Proofreader: has AI replaced it?

AI tools doing this job

GrammarlyWriting tool that the authors of a peer-reviewed study published in 2024 count among AI tools used in editing and proofreading. 13 of 17 professional editors and proofreaders surveyed in Egypt said they use it, more than any other tool; the authors count it among AI tools and note that generative AI is integrated within applications like Grammarly. The study does not measure effects on jobs.source ↗
ChatGPTGeneral-purpose LLM. A study of job postings on one large freelancing platform from January 2021 to September 2023 (Teutloff et al., Journal of Economic Behavior and Organization, online January 29, 2025) found that after its release, demand for substitutable skills such as writing and translation fell by 20-50% relative to the counterfactual trend. A 'Writing & proofreading' cluster is among those with a negative estimate; the paper gives no separate figure for it.source ↗

Task breakdown

Task	Status	Tool	Source
Freelance writing and proofreading jobs on one online platform (2021-2023 postings)	Partially replaced	ChatGPT	Journal of Economic Behavior and Organization
Editing and proofreading with AI tools (17 professionals in Egypt, 2024 study)	Augmented	Grammarly	Discover Artificial Intelligence

How we rate →

Evidence

This page covers proofreaders and copy markers (BLS code 43-9081). BLS defines them as workers who read transcript or proof type setup to detect and mark for correction any grammatical, typographical, or compositional errors, and the definition adds: "Excludes workers whose primary duty is editing copy." Copy editors are counted under editors, a separate BLS occupation with about 105,000 jobs in 2025, so the newsroom copy desks discussed below are context for this job, not part of its count. Publishers are the largest employer. The May 2016 OEWS survey put 3,030 of 11,430 proofreaders in newspaper, periodical, book and directory publishers, and the May 2023 survey put 1,390 of 5,490 there.

Employment comes from the BLS Occupational Employment and Wage Statistics (OEWS) survey, national estimates for May of each year: 18,070 in 2005, 12,770 in 2010, 11,430 in 2016, 11,420 in 2017, 9,820 in 2018, 7,730 in 2019, 6,610 in 2020, 5,340 in 2021, 5,120 in 2022, 5,490 in 2023, 5,160 in 2024 and 4,580 in 2025. That is about 75% fewer jobs in 2025 than in 2005. Most of the fall came before ChatGPT was released on November 30, 2022: jobs fell about 57% from 2005 to 2019, and the steepest drop, about 53%, came from 2017 to 2021. Since 2022 the count has moved from 5,120 to 4,580, about 11% lower, after a rise in 2023. Caveats: OEWS is a survey estimate, and it does not cover the self-employed. BLS says it "does not encourage the use of OEWS data for time-series analysis," and its switch to a new estimation method (MB3) with the May 2021 estimates affects comparisons across that year. 2020 includes the pandemic. The May 2010 estimates were the first based partly on the 2010 occupational classification, and the estimates now use the 2018 version; the occupation title is the same in every year I used, but I did not check whether its boundaries shifted.

BLS has no full Handbook profile for proofreaders. Its table of occupations not covered in detail counts 8,500 jobs in 2025, a count that includes the self-employed (25.6% of the total in BLS's projections table), and projects a 1% decline from 2025 to 2035; the May 2025 median wage was $51,120. Leaving out the self-employed still gives about 6,300 jobs, more than the 4,580 in OEWS, so self-employment explains only part of the gap between the two counts. That page gives no reason for the projected decline and does not mention AI or software. The only BLS mention of AI for this occupation I found is its AI exposure table, published with the 2025-35 projections, which rates proofreaders and copy markers "Very high" in relative AI exposure. BLS says that generally means a larger fraction of an occupation's tasks can be completed or assisted by AI than in other occupations, and that LLMs have been observed performing some of its tasks. It also warns that "Exposure does not imply job loss, productivity gains, automation probability, or wage effects," that an exposure category "is not a worker replacement estimate," and that the theoretical sources describe AI capabilities as of no later than mid-2023.

