Ahrefs surveyed 301 marketers for the second year running, and 280 of them use AI to make content. Some of the reported changes since 2025 are larger than the sampling noise and some are not. The survey does not say how respondents were found or when it ran.
Table of contents
Key takeaways
- 93% of 301 respondents use AI to create content, against 87% in 2025. Our 95% margin on the 2026 figure is about ±3 points.
- Using AI to check AI output doubled, from 14% to 28.6% of 280 AI users. That change is well outside the noise.
- The rise in mostly unedited AI content, from 4% to 7%, sits inside a ±3-point margin. It may be noise.
- 2.7% trust AI content with no human review. If the base is all 301 respondents, that is about eight people.
What this is about
Glen Allsopp of Ahrefs published the second annual survey of how marketers use AI on September 30, 2026. Of 301 respondents, 280 (93%) use AI for content. Most say they publish a larger AI share than a year ago. The question for a reader is which changes since 2025 a sample this size can actually support.
The design
This is a self-reported survey, not an experiment. It measures what marketers say they do, not what they publish. There is no control group and no check against real output.
What the article states:
- Sample: 301 respondents, 280 of whom use AI. Most percentages below use the 280 as their base.
- Comparison: a 2025 survey with more respondents. Allsopp writes that, for that reason, comparisons will cover only high-level answers.
- Question changes: the 2025 tool question took one answer. The 2026 version allowed several, so tool shares are not comparable year on year.
What the article does not state: when the survey ran, how respondents were recruited, how many took part in 2025, or the job and company-size mix behind each figure. Ahrefs sells an AI writing product, which the article promotes at the end. That does not make the answers wrong. It does mean the publisher has an interest in the result.
The numbers
The margins below are ours. We calculated a standard 95% interval from each reported share and its base. That formula assumes a random sample, and this one is not random. So read each margin as the least uncertainty the figure carries.
| Measure | Base | 2025 | 2026 | Change | 95% margin, 2026 |
|---|---|---|---|---|---|
| Use AI to create content | 301 | 87% | 93% | +6 pts | ±2.9 pts |
| Use AI to check AI output | 280 | 14% | 28.6% | +14.6 pts | ±5.3 pts |
| Use AI for data analysis | 280 | 28% | 60% | +32 pts | ±5.7 pts |
| Publish no pure AI content | 280 | 50% | 40% | −10 pts | ±5.7 pts |
| Disclose AI use to readers | 280 | 16% | 22% | +6 pts | ±4.9 pts |
| Over 75% of output is pure AI | 280 | 4% | 7% | +3 pts | ±3.0 pts |
Source: Glen Allsopp, Ahrefs, September 30, 2026. Margins calculated by SEO After AI. The 2025 sample size is not published, so the margin on each change is wider than shown.
Read top to bottom, the table sorts itself. The checking, data-analysis and no-pure-AI rows moved far more than any plausible noise. The adoption row probably moved, but six points against a ±2.9 margin, plus an unknown margin for 2025, is a modest claim. The disclosure and pure-AI rows could be noise. A three-point rise with a three-point margin is not a finding.
Several figures have no 2025 comparison:
- 85% of AI users say they produce content faster, 47.9% much faster. 2.5% say AI slowed them down.
- 74% of AI users say their AI share grew. Among those who publish no pure AI content, 58% still say it grew. So “more AI” mostly means more assisted writing, not more unedited text.
- 56% of AI users rate human writing above their own AI output.
- 41% of AI users say AI saved them nothing on writers, and 26% do not know. Only 28% work with outside writers at all.
The split behind the quality question is the most interesting result, and the weakest. 43% of founders and owners say their AI content beats human writing, against 17% of junior and mid-level marketers. The article does not give the size of either group.
What could still be wrong
Who answered. A survey promoted by an SEO tool vendor reaches that vendor’s audience. People who use AI heavily are more likely to answer a survey about AI. Both effects push adoption figures up. The 93% describes these respondents, not marketers in general.
Two different samples. The 2025 and 2026 groups are not the same people, and the 2025 group was larger. If its job mix differed, part of each change is a change in who answered. Allsopp flags this himself, which is to his credit.
Changing questions. The tool question changed from one answer to several, and some write-in answers were folded into existing options. Any reading of tool shares as market share is wrong by design. The article says so.
What people say versus what they do. “7% publish mostly pure AI content” is a claim about oneself. Our own detector test on 12 of our articles showed how far labels and measurements can drift apart.
The founder split. Without group sizes, 43% against 17% could rest on a few dozen answers in each group. It is a lead, not a result.
What to do on Monday
1. Quote this survey for the large changes only. The rise in AI-checking-AI and the drop in “no pure AI content” survive any reasonable doubt. The disclosure and pure-AI rises do not. 2. If you check AI drafts with another AI, add one step a model cannot do. Have a person check each number against its source. A second model reads the same draft with the same blind spots. 3. Run the same survey inside your team, with dates. Ask what share of last month’s published pages was mostly unedited AI. Then sample ten pages and check. You will learn more from the gap than from either number. 4. Track your AI share against outcomes, not against the industry. Our notes on measuring search after AI answers cover what still counts as a signal.
FAQ
Does this survey show that AI content performs worse?
No. It measures opinions and habits, not traffic or rankings. 56% of AI users rate human writing higher, but that is a judgement of quality, not a measurement of results.
Why is Claude listed as the most-used tool?
83% of AI users chose it, against 76% for ChatGPT and 58% for Gemini. The question allowed several answers this year and one last year, and the sample is one vendor’s respondents. It describes this group, not the market.
How many people does the 2.7% represent?
The article does not state the base for that question. If it is all 301 respondents, about eight people trust AI content with no human review. Any percentage that small should be quoted with its count.
