The email came in on a Tuesday. Four years with the same SaaS client at a decent 8c a word, roughly $2,800 a month you could plan a life around. They are “restructuring content production”. The new arrangement: their marketing coordinator generates the drafts, you clean them up at $40 an hour, three hours a piece. They say it like it is a promotion. Meanwhile the job board has 20 articles at $25 each and 340 applicants, and your pitch reply rate has gone from one in six to one in thirty.
Here is the honest answer to whether content writers should pivot, and it applies to anyone doing freelance writing for a living: the commodity end of this market has collapsed and it is not coming back, but the work that requires a human to know something, go somewhere, ask someone, or be legally accountable for a claim has held its price and in several segments is paying more than it did in 2022. The hard part is that most working writers built their business in the middle, and the middle is what disappeared. So the real question is not “is there hope”. It is: what share of your invoices last year came from work a model can now do for nothing, and how fast can you replace it.
What the labour data actually shows
Start with the US Bureau of Labor Statistics, the least emotional source available and the only one not selling a course.
Writers and authors: 140,300 jobs in 2025, projected to change by 0% through 2035, about 500 fewer positions. Median pay $76,910 as of May 2025. Sixty-five percent of the occupation is self-employed. BLS states plainly that “increasing use of artificial intelligence (AI) for writing is projected to dampen demand for these workers”, while still projecting roughly 11,900 openings a year from people leaving the field.
Two neighbouring occupations tell you more than the headline. News analysts, reporters and journalists: 47,100 jobs, down 6% to 2035, median $62,200. Technical writers: 46,400 jobs, up 1%, median $90,390, with BLS attributing the slow growth to AI tools making those workers “more productive” rather than to replacement.
The pattern there is not “writing is dying”. The closer your work sits to explaining a documented system to a specific audience with consequences attached, the better it is holding. The closer to general-interest words sold against advertising, the worse. One caveat: BLS projections are models, not measurements, and an occupation that is 65% self-employed is one the household surveys capture badly. Trust the direction, not the decimals.
The sharpest evidence on freelance specifically comes from Ozge Demirci, Jonas Hannane and Xinrong Zhu, whose study of a major global freelancing platform found a 21% decrease in job posts for automation-prone writing and coding work within eight months of ChatGPT’s launch, relative to manual-intensive categories. It went through peer review at Management Science rather than a press office.
The finding almost nobody quotes is the one that matters most to you. The writing jobs that remained were more complex and paid higher rates. That is the whole shape of it: fewer briefs, harder briefs, better money per brief, and far more competition for each one.
Sort your own invoices into three piles
Not the market in general. Your last twelve months of billing, by dollar value.
Pile one: gone, and not coming back
Volume SEO articles written to a keyword and a word count. Product descriptions at scale. Meta and category copy. Summarising a report someone sent you. Generic “what is X” explainers. Rewritten press releases. Listicles. First drafts of anything where the real brief was “words on this topic, by Friday, cheap”.
This work is not gone because the machine writes it better than you. On a good day it does not. It is gone because the machine writes it at roughly the quality the buyer was already accepting, at a marginal cost near zero, and no hourly rate competes with zero. Any argument starting “but AI content is bad” answers a question the buyer never asked. They were never buying good. They were buying adequate, in bulk.
Pile two: worth more than it was
- Original reporting and interviews. A quote from a named person who said it to you, this month, about their actual situation. There is no training data for a conversation that has not happened yet.
- Subject-matter expertise the model cannot access. Not “I researched this thoroughly”. You were a paediatric nurse, you ran payroll for 400 people, you hold a Series 7. The value is what you know that is not written down anywhere.
- Regulated and high-liability copy. Health, financial services, legal, pharma, insurance, employment. Someone has to own the claim, and that recently stopped being a soft argument.
- Brand voice with a real person attached. Founder ghostwriting, executive bylines, newsletters where the audience has a relationship with a name. The client is buying a voice they can point at.
- Editing and fact-checking machine output. Real demand, growing, and not the safe harbour it looks like. Caveat in the rates section.
- Content strategy and information architecture. What should exist, what should be merged, who it is for, what the brief says. Cheap generation makes deciding what to generate more valuable, not less.
- Research-led writing from primary sources. The filing, the statute, the dataset, the court record, and something nobody has published yet.
