Is AI taking technical writing jobs? What the data says

No, not in the “robots took the jobs” sense, at least not yet. The Bureau of Labor Statistics still projects slight growth in technical writing employment through 2035, and we found no official or reputable data showing that AI has already cut technical writing jobs. AI is changing the work, though. It is taking over more of the drafting and reformatting, and BLS says that may slow hiring because each writer gets more done. The writers who stay in demand are the ones who own accuracy, structure and the conversations with engineers.

What AI speeds up here

Read the BLS numbers before the headlines

The BLS Occupational Outlook for technical writers gives you the clearest baseline:

• 46,400 jobs in 2025.
• Median pay of $90,390 a year, or $43.46 an hour (May 2025).
• Projected growth of 1% over 2025 to 2035, which BLS calls “slower than the average for all occupations.”
• About 3,200 openings a year, mostly to replace people who change occupations or leave the workforce.

On AI, BLS says employment growth “may be slowed by artificial intelligence (AI) tools that allow these workers to be more productive.” So BLS expects AI to slow growth by making each writer more productive, which is different from replacing writers. These are also projections. They don’t count any jobs already lost.

BLS tracks writers and authors as a separate occupation, and its outlook there is harsher: it says AI for writing “is projected to dampen demand,” and the median pay is $76,910. If most of your work is marketing copy or generic blog posts, you face more direct pressure than someone writing API references or installation guides for complex products. Our breakdown of what’s shrinking and what’s paying in writing work covers that side.

Why technical writing ranks high on AI “applicability”

The scarier-sounding data point comes from Microsoft Research. In “Working with AI: Measuring the Applicability of Generative AI to Occupations”, the researchers mapped Bing Copilot conversations to the job tasks listed in the O*NET database. Technical writers scored 0.373, which puts them 13th among the paper’s top 40 occupations by AI applicability.

The authors also warn against assuming that high-applicability occupations will be automated or lose jobs or wages. They say the outcome depends on business decisions that are hard to predict, and that their scores show where AI is likely to change work, not how employment will change.

In practice, people already bring a lot of technical writing tasks to chatbots, so the job will look different. Some companies may use AI to get more documentation out of the same team, and others may use it to cut the team. The research can’t predict which will be more common.

The STC closure is a money story

In January 2025 the Society for Technical Communication, the field’s long-running professional body, announced it would shut down. It blamed “financial liabilities coupled with falling membership numbers” and debt and operating costs that outweighed revenue. STC’s statement doesn’t mention AI. Some commentators described an industry under pressure from AI, but STC never gave that as a cause. Be skeptical if someone cites the closure as proof that AI killed technical writing.

The human-only part

A chatbot can produce a draft in seconds, but a company can’t ship that draft until someone vouches for it. Vouching for it is the paid work:

• Running every command, clicking every menu path and confirming every default value against the current build.
• Getting the facts that live only in engineers’ heads, such as the undocumented limit, the workaround support keeps giving out, or the feature flag nobody wrote down. That takes interviews, follow-ups and some diplomacy.
• Making audience calls, like deciding a page has to work for a nervous first-time admin rather than a power user.
• Owning the result. When the docs are wrong and a customer breaks production, someone answers for it, and AI tools don’t sit in that meeting.

If you’re building or protecting a technical writing income, put your energy into these and let AI handle the reshaping and polishing.

Hand AI the boring half of a documentation job

This workflow keeps you in charge of the facts and gives the mechanical parts to AI.

1. Gather the source material: specs, tickets, release notes, recorded explanations and screenshots. Before you paste anything into a consumer chatbot, check your employer’s or client’s policy on AI tools, and strip out anything confidential or unreleased.
2. Write a five-line brief covering the audience, the task the reader needs to complete, the product version, the doc type and your top style rules.
3. Ask for an outline and a gap list, not a draft. The prompt below does this.
4. Take the gap list to your subject matter expert. Engineers answer a list of specific questions much faster than “can you review this?”
5. Draft with their answers. AI now has real facts to work with, so the draft is worth editing.
6. Test every step in the real product. Clients are paying you for this step.
7. Run a consistency pass. Paste in your style rules and the draft, and ask for a list of violations rather than a rewrite, so you decide what changes.
8. Do a final read for tone, flow and anything that “sounds right” but isn’t.

