Editor's note: This article was also published in The Surveyor, YouGov America's email newsletter. Subscribe to The Surveyor for regular updates on YouGov's polling.
This article was written entirely by a human.1 That's increasingly rare: more than one-third of recent web pages show signs of AI writing. And a new YouGov poll shows that 45% of Americans at least sometimes use AI tools to help them write, including 17% who do so at least weekly.
But there's a counterintuitive finding about who is using AI to write. It's not the least confident writers who are most likely to let AI shape their words. Instead, more of the confident, prolific writers are using AI to write. What follows is a deep dive to understand this finding.
The groups who use AI most for their writing
Using AI tools to write is becoming more common. Since last August, the share of U.S. adult citizens who say they never use AI to write has fallen to 51% from 60%, while the share who use AI to write at least weekly has risen to 17% from 12%.
AI writing assistance is most common among younger adults: 36% of 18- to 29-year-olds and 39% of 30- to 44-year-olds use AI to write at least monthly, compared to 19% of 45- to 64-year-olds and 8% of those 65 and older.
AI-assisted writing is less common among poorer Americans. 60% of those with family incomes under $50,000 say they never use AI to write, compared to 43% of those with higher incomes.
College graduates are more likely to use AI to write than non-graduates. 35% of those with four-year degrees use AI to write monthly or more, and 39% never use AI for writing. That compares to 19% of adults without college degrees who use AI to write monthly or more and 57% who never do this.
Americans who are turning to AI to help write are generally not those Americans who see themselves as bad writers.
Overall, 27% of Americans say they are very good at writing, while 47% say they're somewhat good at writing, and 21% say they're either not very good (15%) or not at all good (6%). (Because of the relative size of the groups, for much of the analysis that follows we'll combine the self-described not very good writers and the not at all good writers, often calling them the least confident or worst writers.) Younger adults and higher-earning Americans are slightly more likely than older or poorer Americans to say they're at least somewhat good at writing. Americans with college degrees are much more likely than those without to rate their writing skills highly.
People who describe their writing less positively are less likely to use AI to write. 64% of those who say they're not very good at writing or not at all good at writing never use AI to write, compared to 47% of those who say they're somewhat good at writing and the same share of those who say they're very good.
A group that's particularly likely to use AI to help write is people who read books but aren't the heaviest readers. 38% of those who say they've read three or four books in the past 12 months use AI to write at least monthly, as do 36% of those who read five to nine books. Use of AI for writing is less common among those who did less reading in the past year: 23% of those who read one or two books use AI to write monthly, as do 17% of those who read no books. AI-assisted writing also falls off among the heaviest readers, although some of these changes fall within the margins of error: 30% of those who read 10 to 19 books in the past year and 27% of those who read 20 or more use AI to write monthly.
Those who write and those who don't
Why is using AI for writing more common among Americans who are confident in their writing ability than among worse writers, who might seem to most need assistance writing? The answer is that most Americans don't write very much. People who think their writing is bad are less likely to write, and people who aren't writing anything have no writing for AI to help with.
YouGov asked Americans how often they do 10 different types of writing. Most Americans (70%) say they write text messages daily, and 94% ever text. 93% also ever send emails, but other types of writing are far less common. 76% say they ever write letters or notes and 50% write work documents. Less than half of Americans say they ever do each of these other forms of writing: journaling (44%), speeches or presentations (40%), nonfiction (36%), fiction (34%), poetry or songs (33%), or papers for school or training programs (32%).
More confident writers are more likely than the least confident writers to ever do each of these types of writing. For example, 51% of very good writers ever write nonfiction, compared to 19% of the least confident writers. 47% of very good writers write poetry or songs, compared to only 13% of the least confident writers.
And more confident writers are also more likely to do multiple kinds of writing. Because texts and emails are both so common, they aren't very useful for differentiating between groups of writers. But the other eight forms of writing from the survey do. 60% of those who say they're very good writers say they ever do at least four of those eight types of writing, including 18% who do all eight. In contrast, among the least confident writers, only 22% do at least four of the eight types of writing, and only 2% do all eight.
The big split here is between those who say they're at least somewhat good at writing, and those who say they're not very good or not at all good. For example, 20% of very good writers and 25% of somewhat good writers only do one or fewer of the eight types of writing — but 50% of the not very or not at all good writers do one or none.
Returning to the question of who uses AI to write, frequent AI-assisted writers are also more likely than those who seldom or never use AI to write to say they do many types of writing. 70% of those who use AI to write weekly or more ever do at least four types of writing, compared to 29% of those who never use AI to write. Here again, the big difference is between those who never use AI to write and everyone else who even occasionally writes with AI — 71% of the former group do fewer than four of the types of writing, compared to 38% of everyone else.