Pile three: genuinely uncertain, and anyone who tells you otherwise is guessing
Long-form B2B thought leadership, because it turns on whether buyers keep caring who wrote something. Technical documentation, which BLS calls stable while every documentation team I know pilots generation. Book ghostwriting, localisation, transcreation. And the edit-the-machine category itself, which could professionalise into a respected role or compress into $25-an-hour cleanup, with both paths currently open. Build your plan so a bad outcome in pile three is survivable rather than fatal.
What surviving freelance writing rates actually look like
The best public rate data with a stated methodology is the Editorial Freelancers Association’s 2026 Editorial Rate Chart: over 1,100 respondents, more than a third of the membership, collected November 2025 to mid-January 2026 and covering 2025 rates. It is self-reported and EFA membership skews toward book and academic publishing, so read it as a well-documented slice rather than the whole market. Selected figures:
- Ghostwriting, articles and essays: 15 to 30 cents a word, or $75 to $100 an hour
- Ghostwriting, full-length nonfiction: 33 to 50 cents a word, or $85 to $100 an hour
- Fact-checking: $50 to $60 an hour
- Copyediting, medical: 3.0 to 5.0 cents a word, or $60 to $75 an hour
- Developmental editing, nonfiction: 4.0 to 5.0 cents a word, or $57.50 to $70 an hour
- Copyediting, fiction: 2.0 to 2.7 cents a word, or $40 to $50 an hour
Two things fall out of that table. First, the spread: ghostwriting an article at 30 cents against copyediting fiction at 2 cents is a fifteen-fold difference inside one profession, and the gap tracks how much judgement and accountability the work carries. That is the bifurcation, with numbers on it.
Second, a warning about the popular escape route. “I’ll pivot to editing AI output” sounds safe until you notice that per-word editing rates sit well below per-word writing rates, and that the client paying you to fix a machine draft is doing it to cut the line item. Watch the hourly column. Editing is a good business when you are the named accountable editor on high-risk copy at $60 to $75 an hour, and a bad one when you are the cheapest cleanup for infinite drafts.
Which leads to the pricing change that matters more than any niche choice: stop selling words by the unit. The per-word rate was a proxy for effort when producing text was the expensive part. Text is now free, so pricing by the pound prices you against the free thing. Move new clients to project fees, day rates or retainers scoped by outcome. You will lose the ones who only understand cents per word, and those are pile-one clients anyway.
Liability just became a job description, in writing
Article 50 of the EU AI Act, which applies from 2 August 2026, requires deployers of generative AI to clearly label AI-generated text published to inform the public on matters of public interest. The Commission scopes that broadly: politics, public administration, justice, fundamental rights, public security, health, environment, consumer safety, and economic or cultural developments relevant to public debate.
Then the part that should interest you. There is an exemption where the content has undergone human review or editorial control with a person or organisation holding ultimate legal responsibility for the publication. The guidance is explicit that superficial checks do not qualify, naming spell-checking and grammatical correction as what is not enough. It requires deliberate examination by a qualified person with authority to approve, alter or reject the substance on substantive grounds.
Read that as a freelancer. A major jurisdiction has written into law that qualified human editorial judgement, exercised by someone who takes responsibility, changes the legal status of a piece of content. That is not a marketing angle. It is a role with a statutory function, and clients publishing into the EU on those topics need someone doing it.
Outside the EU there is no equivalent general labelling rule for text as I write this, but the underlying exposure never moved: a publisher is responsible for what it publishes and an advertiser for its claims, whatever produced the draft. That is why health systems, banks, law firms and insurers are the segment least interested in cutting your rate. They are not buying prose. They are buying someone who will put their name on a claim and can be asked, under oath if it comes to that, what the basis for it was.
If you want that work: name the sign-off in your contract, price accountability separately from drafting, keep research files and version history by default, and put your own AI use policy in writing. Clients in these sectors increasingly ask, and a clear answer wins work where being cagey loses it.
What buyers are paying for now, and how to prove you have it
The Content Marketing Institute’s enterprise research, fielded 24 June to 14 August 2025 across 1,229 respondents including 296 enterprise marketers, found 95% using AI-powered marketing applications and 86% using AI for content creation. Then the gap: 84% said productivity improved and 76% said operational efficiency improved, but only 53% said content quality improved and just 38% said content performance improved.
That gap is your market. They bought volume, got volume, and did not get results. Whoever closes the 46-point distance between “we make more content” and “our content works” is not competing with a text generator at all.
Proving you can takes different assets than the ones in your current portfolio: a beat you can name in six words (“revenue cycle management for hospital finance teams” beats “B2B SaaS and tech”); verifiable expertise in the form of a licence, credential or prior operating role; three people in that beat who will take your call this week; and clips whose value is an interview you conducted, a number you generated, or a judgement you defended.
