A new Ahrefs survey presents an adoption-and-trust gap for content teams. In a sample of 301 marketers, 93% reported using AI to create or help create content, while only 2.7% said they trusted AI-generated content without human review.
What the survey found
The September 30, 2026 report says 69% of respondents publish a larger share of AI content than a year earlier. Among AI users, 85% report faster production, 56% rate human writing more highly and 22% disclose their AI use to website visitors.
Claude was used by 83.2% of AI users, followed by ChatGPT at 75.7% and Gemini at 57.9%. Respondents could choose multiple tools: these figures are usage rates within the sample, not market shares. Ahrefs also cautions that the previous survey had more respondents and that tool-question formats differed.
Trust is a different measure from publishing behavior
The 2.7% result concerns stated trust. It should not become a claim that exactly that proportion publishes without checking. A respondent can distrust a draft yet publish it under time pressure; another can express confidence while retaining a formal approval process. Neither behavior can be inferred from the trust answer alone.
Likewise, use of AI can describe several stages of production. A team might ask a model for an outline, revise an existing paragraph or generate an initial draft. Adoption by itself does not tell a reader how much of a finished article came from a model or how carefully the material was checked.
The editorial implication: measure completed work
For NetContentSEO readers, the practical question is what faster drafting changes in the finished product. An editorial team could measure elapsed time from brief to approved publication, correction effort and the number of claims that remain unresolved. Those are proposed operating measures, not outcomes established by this survey.
Consider a hypothetical product guide. A model produces a polished draft quickly, but the editor discovers that a feature applies to an older version. The relevant cost includes finding and correcting that error, checking comparable claims and updating the conclusion. Drafting speed alone would miss that work.
A useful review process would require the author to retain sources for consequential claims and distinguish evidence from interpretation. The reviewer should check that numbers use the right denominator, that dates refer to the right event and that the headline does not promise more than the source establishes. These checks are particularly valuable when a readable paragraph makes an uncertain assertion feel settled.
Separate factual review from editorial judgment
Checking sources answers whether a claim is supported. Editorial judgment asks whether the article is useful to its intended reader: does it address the actual question, include necessary context and contribute an original explanation? Teams can explicitly assign responsibility for both before publication.
A second AI pass may help flag material for attention, but a publisher should still decide who is accountable for resolving the flagged questions. Asking another model to agree with a draft is not, on its own, a documented connection between the article and its evidence. A practical pilot could compare the issues found by automated checks with those found by an editor.
Make transparency specific
A publication considering disclosure could explain how it uses AI and who approves the final text. For example, a policy might describe assistance with outlines and drafting while identifying the human responsibility for fact-checking. That is a suggested approach, rather than a rule or a measured consequence of the survey.
The Ahrefs findings offer a snapshot of respondents’ practices and attitudes. They do not establish the adoption rate for every marketer, the quality of all AI content or an effect on search rankings. For an individual content team, the useful next step is to test its own process against a clear standard for accuracy, usefulness and review time.