The long fall coincides with the decline of print, a trend that is not AI, though no source I found attributes the proofreader decline to it. BLS reported in April 2017 that newspaper publishers lost over half of their employment from January 2001 to September 2016, from 412,000 to 174,000. Its current Handbook page for editors projects a 1% decline for that occupation from 2025 to 2035, saying that as traditional print newspapers and magazines lose ground to other media formats, employment of editors who work for those publications is projected to decline; that page does not mention AI. Newspapers also cut copy desks for cost and workflow reasons years before generative AI. Poynter reported on June 2, 2017 that The Denver Post eliminated copy editors in 2012, that five Gannett papers cut copy desks in 2014, and that The New York Times was offering buyouts to reduce the number of editors. The Times' managing editor said: "We are eliminating a separate copy editing desk, but we are not doing away with copy editing," and called the old system "a legacy of the print era." The same article noted that computers can flag spelling mistakes; spell-checkers and rule-based grammar checkers are plain software, not AI, by this site's rules.

I found no US publisher that said it cut proofreaders because of AI. The one case that names proofreaders next to AI is in Germany and is disputed. In June 2023 the tabloid Bild announced about 200 job cuts. The Irish Times reported that Bild would part with colleagues whose tasks in the digital world can be replaced by AI or other processes, and that the functions of editors, layout staff, proofreaders, secretaries and picture editors would no longer exist in their current form. The same report tied the cuts to a digital-only strategy and a goal of boosting revenue by EUR 100 million over three years, and said six regional editions would be cut. A Bild spokesperson told CNN the job losses had "nothing to do with the use of AI." In the UK, the NUJ said Reach, publisher of the Mirror and Express, sought to cut "over 50 sub-editors" among about 70 print production staff in its UK nationals in late 2025 (Press Gazette, October 30, 2025). The union said Reach uses AI to duplicate content across publications, while a Reach spokesperson said the company had redeployed over 100 at-risk staff to roles supporting its ambitions in video, digital subscriptions and its Live News Network. No source I read says AI took over the sub-editors' work, and UK sub-editors are closer to copy editors than to proofreaders.

The clearest AI evidence comes from freelance platforms, and it covers proofreading only as part of wider groups. Teutloff and co-authors (Journal of Economic Behavior and Organization, online January 29, 2025) studied daily job postings from January 2021 to September 2023 on a leading freelancing platform that they do not name. They found that demand for skills that generative AI can substitute, such as writing and translation, fell by 20-50% relative to the counterfactual trend after ChatGPT, with the sharpest decline in short-term (1-3 week) jobs. Their chart of individual skill clusters places a "Writing & proofreading" cluster below zero, but the text gives no figure for it, and the authors say cluster-specific estimates should be interpreted with caution. One writing-related cluster, labeled fiction writing and editing, showed no significant decrease; it is not a copyediting or proofreading cluster. Hui and Reshef, summarizing their Upwork study for Brookings (July 8, 2025), wrote that freelancers offering copyediting, proofreading and other text-heavy tasks saw about 2% fewer new monthly contracts and about 5% lower total monthly earnings on the platform after ChatGPT. Their data cover only the first six to eight months after the release.

AI tools are in use among editors and proofreaders, but I found no representative survey of how many use them. In a peer-reviewed study published in 2024, all 17 professional editors and proofreaders surveyed from a society of professional editors and proofreaders in Egypt used AI tools in editing and proofreading and said the tools helped their work. Grammarly was the most used (13 of 17), followed by ChatGPT (8). This site counts Grammarly's machine-learning tools and LLMs such as ChatGPT as AI.

Limits: OEWS counts fewer than 5,000 proofreaders, and no source I found explains the 2017-2021 fall, which may reflect the decline of print, the 2021 change in survey method, or causes no source names. No source ties any part of the OEWS decline to AI, and none separates AI from these other causes. The freelance studies cover one platform each, mix proofreading with other writing work, and end in 2023, so they say little about salaried proofreaders. The Bild and Reach cases are outside the US data and are about wider newsroom staff. The only survey of AI use is small and from Egypt. I could not read the Guardian's report on Bild, and the BLS data API was not reachable, so the 2025 figure comes from the May 2025 national table.

The status is augmented. AI writing tools are in use, freelance demand for writing and proofreading fell after ChatGPT, and BLS rates the occupation's AI exposure very high. But most of the job loss came before generative AI, BLS projects only a 1% decline to 2035 and gives no AI reason for it, and no source shows employers replacing proofreaders with AI at scale. Partial is arguable from the two freelance studies, so the weaker status wins. Confidence is low: the freelance evidence covers only a slice of the work, proofreading is mixed with other writing in both studies, and the recent fall in the small OEWS count is unexplained.

References

Suggest a correction on GitHub →