The four groups
While self-professed good writers are more likely to use AI to write, it's not a perfect correlation. In particular, there are differences among more frequent writers based on whether they use AI to write. YouGov analyzed the results and found it useful to classify Americans into four general categories based on how often they write and whether they ever use AI to write:
- 39% of Americans rarely write and never use AI to write
- 17% rarely write but do sometimes use AI to write
- 16% often write and never use AI to write
- 28% often write and do use AI to write
Our methodology: Frequency of writing comes from the eight different types of writing YouGov asked Americans if they do (after excluding the near-universal categories of text messages and emails). Those who ever do at least four of the eight types of writing are classified as writing often, and those who do three or fewer are classified as writing rarely. People who aren't sure how often they write in a genre are counted as never writing in it. The cutoff at four writing genres was chosen as a compromise because it created four groups of sufficient sample size with distinctive behavior. AI usage is based on how often Americans say they use AI to write, with AI users those who say they use AI to write daily, weekly, monthly, or rarely, and non-users those who say they never use AI to write or aren't sure how much they use AI to write.
Younger adults are more likely to write often than are older Americans: 57% of adults under 30 ever do at least four of the eight types of writing, compared to 27% of those 65 and older. But both groups are about equally likely to write without using AI. 20% of adults under 30 and 19% of those 65 and older write often and never use AI. Where there's a big gap between age groups is in AI-assisted writing: 37% of 18- to 29-year-olds write often and use AI, compared to 8% of adults 65 and older.
Among those who say they're very good writers, 60% write often. More of them say they do use AI (37% of very good writers) than say they don't (24%).
https://datawrapper.dwcdn.net/lZOqA/1/
Among Americans who rarely write, about one-third (30%) say they ever use AI to help them write — compared to two-thirds (64%) of those who write often.
This pattern holds in a range of demographics, including age and educational attainment — with the exception of Americans 65 and older. Among these older Americans, AI usage for writing is similar among both those who rarely write (25%) and those who often write (31%).
https://datawrapper.dwcdn.net/ZuFE9/1/
Among Americans who do use AI to write, 62% write often, compared to 29% of Americans who never use AI to write.
YouGov asked people who use AI to help write what kind of things they have AI do for them, from conducting research or generating ideas, to having AI write parts of documents or whole documents. There were only minor differences between those who write rarely and often used AI in their writing, and similarly minor differences in AI usage for writing among those who say they're very good, somewhat good, or not very good writers.
Those who write often without AI really dislike AI
Americans who don't use AI to write tend to have more negative views about AI than those who do — and this is especially true for those who write often and who don't use AI. Overall, 68% of Americans say they prefer to read something written by a human, 3% prefer AI writing, and 20% say it doesn't matter. But 86% of those who often write and don't use AI to write say they prefer human writing, compared to 72% of those who rarely write and don't use AI to write, 58% of those who often write and use AI to write, and 57% of those who rarely write and use AI to write.
https://datawrapper.dwcdn.net/OImJA/1/
YouGov asked Americans about whether AI writing was generally better or worse than human writing in 10 different categories. Many say AI is worse than humans at expressing emotion, being original, and being funny, while fewer say AI is worse than humans at clarity or correct grammar and spelling.
But in each of the 10 categories, Americans who write often and don't use AI to write are the most likely of the four groups to say AI is worse than humans — sometimes by significant margins. For example, 65% of those who write often and don't use AI say AI is worse than humans at sounding like a real person, compared to 47% of all Americans.
https://datawrapper.dwcdn.net/pfdSO/1/
Methodology: The August 14 - 17, 2026 poll was conducted among 1,092 U.S. adult citizens. Respondents were selected from YouGov's opt-in panel to be representative of U.S. adult citizens. A random sample (stratified by gender, age, race, education, geographic region, and voter registration) was selected from the 2019 American Community Survey. The sample was weighted according to gender, age, race, education, U.S. region, 2024 presidential vote, 2020 election turnout and presidential vote, baseline party identification, and current voter registration status. 2024 presidential vote, at time of weighting, was estimated to be 48% Harris and 50% Trump. Demographic weighting targets come from the 2019 American Community Survey. Baseline party identification is the respondent's most recent answer given around November 8, 2024, and is weighted to the estimated distribution at that time (31% Democratic, 33% Republican). The margin of error for the overall sample is approximately 4.0%.
Image: Getty (karetoria)
What do you think about the election, American politics in general, and everything else? Have your say, join the YouGov panel, and get paid to share your thoughts. Sign up here.
Notes
- We did use AI tools to help produce the typology analysis in this article, however.