Largely yes for the mechanical layer, and no for the part that requires judgement. Automated checking now catches spelling, grammar and most punctuation errors better than a tired human reading at speed, and that has removed a category of work that used to be paid by the hour. Federal figures put editors at a 1 percent decline from 2025 to 2035, taking the occupation from 105,000 positions down by about 1,100, with median pay of $77,920.
A 1 percent decline sounds mild for a field this exposed to artificial intelligence. The reason it is not worse tells you which parts of editorial work have survived and which have not. If you find yourself wondering whether to stay in this work, the honest answer depends entirely on which of those parts you actually do.
Key points
- Mechanical correction is effectively automated. Spelling, grammar, agreement and standard punctuation are handled reliably and instantly.
- Editorial employment declines 1 percent to 2035, losing about 1,100 positions from a base of 105,000.
- Consistency and house style still need a person, because the rules are organisation-specific and often unwritten.
- Factual and legal checking is the durable work. Names, figures, claims and anything that creates liability.
- The failure mode inverted. Errors used to look wrong. Now they read perfectly and are still wrong.
What proofreading actually involves
The word covers several activities that are frequently conflated, and they face very different pressure.
Mechanical correction. Spelling, grammar, subject-verb agreement, punctuation. This is what most people mean by proofreading and it is the part that automated.
Consistency checking. Whether a document uses one spelling of a name throughout, whether headings follow the same pattern, whether numbers are formatted the same way. Partly automatable where the rule is explicit, difficult where it is not.
House style. Applying an organisation’s own conventions, which are frequently undocumented, occasionally contradictory, and exist mostly in the heads of the people who have worked there longest.
Factual verification. Checking that a name is spelled correctly, a figure matches the source, a date is right, a quoted person actually said that. This is where the expensive mistakes live.
Legal and compliance review. Catching claims that cannot be substantiated, statements that create liability, or content that breaches regulatory requirements in a specific sector.
Sense checking. Noticing that a sentence is grammatically perfect and means nothing, or means the opposite of what was intended.
The part that did not change
It is worth being precise about what survived, because it defines what an editor should be good at now.
Someone still has to catch the error that reads correctly. An automated checker verifies that text conforms to rules of language. It cannot know that the company was founded in 1997 rather than 1979, that the executive quoted left the business last year, or that the percentage in paragraph three contradicts the table in paragraph seven. It does not think about the subject at all; it processes the words. These are the errors that cause retractions, and they are invisible to any tool operating on the text alone.
Someone still has to hold the whole document in mind. Consistency is not a local property. Whether this chapter contradicts that appendix, whether the tone shifts halfway through, whether a term is defined after it is first used, all require reading the thing as a whole and remembering it.
Someone still has to take responsibility for publication. In regulated sectors this is formal, with named sign-off. Elsewhere it is informal but real, and it is the reason organisations keep editorial staff even when the mechanical work has gone.
Where automation genuinely performs
Being honest about this matters more than defending the profession.
Modern checking tools catch mechanical errors at a rate that exceeds careful human proofreading, and they do it instantly across any length of document. They enforce documented style rules consistently, flag readability problems, detect inconsistent terminology when it is defined in advance, and translate between languages at a standard adequate for many purposes.
For organisations that previously published unchecked text, this is a straightforward improvement. For those that employed people primarily to catch typos, artificial intelligence substitutes for most of that work, and pretending otherwise would not help anyone planning a career.
There is one hazard, and it is the defining problem in this field now. Generated and heavily corrected text reads fluently regardless of whether it is accurate. Reviewers historically used awkward phrasing as a rough signal that a passage had not been checked. That signal has gone. Everything reads well now, including the parts that are wrong, so verification has to be deliberate rather than triggered by something looking off.
What replaced the old workflow
The practical shape of the job changed rather than the job disappearing, and it is worth describing concretely.