For tools, a general chatbot like Claude or ChatGPT is enough, and the free tiers cover occasional drafting (see where free tiers stop). If you use one daily and keep hitting limits, Claude Pro costs $20 a month, or $17 a month billed annually. Max plans start at $100 a month, which most solo writers don’t need. Our guide to AI tools for freelance writers compares the other options.

Prompt you can copy

You are helping a technical writer plan a [doc type, e.g. quickstart guide] for [product name and version].

Audience: [who reads this, e.g. backend developers new to our API]
Goal: after reading, the reader can [one specific task].
Source material is between the ### markers below.

Do three things, in this order:
1. Propose an outline with numbered sections and a one-line purpose for each.
2. List every fact the outline depends on that is NOT stated in the source material. Label each one MISSING and phrase it as a question to ask an engineer.
3. Do not write any procedure steps yet. Do not invent commands, parameters, menu names, file paths or default values.

Style rules: [e.g. second person, present tense, sentence case headings, "select" not "click"].

###
[paste spec, ticket or notes here]
###

Step 2 is the one to keep even if you throw away the outline. The MISSING list doubles as your interview agenda.

Where AI falls over

• Invented specifics. If you ask for a procedure without enough source material, you’ll get command flags, API endpoints and menu paths that look right but don’t exist. These are the most dangerous errors because they pass a skim.
• Outdated interfaces. Models often describe how a well-known product used to work. If the UI changed last quarter, expect the old menu names.
• Confident defaults. AI will state timeouts, rate limits and file size caps without hesitation. Treat every number as unverified until you’ve checked it.
• One template for everything. Every page becomes Overview, Prerequisites, Steps, Troubleshooting, even when you needed a reference table or a conceptual explainer.
• Unrequested rewrites. In an editing pass it may soften a carefully worded warning. Ask for a list of suggested changes instead.
• Confidentiality slips. Pasting an unreleased roadmap into an unapproved tool can cost you a client faster than any typo.

Check AI output before it ships, every time. Fluent writing is no evidence that the facts are right.

If you freelance: copyright and taxes

Copyright

The US Copyright Office’s Part 2 report on copyrightability (January 2025) concluded, in summary, that with current technology, prompts alone don’t give you enough control to count as the author of AI output. Using AI to assist your own original expression doesn’t by itself stop the whole work from being protected, and your human-written material stays protected, but purely AI-generated portions may not be. Our inference is that if you hand a client mostly unedited AI output, you may not be able to give them full copyright in that text, which matters if your contract assigns IP to them. Be upfront about how you use AI, and make sure your own writing and editing are substantial.

Taxes

Freelance documentation income goes on Schedule C. Once your net self-employment earnings reach $400, you also owe 15.3% self-employment tax, filed on Schedule SE. For tax years after 2025, a client only has to send you a 1099-NEC once they’ve paid you $2,000 or more, but the IRS is clear that gig income is taxable whether or not a form arrives. If you expect to owe $1,000 or more, you’ll generally need to make quarterly estimated payments with Form 1040-ES. This is general information, not tax or legal advice, so check your own situation with a tax professional.

AI training gigs: a side door, not a safety net

Some technical writers pick up extra work evaluating chatbot responses. DataAnnotation’s FAQ lists $25 to $50+ an hour for chatbot evaluation and $40 to $125+ for domain-specific work. Its homepage quotes different “up to” figures, so treat those as ceilings. Its Starter Assessment takes about an hour and can only be taken once. Outlier doesn’t publish rates at all. Our guide to AI writing jobs from home compares the platforms.

This market is unstable. Scale AI cut 500 contractors in July 2025. In September 2025, xAI laid off about 500 data annotators as it moved toward specialist tutors. Mercor offered a replacement Meta project at $16 an hour, $5 less than before. The trend favors people with real domain expertise, which technical writers often have, but none of it is steady income. Scammers also borrow these platforms’ names, so remember the FTC’s rule of thumb: “Never pay anyone to get paid.”

Try it in 20 minutes

1. Find an open-source project on GitHub with a thin README.
2. Paste the README into the free tier of Claude or ChatGPT, along with the prompt above.
3. Count the MISSING items it lists.
4. Then ask it to write the quickstart anyway, and to mark every step it couldn’t have known from the README.

The invented steps show you which part of the job still needs a person. The MISSING list shows how much prep time AI can save you before your next engineer interview.