Here is a test worth running this week. Take the three portfolio pieces you are proudest of, paste the original brief into a competent model, and read what comes back. If it lands within 80% of what you delivered, that piece has stopped being a sales asset, whatever it earned at the time. If it comes back confidently wrong in ways only someone in your field would catch, keep the comparison and show it to prospects. It is the most persuasive sales document available to a writer right now.
A six to twelve month transition plan with dates on it
Weeks 1 to 4. Do the invoice audit. Twelve months of billing sorted into the three piles by revenue, not by job count. If pile one is over 60% of your income, you are not deciding whether to change but how fast. Count your months of runway. That number sets the aggressiveness of everything below.
Weeks 2 to 8. Choose one beat and commit. Three criteria: you have a real credential, prior role or genuine access; buyers in it have budgets and compliance exposure; it is not “AI” or “marketing”, where every displaced writer is currently running. Do not pick two as a hedge. The hedge is what stops you becoming known for either.
Weeks 6 to 16. Build three proof pieces. Each must contain something that did not exist before you made it: an interview, a small dataset, a document analysis, a test you ran. Self-publish if nobody commissions them. This is the only unpaid work in the plan I would defend.
Weeks 8 to 24. Reprice everything new. No per-word quotes. Project fees or day rates that assume the thinking is the product. Track how often prospects refuse, because if nobody refuses your price it is too low.
Weeks 12 to 32. Re-scope or release the commodity clients. A script that works: “I’ve moved my practice to interview-led and compliance-reviewed work, and I can’t do volume drafting at a rate that makes sense for either of us. I can do fewer pieces at a higher standard, or recommend someone for the volume.” Say it once, do not apologise.
Weeks 16 to 36. Add one adjacent revenue line on a different demand curve: content strategy and briefs, technical editing, research services, or AI evaluation and data work, which pays for exactly the judgement you already have.
Month 9 to 12. Decide on evidence, not mood. After six months of real effort on the new beat, do you have two retained clients or one anchor at the new rate? If yes, compound it. If no, pivot properly: in-house content strategy, communications, technical writing, a subject-matter role that uses your domain, or AI evaluation and annotation. In-house content roles are under pressure too. A pivot changes your risk profile. It does not remove risk.
Related reading
Questions writers ask after this one
Is editing AI output a real career or a holding pen? Both exist and the difference is who is accountable. If you are the named editor of record on regulated or high-liability content, with authority to reject the substance, that is a defensible role with a legal function behind it and hourly rates to match. If you are the cheapest available polisher of unlimited drafts, the rate only goes one direction, and the volume of drafts you must handle to keep your income flat rises every year.
Should I tell clients I use AI in my process? Have a written policy and lead with it: which stages you use it for, which you never do, how you verify facts, what you will attest to on request. Writers who volunteer a clear policy get fewer awkward conversations and more regulated work. Writers who avoid the subject get asked anyway, in worse circumstances.
I am three years in with no specialism. Specialise or leave? Specialise, with a defined window to find out. Three years of general content experience has little market value now, but it is enough foundation that adding one real domain gets you somewhere inside a year. What does not work is another three years of being available for anything.
If this is the work you want more of rather than less: AB7 Solutions runs content operations and SEO, AEO and GEO programmes, so we commission the editorially-reviewed, subject-matter, accountable-author work described above, and we are consistently short of people who can do it. We also run data annotation, labelling, HITL, RAG evaluation and red teaming projects, which reward what writers already have: reading closely, spotting the confident falsehood, explaining in one sentence why it is wrong. To talk about either side, call +1 321 341 7733 or email ab@ab7solutions.com or director@ab7solutions.com. More at www.ab7solutions.com.
Sources: US Bureau of Labor Statistics, Occupational Outlook Handbook, Writers and Authors, Technical Writers and News Analysts, Reporters, and Journalists (May 2025 wage data; 2025 to 2035 projections); Ozge Demirci, Jonas Hannane and Xinrong Zhu, “Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms”, Management Science; Editorial Freelancers Association, 2026 Editorial Rate Chart (1,100+ self-reported respondents, data collected November 2025 to mid-January 2026, covering 2025 rates); European Commission, Transparency obligations under Article 50 of the AI Act (applies from 2 August 2026); Content Marketing Institute, Enterprise Content and Marketing Trends: Insights for 2026 (1,229 respondents, 296 enterprise marketers, fielded 24 June to 14 August 2025).
Related AB7 service: AI automation and AI agent services.