Human-Reviewed AI News Has an Audience: 43% Would Trust It, Consumer Research Reveals

A small U.S. survey found that people were considerably more comfortable with AI-assisted news when a professional journalist reviewed the work. The results offer a clear signal for newsrooms testing generative AI: editorial responsibility still has to sit with a person.

A survey of 100 U.S. respondents found a meaningful trust gap between an AI draft reviewed by a journalist and an article produced entirely by AI. About 43% said they would completely or mostly trust the journalist-reviewed version. The comparable figure for an entirely AI-created article was 26%.

Distrust followed the reverse pattern. About 24% said they would distrust the journalist-reviewed article, while 52% said they would distrust the article produced entirely by AI. For readers, the presence of a journalist appears to signal more than a final spelling check. It indicates that someone with editorial training has examined the claims, judged the sources and accepted responsibility for what reaches the public.

The survey points toward a practical role for AI in journalism: a tool that helps with reporting tasks while a journalist controls verification, context and publication.

The strongest result came from the comparison between reviewed and unreviewed work:

The 17-point gap in trust is large enough to shape newsroom policy, even with the survey’s small sample. Human review changed how respondents evaluated the finished article, and the change went beyond a modest preference.

That response makes sense because readers usually have limited ability to inspect an AI system’s work. A language model can produce fluent copy while mixing accurate details with unsupported claims, invented citations or missing context. A journalist can check whether a source actually supports a statement, determine whether an important viewpoint has been left out and decide whether the story deserves publication at all.

Those tasks require judgment. They also create a clear person or editorial team that can answer questions when a story is challenged.

The survey’s central message is simple: readers show more trust when a journalist remains responsible for the published version.

The finding fits broader research on public comfort with AI-produced news. The Reuters Institute’s 2026 research described a continuing gap between comfort with human-led news production and comfort with AI-led production. The report also found that people who already trust news tend to feel more comfortable with different forms of AI use. That suggests a newsroom’s existing reputation still matters when it introduces automated tools.

A survey of 100 U.S. respondents found that people were more willing to trust AI-assisted news when a professional journalist reviewed the work. Forty-three percent completely or mostly trusted a journalist-reviewed AI draft, compared with 26% for an entirely AI-created article, while distrust rose from 24% to 52%. Respondents generally accepted AI for supervised tasks such as drafting, summarizing and transcription, but support fell sharply when AI moved closer to publishing authority. The findings suggest that newsrooms have a stronger public case for using AI as an assistant than as an autonomous publisher, provided they verify claims, disclose material use and keep human accountability visible.

Respondents were asked to select the highest level of newsroom AI use they would accept. Their choices showed a preference for human involvement at several points in the editorial process.

The results show a steep drop in acceptance as AI moves closer to publication authority. Drafting, summarizing and transcription received more support because each task leaves a journalist with a visible role in checking the work. Routine story production with occasional review received much less support, and direct publication without human review received the least.

In total, 65% selected at least one use that retained some form of human review. That figure should be read carefully. It does not mean that 65% supported every supervised use, and it does not establish a general approval rating for newsroom AI. It does show that many respondents could imagine a limited role for AI when a journalist continues to evaluate the output.

A newsroom acting on this signal would need to define what review means. A quick read for grammar would fall short of the public expectation suggested by the survey. A meaningful review would involve several checks:

That process treats AI output as a draft with uncertain reliability. It gives the journalist authority to rewrite, reject or delay the work.

Respondents also selected the single measure that would do the most to increase their trust in AI-assisted news. No single safeguard won majority support, which suggests that trust depends on how the reporting can be inspected and who stands behind it.

The leading choices were:

Links to original sources ranked first among the available safeguards. That preference gives readers a way to test important claims for themselves. It also places pressure on newsrooms to preserve the evidence behind a story instead of presenting a polished article as a self-contained product.

Clear disclosure ranked second. Readers appear to want to know when AI played a material role in creating an article, but disclosure alone may leave important questions unanswered. A label can tell readers that AI was involved. It cannot show whether the facts were checked, whether the journalist had authority to change the draft or whether the publisher will correct errors.

Independent editorial oversight and a visible corrections policy address those questions more directly. A named journalist ranked lower as a single preferred safeguard, but a byline still has practical value when it reflects genuine responsibility. The name should identify the person who reviewed the work, confirmed its reporting and can respond to concerns.

The question allowed only one answer. As a result, the percentages show which measure each respondent considered most important, not the total level of support for every measure. A reader could value source links and disclosure at the same time while choosing only one option.

The right question for newsrooms is not whether AI belongs in journalism, but where editorial responsibility must remain non-negotiable. This survey points to a practical boundary: use AI to expand a journalist’s capacity, not to replace the journalist’s judgment. Human review should mean source-level verification, contextual judgment and authority to rewrite or reject the draft—not a quick grammar check before publication. Trust will depend less on the novelty of the tool than on whether readers can see the evidence and identify who is accountable.

