■ CRITICAL RISK ■ Media & Communication
For catching typos and grammar slips, yes — that's a solved software problem free in every text box. The surviving profession is a thin specialist layer: legal and medical proofing, publishing quality control, and now, ironically, checking AI's own output.
“Grammarly doesn't need coffee to catch your typos.”
Our AI replacement risk score — how we score jobs
Proofreading is the last quality gate before text goes public: catching typos, grammar errors, punctuation slips, formatting inconsistencies, and — in the traditional publishing sense — comparing proofs against the marked-up manuscript to verify corrections were made. It's distinct from copyediting (which fixes style and clarity) and from editing proper (which fixes the writing), though small shops always collapsed the three into one underpaid person.
The commodity layer is gone. Spellcheck and grammar checkers ship in every writing surface; tools like Grammarly catch a typical proofreader's bread-and-butter errors instantly and at zero marginal cost; and large language models push further, flagging awkward constructions, inconsistent terminology, and even factual oddities. For the routine business document, blog post, or email — a large share of what freelance proofreaders actually billed for — the machine pass is good enough that clients stopped paying for a human one. Rates and volumes in the freelance market show it. That collapse of the general market is the substance of our 95.
The residue is where errors are expensive and accountability matters. Legal proofreading survives because a wrong number in a contract has consequences software can't be sued for; medical and pharmaceutical labeling runs under regulatory scrutiny; book publishers keep proofreaders because a printed error is permanent and brand-damaging; and financial documents get human eyes for the same reason. The newest niche is the strangest: organizations generating text with AI need humans to verify it — checking not just spelling but whether the confident-sounding sentence is true, a failure mode traditional proofreading never had to consider. The role that survives is closer to quality assurance with domain expertise than to typo-hunting, and it employs a fraction of the people the general market once did.
Automatability: our editorial assessment of current and near-term AI capability
The general market already collapsed — free grammar tools ended paid proofreading for routine text over the past decade, and LLMs have deepened the cut since. Through the rest of the 2020s, expect remaining demand to concentrate in regulated and high-stakes niches (legal, medical, financial, book publishing) and in verification of AI-generated content, with generalist freelance proofreading continuing to shrink toward hobby-income territory.
As a generalist, it's increasingly hobby income — free tools handle routine text well enough that most clients stopped paying humans for it, and rates reflect that. Livings still exist in specialization: legal and medical proofreading, book publishing quality control, financial documents, and academic work for non-native English writers. The common feature is stakes and accountability, not comma expertise.
The ones requiring context and stakes-awareness: a technically grammatical sentence that says the legally wrong thing, a plausible-looking figure that's off by a factor of ten, terminology that's consistent but incorrect for the field, and layout-level problems in designed documents. Most importantly, AI tools confidently miss their own class of failure — fluent, well-spelled sentences that are simply false — which human verifiers now get paid to catch.
Increasingly, yes — though it's better described as content verification or QA. Companies generating marketing copy, documentation, and reports with LLMs need humans to check accuracy, brand fit, and legal exposure before publishing, because the failure mode isn't typos, it's confident falsehood. It pays better than traditional proofreading precisely because it requires domain judgment, and it's currently one of the few growing niches in the field.
It's the natural climb. Copyediting and developmental editing involve judgment about clarity, structure, and audience that automates far less cleanly than error-catching — and clients perceive the value difference, which shows in the rates. AI pressures editing too, but as a drafting assistant rather than a replacement. Add domain specialization on top and you've rebuilt the career one durable level up.