A document used to arrive, a proofreader read it once for mechanical errors and once for sense, and returned it marked up. Both passes took time proportional to length. Now the mechanical pass happens before a human sees the text, which removes perhaps half the hours and none of the responsibility.
What fills the space is verification. Checking that the person named in paragraph two holds the job title given. Confirming the statistic against the report it came from rather than against the press release about the report. Noticing that two sections of a long document give different figures for the same thing. Establishing whether a claim about a product can actually be substantiated, which frequently means asking someone who knows.
This is slower per page than mechanical correction ever was, and it is the part clients most often try to skip. Editors who can explain why it matters, in terms of what a published error would cost, tend to keep the work. Those who present themselves as careful readers compete directly with software.
Why the decline is smaller than the automation suggests
Three factors slow the fall.
Volume of published text grew enormously. Cheaper production means more content, and more content means more opportunities for the errors that automation does not catch.
The cost of a factual error rose. Published mistakes spread further and faster than they used to, and correcting them publicly is expensive in reputation terms. That raises the value of the checking that only a person does.
Regulated sectors require named accountability. Financial promotions, medical information, legal notices and safety documentation all involve someone signing that the content is accurate. That requirement is written into rules rather than chosen, and it does not soften as tools improve.
Translation and localisation grew. Organisations publishing in several languages need someone who can confirm that a translated version says what the original does, which is a verification task rather than a language one.
What to know before deciding
| Measure | Editors, 2025 |
|---|---|
| Median annual pay | $77,920 |
| Number of jobs | 105,000 |
| Projected change, 2025 to 2035 | -1 percent (Decline) |
| Projected employment change | -1,100 |
| Typical entry-level education | Bachelor’s degree |
The comparison with adjacent verification work is the most useful thing here.
Technical writers are forecast flat at 1 percent growth with median pay of $90,390, doing work that also involves producing and checking text. The difference is domain knowledge and accountability for accuracy against a system. Court reporters hold flat at 0 percent at $72,420, protected by a certification the law requires.
The pattern across all three is consistent. Where the work is checking text against rules of language, automation wins. Where it is checking content against reality, or where a regulation names a responsible person, the work persists.
For context on how exposure is measured, the Bureau publishes AI exposure categories for 831 occupations and states plainly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.”
What actually changes over the next decade
- Mechanical checking disappears as a paid task. Nobody pays by the hour for what a tool does instantly.
- Fact-checking rises in relative value. It is the error class that survives automation and causes the most damage.
- Volume increases the workload per editor. More text produced means more to review, even as each piece needs less mechanical attention.
- Style guides get formalised. Organisations that want automated enforcement have to write down conventions that used to be tacit.
- Verification replaces reading for feel. The old habit of skimming for signals of carelessness no longer works.
- Rates split. Commodity checking is priced against software; substantive editing and fact-checking hold their value.
Decision framework
Five questions if you proofread or edit now.
- What share of your work is mechanical correction? That is your directly measurable exposure, and it is usually higher than people estimate.
- Do you check facts, or only language? Fact-checking is the defensible position. It requires curiosity and access to sources rather than grammar knowledge, and it means being willing to ask an author questions they may not like.
- Do you have a subject area? Editors with genuine expertise in medicine, finance, law or a technical field are far harder to replace than generalists.
- Are you in a regulated context? Sectors with named sign-off requirements retain editorial roles that unregulated publishing has shed.
- Will you review generated text properly? Someone has to catch the fluent sentence that is wrong. That habit is now the core professional skill.
That last point transfers well beyond editorial work. Knowing where a model’s confident output fails, and verifying against sources rather than appearance, applies anywhere these tools are used. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
Your next step
If you edit for a living, separate your last month of corrections into mechanical fixes and substantive ones. The proportion tells you honestly how much of your work a tool already does, and it is usually the more uncomfortable number that people avoid calculating.
Then pick a subject area and learn it properly. Editors who understand the field they work in catch the errors that matter, and that expertise is what separates a rate priced against software from a rate priced against expertise.