The survey produced a more complicated result when respondents imagined a community without a local reporter. About 33% said their preferred approach would depend on the story. Another 20% preferred AI-assisted coverage reviewed by a journalist, and 18% selected fully automated local coverage.

About 19% were unsure. Only 10% preferred no coverage over AI-generated coverage.

The two AI coverage options together accounted for 38% of responses, nearly four times the share who preferred no coverage. That result shows why local news may become a major testing ground for carefully limited automation. A community may face a choice between imperfect automated coverage and a major reporting gap, especially for routine public meetings, election results, weather updates or basic government notices.

Story type still matters. A routine agenda item may be suitable for structured automation with clear checks. A school shooting, contested election, public corruption investigation or health emergency requires source evaluation, sensitivity and context that cannot be reduced to filling a template.

The largest group, 33%, wanted the decision to depend on the story. That response supports a tiered newsroom policy based on risk and complexity. A useful policy could separate assignments into categories:

Such a system gives editors a way to consider the public stakes of a story before choosing a tool. It also prevents the phrase “AI-assisted” from covering a wide range of very different editorial practices.

Generative AI can save time on tasks that consume newsroom capacity without requiring the system to make the final editorial judgment. Transcribing an interview, sorting documents, identifying repeated information and producing a first draft can help journalists spend more time on reporting and verification.

The public response suggests that this support role has a stronger foundation than AI autonomy. Readers may accept assistance because they can still identify the journalist’s role in the process. Once the system becomes the publisher, accountability becomes harder to locate.

A newsroom also has to account for failure modes that are easy to miss in fluent copy:

A journalist’s review should focus on those risks. The job involves more than making the prose sound natural. It involves asking whether the article is true, supported and proportionate to the evidence.

The Associated Press offers one example of how a major news organization has approached this responsibility. Its updated AI standards include disclosure when generative AI plays a material role in published content, while its broader policy retains human oversight and places limits on generative AI in news photography. Such policies do not settle every newsroom question, but they show how AI use can be tied to existing editorial standards.

A useful disclosure gives readers enough information to understand what happened. A vague statement such as “AI was used in the production of this story” leaves too much room for guesswork.

A stronger note could explain:

The level of detail can vary by story. A short automated weather update may need a brief notice. A public records investigation supported by AI-assisted document analysis may need a fuller explanation of the workflow and source material.

The goal is clarity. Readers should be able to distinguish between AI helping with production and AI making the editorial decision.

Research from the Frontiers in Artificial Intelligence review also suggests that disclosure works in context. An AI label does not produce a uniform response across all audiences or story types. Outlet reputation, source cues, political identity, prior knowledge and the subject of the article can all affect how readers interpret the label.

That finding gives newsrooms a practical reason to pair disclosure with evidence. A label may prompt scrutiny. Source links, visible corrections and accountable editing give readers something concrete to examine.

The findings offer a useful snapshot, but they should not be treated as a precise estimate of every U.S. adult’s view.

The survey included 100 U.S. respondents and was conducted through Pollfish on Aug. 19, 2026. The results use post-stratified weights from the Pollfish export. Pollfish generally uses age and gender benchmarks for U.S. post-stratification and may include income when enough information is available, but the supplied export does not identify the exact variables used for this survey.

Pollfish recruits people through opt-in digital experiences. That means the sample was not a probability sample, so a conventional margin of sampling error is unavailable. Weighting can address some demographic imbalances, but it cannot remove possible coverage, nonresponse, wording, order or measurement errors.

The sample size also means that small percentage differences should be handled carefully. Rounded figures can obscure the underlying weighted values, and totals may vary slightly because of rounding. The 43% and 26% trust figures provide a strong directional result, while a larger study would be needed to estimate how stable the gap is across age groups, education levels, political identities, news habits and local communities.

Other research points in a similar direction while adding context. Pew Research Center has reported that most U.S. adults do not get news from AI chatbots, and fewer than 1% prefer chatbots as a news source. Gallup has found that source verification and human editorial verification are among the measures that can increase trust when news organizations disclose AI use. A preregistered study in Scientific Reports also examined differences between human-written, AI-assisted and fully AI-generated news, including how disclosure affects perceived quality and intended engagement.

Together, these findings suggest that public expectations are still forming. People may use AI tools in daily life while holding a higher standard for news, where accuracy, public accountability and source transparency carry special weight.

The survey gives editors a starting point for a policy that readers can understand. A responsible newsroom could build its approach around five rules:

This approach allows a newsroom to use automation without handing over editorial authority. It also gives readers a basis for deciding how much confidence to place in a story.

The survey’s most useful finding is the boundary it draws. Many respondents could accept AI in a newsroom when a journalist reviewed the work. Very few accepted publication without human review. That boundary gives editors a direct public signal about where accountability should remain.

For now, AI has a stronger case as a newsroom assistant that extends a journalist’s capacity. Trust depends on the journalist checking the reporting, making the judgment and standing behind the published article.