■ CRITICAL RISK ■ Media & Communication
For volume SEO content, the replacement isn't looming — it's on the invoice. LLMs already produce serviceable blog posts for pennies, and the content-mill tier of this job has collapsed. Writers with genuine expertise, reporting skills, or a distinctive voice are being promoted to editor; everyone else is competing with a subscription.
“ChatGPT writes blog posts at 3 AM without needing 'creative space.'”
Our AI replacement risk score — how we score jobs
Web content writing, in its dominant commercial form, was never literature — it was search-intent fulfillment. Take a keyword brief, research the top-ranking pages, produce 1,500 optimized words on 'best CRM for small business' or 'how to unclog a drain,' hit the H2 structure the SEO tool suggests, and file by Thursday. Agencies and content mills built an entire industry on this, paying per word for prose whose main quality bar was 'ranks and doesn't embarrass the client.'
That job description reads like an LLM product spec, because functionally it is. ChatGPT, Claude, and purpose-built content platforms generate competent commercial copy instantly, and the market noticed: freelance platforms saw writing gigs contract after 2023, content agencies restructured around AI-assisted workflows, and rates for commodity writing fell through the floor. The workflow that survives isn't 'human writes' but 'human briefs, AI drafts, human edits' — which needs perhaps a third of the writers, retitled as editors. Google's crackdowns on mass-produced content sting the worst offenders, but well-edited AI-assisted content ranks fine, so the moat many writers hoped for never filled.
What still commands money: actual expertise (a nurse writing medical content, an engineer explaining APIs), original reporting and interviews an LLM cannot conduct, brand voice development, content strategy, and accountability — someone whose name and reputation stand behind claims in regulated industries. E-E-A-T pressures from search engines push the same direction: content needs demonstrable human experience to differentiate. The 76 risk score reflects a bifurcated fate — the commodity majority of this occupation is already substantially automated, while a smaller expert tier climbs the value chain.
Automatability: our editorial assessment of current and near-term AI capability
The disruption arrived in 2023 and has been compounding since. Commodity content rates collapsed first; agency restructuring followed; the editor-supervising-AI workflow is now standard. Through the late 2020s, expect continued contraction of pure writing roles and growth in hybrid strategist-editor positions. There is no 'before the wave' left to prepare during — anyone in this field is already treading water in it, and the differentiation strategies are the swim strokes.
For the commodity tier, largely yes. Volume SEO posts, product descriptions, and generic how-to content are now routinely AI-generated with light human editing, and freelance rates for that work collapsed accordingly. What hasn't been replaced: writers with real expertise, original reporting, distinctive voice, or strategic skills. The occupation didn't vanish — its bottom two-thirds got automated while the top third changed job descriptions.
Not the 2019 version of one. Entering the field now means entering as an AI-fluent editor-strategist with a specialty: pick a domain you genuinely know, learn to direct and edit AI drafts to a high standard, and develop skills models lack — interviewing, original research, brand judgment. The pure freelance blog-post market is a race to the bottom against software that works for a monthly subscription fee.
Google penalizes unhelpful mass-produced content, whoever wrote it — its spam updates have hammered sites publishing thousands of thin AI pages. But it has stated that well-made AI-assisted content can rank, and in practice edited AI content ranks constantly. The practical bar is quality and demonstrated experience (E-E-A-T), not authorship method. Which means 'Google will save human writers' was a hope, not a strategy.
Work anchored in things a model doesn't have: firsthand experience (product testing, travel actually taken, procedures actually performed), access (interviews, embargoed information, community trust), accountability (regulated industries needing a credentialed human behind claims), and voice tied to a known persona. Also the meta-work: strategy, briefs, and editing. If your output could be swapped with a prompt's output without the client noticing, that's the signal to move.