style guides. The two real gates are sector-specific rather than academic: a security clearance plus export control eligibility for US defence work, and demonstrable familiarity with a quality management system for medical device work.
What writing samples should a technical writer submit, and how many?
A technical writer should submit three to five samples chosen for coverage rather than for polish, with a one-page index in front of them. The set that passes in most markets is one procedure with prerequisites, numbered steps and a verification step; one reference page such as an API, parameter, error code or parts reference, where completeness is the test; one conceptual piece showing you can explain a system's model rather than its buttons; one before-and-after edit with margin notes saying what you changed and why; and one piece that shows you structured content so a retrieval system can answer from it, which is the sample almost nobody sends in 2026. The index should give each sample an audience, a purpose, the constraint you worked inside, and an explicit statement of what you wrote versus what you edited, because reviewers assume the worst about unattributed documentation. Send two strong pages rather than a twenty-page document, and never send anything you cannot discuss line by line.
How does a technical writer build a portfolio when all their work is confidential?
A technical writer with no shareable work has four legitimate routes, and all of them beat an apology in a cover letter. Contribute documentation to an open-source project you actually use: find the documentation issue nobody wants, get the pull request merged, and you have a public, dated, attributable artefact with a visible review trail, which is the single best investment for a developer documentation candidate. Write a complete working quickstart for a public API that currently has a bad one, including the code and the error you hit on the way. Rebuild a document you wrote with a fictional product name and scrubbed identifiers, labelled clearly as a sanitised rewrite, which is accepted practice when declared. Or document something physical, such as the maintenance procedure for a bicycle hub or an espresso machine, with real part numbers and real values, which proves more about hardware writing than an essay does. If you are coming from support, ask whether you can use two knowledge base articles you wrote that are already public.
Which industries still hire technical writers steadily?
Technical writers are hired most reliably where somebody other than the marketing department requires the document. That means aerospace and defence, including the tier-one and tier-two supply chain, which is the steadiest employer of technical writers in the US and UK and the one most applicants overlook; medical devices, diagnostics and life science instruments; semiconductors, test and measurement and industrial hardware; industrial automation, robotics, energy, grid and nuclear; enterprise infrastructure and developer platforms such as cloud, data, security, identity, payments and observability; automotive service information; telecommunications and network equipment; financial services procedures, policy and controls; government agencies and their contractors; and AI companies themselves, which need API documentation, integration guides and system documentation. The thin end for a technical writer is generalist content roles at small software companies, where documentation has often been folded into a product or support job. A large share of the steady work is contract through staffing firms, which is normal in this field rather than a warning sign.
What does a technical writer interview actually test, and what is the writing exercise like?
A technical writer interview tests five things and only one of them is prose: whether you can learn something complex quickly and explain it at the altitude the audience needs, whether you can get information out of a busy engineer, technician or clinician who does not want a meeting, whether you have structural judgment (knowing when a reader needs a task, a reference, a concept or a tutorial rather than blending all four into one page), whether you will verify what you wrote by actually doing it, and whether you can take and give editorial feedback without it becoming personal. The exercise is predictable and worth practising: an edit test on a page deliberately full of passive constructions, undefined terms, buried steps and inconsistent naming, where the list of changes with reasons is what is graded rather than the prose; a twenty to thirty minute subject matter expert interview followed by a write-up; one reference page written from a specification, a changelog or a pull request; restructuring a sprawling page into task, reference and concept; or reviewing a machine-generated draft and marking what you would not publish. A capped take-home for a technical writer is normally two to four hours, and it is reasonable to ask for the cap if the brief does not state one. Prepare one story in particular: a time the documented steps did not work, what you found, and that you filed the bug rather than writing around it.
How did AI change what good documentation looks like?
For a technical writer the biggest change is that documentation acquired a second audience: assistants and agents that read it in fragments, remember nothing between questions and cannot ask a follow-up. That makes several previously optional habits load-bearing. Sections have to be self-contained, which rules out "as described above" and a procedure whose prerequisites live on another page. Headings have to be literal and question-shaped because they are what gets matched. One task per page, named in the title. Terminology consistent to the word, because a concept split across three synonyms is a concept retrieval answers badly. Nothing load-bearing inside only a screenshot or a diagram. Error codes, parameters, limits and defaults in complete explicit tables rather than in prose. In developer documentation there is also a machine-readable layer that a technical writer may be asked to own: the quality of the OpenAPI specification, plain-Markdown versions of pages, the llms.txt community convention, and a Model Context Protocol server exposing the reference to agents.
Has AI changed technical writing in hardware, aerospace and medical devices as much as in software?
No, and a technical writer who claims otherwise in an aerospace or medical device interview loses credibility in the first ten minutes. In those environments the binding constraints are publication specifications such as S1000D or DITA, configuration control, validated tool chains, translation memory and a signed review cycle, and none of those yield to a faster draft. Assistants help at the edges: terminology checking, Simplified Technical English compliance, finding reuse candidates across a topic library, drafting a change summary, summarising a long specification before you interview an engineer. The core of the work, deciding what a technician needs at a maintenance step, in the format a specification dictates, inside a system that records who approved it, is substantially the job it was five years ago. There is also a hard constraint that software teams do not face: controlled technical data under export regulations and content inside a validated quality system must not be pasted into a consumer tool at all, which rules out the casual workflow entirely.
How much do technical writers get paid, and where should I check?
A technical writer should check the authoritative wage series rather than a forum consensus, because pay in this occupation varies more by industry and specialism than by years of experience. In the United States the right source is the Bureau of Labor Statistics Occupational Employment and Wage Statistics data for SOC 27-3042, Technical Writers, which has its own code with state and metropolitan wage distributions, so read your own metro and the whole distribution rather than a national median; O*NET's 27-3042 entry is also worth reading because many employers build job descriptions from it. Outside the US, use the ONS Annual Survey of Hours and Earnings in the UK, Canada's National Occupational Classification technical writer code with Job Bank provincial ranges, and Australia's ANZSCO. Then read fifty live postings for your city, level and industry, since pay transparency rules mean many now carry a real band. Four things move the number for a technical writer: clearance level in defence work, the structured authoring and regulated specialisms, documentation engineer roles paid on engineering bands, and industry, where changing sector usually pays better than changing title.
Can you move into technical writing from support, QA or engineering?
Yes, and those are the most common entry routes into technical writing rather than exceptions. A support agent who has written the top fifty articles in their company's knowledge base is already doing a technical writer's job and should say so explicitly on the resume, with the article count, the ticket categories and what happened to volume. A QA engineer has the two scarcest habits in the field already: reading a specification carefully and actually running the procedure. An engineer moving into documentation starts ahead on subject matter and behind on structure and audience, and should invest specifically in task, reference and concept separation. A clinician, technician or field engineer moving into regulated or hardware documentation is often the strongest candidate in the pile, because the employer can teach the authoring tool far more easily than the domain. The move is usually easiest internally first: take the documentation nobody owns, do it well, and apply externally a year later with real samples.
Put this on a resume in about a minute
Paste your history once and point it at the Technical Writer posting you are looking at. No account, no card.
Build my resume free More roles