Skip to main content
Asklet
Sign in Start free
All posts

Measuring content

Do readers trust AI-written articles?

By 7 min read

Many teams now use AI somewhere in their writing: for research, outlines, first drafts or whole articles. The question that follows is whether readers mind, and whether it costs you their trust.

Below is what recent surveys found, what they can't tell you about your own site, and how to measure it on your pages.

The short answer

In general, many readers say they trust it less. Surveys find most people are comfortable with content made by humans and far fewer with content made entirely by AI. Those surveys measure attitudes to AI as a whole, though, not whether a given page helped. To learn how your own AI-assisted pages do, ask readers on each page whether it helped, and compare those pages with similar ones written without AI.

What the research says about trust in AI-written content

Three recent studies asked readers about AI-made content in different ways:

  • The Reuters Institute's Generative AI and News Report 2025 surveyed about 2,000 people in each of six countries (Argentina, Denmark, France, Japan, the UK and the US). On average, 62% said they were comfortable with news made entirely by humans and 43% with news led by a human with some help from AI. Only 21% were comfortable with news made mostly by AI with a human in the loop, and 12% with news made entirely by AI. More respondents expected AI to make news less trustworthy than more trustworthy.
  • A 2026 survey of 444 US consumers by Clutch and Compose.ly found that 66% are more sceptical of fully AI-written content.
  • Trusting News and Benjamin Toff of the University of Minnesota tested AI disclosures with 10 newsrooms. In their survey, 42% said a note that AI had been used made them less likely to trust a story, 30% said more likely, and 28% said it made no difference.

Across all three, the more of the work AI does, the fewer readers are comfortable with it. In the Reuters Institute survey, comfort rose from 12% to 43% when a human led the work.

What these surveys can't tell you about your pages

The studies ask people how they feel about AI in content. That is a useful warning, but it is a different question from whether a particular page helped the person reading it. A few things to keep apart:

  • Most of the research is about news. A product guide, a recipe or a docs page is read differently, and readers may judge it on whether it worked.
  • Answers about a label are not answers about a page. In the Trusting News work, the disclosure alone moved trust, up for some readers and down for others.
  • "AI-written" covers everything from a spell check to a page no one edited. Readers in these surveys were reacting to broad categories, not to your process.

So the research tells you the risk is real. It can't tell you whether your own pages have it. Your readers can.

Ask whether the page helped

The test for an AI-assisted page is the same as for any other: did the reader get what they came for? Ask it the same way on every page, so the answer doesn't depend on how the page was made:

  1. A Yes / No question at the end

    Such as "Did this page answer your question?". One tap, so it gets the most answers, and every page gets a figure you can compare.

  2. A follow-up that asks for the reason

    A multiple choice question such as "What would make this page better?" with options like "More specific detail", "Fix a mistake", "Clearer steps", "More up to date" and "Something else". "More specific detail" and "Fix a mistake" give readers a quick way to say a page feels generic or wrong, whoever wrote it.

  3. An optional comment

    Last, for readers who want to point at the sentence that lost them.

Tip: Don't ask "Do you think AI wrote this?". It measures how suspicious readers are, and it plants the idea on pages a human wrote.

Compare AI-assisted pages with similar pages

One page's score means little on its own. Comparing groups of pages tells you more:

  • Group pages by how they were made, such as "AI-drafted, edited" and "written without AI". If they live in different sections, compare the sections. If they are mixed, add the AI-assisted pages to their own theme.
  • Compare like with like: AI-drafted how-tos with human-written how-tos, not with news.
  • Use the same question, in the same place, over the same dates for both groups.
  • Wait for at least 10 answers per page, and more per group, before you read the percentages.
  • Read the reasons, not only the score. If AI-assisted pages lose on "More specific detail" and match on everything else, you know what the editing step should add.

If you disclose AI use, check what it does

Trusting News recommends that newsrooms disclose AI use and explain why they used it and how people checked the result. Whatever you decide, you can watch what a note does to your own readers:

  • Add the note to one group of pages and leave a similar group without it for the same weeks.
  • Compare the share of Yes answers and the reasons readers pick, before and after, and between the two groups.
  • Read the comments for mentions of the note itself.

This shows whether readers who see the note still find the page helpful. It doesn't tell you whether to disclose, which is your decision and may also be a legal one in some places.

Example

Example: a personal finance site trying AI drafts

Say you run a personal finance site. You start drafting your glossary pages with AI, and an editor checks and rewrites each one before it goes out. Your longer guides stay written by staff. All pages end with "Did this page answer your question?" and "What would make this page better?".

After two months, the glossary under /glossary/ has 120 answers and your guides under /guides/ have 300. The share of Yes answers is close for both. The difference is in the reasons: on the glossary, "More specific detail" comes up far more often, and the comments ask for examples with real amounts.

So you keep the AI drafts and change the editing step: every glossary page now gets a worked example before it goes out. Two months later you compare the same question on the same pages over the same number of weeks.

Where Asklet fits

Asklet shows the same short survey at the end of every page from one snippet, keeps the path each answer came from, and shows results per page, per path and per theme. It shows counts until a page has 10 responses, so a few early answers can't decide the comparison for you.

It has no way of knowing how a page was written, so the grouping is yours to set up. Readers answer without signing in, and the widget sets no cookies and uses no tracking identifiers.

Your next steps

  1. List which pages are AI-assisted, and how

    For example "AI draft, edited by a person" or "AI outline only".

  2. Ask the same Yes / No question on every page

    Whatever wrote the page, the question is whether it helped.

  3. Add a reason question with "More specific detail" and "Fix a mistake"

    They give readers a quick way to say a page feels generic or wrong.

  4. Group AI-assisted pages so you can compare them

    By section or by theme, against similar pages written without AI.

  5. Compare the reasons, not only the score

    The gap in reasons shows what your editing step should add.

  6. Test any disclosure note on one group first

    Compare the same weeks with and without it.

Try these steps on your own site. Asklet adds the questions to every page with one snippet. Free plan, no card needed. Compare the plans.

Start free

Common questions

Will readers trust my site less if I use AI?

Some will, according to surveys: many readers say they are less comfortable with content made mostly or entirely by AI. Whether your pages lose trust depends on how you use it and how good the result is, and you can measure that on your own pages.

Should I tell readers that AI helped write a page?

That is your call, and in some places it may also be a legal question. Trusting News found that disclosures can lower trust for some readers and raise it for others, and recommends explaining why AI was used and how people checked it. You can compare reader answers with and without a note.

Can readers tell when a page is AI-written?

Sometimes, and not reliably. That is why the useful question is whether the page helped: readers can always answer that one.

How many answers do I need to compare two groups of pages?

At least 10 per page before you read percentages, and enough per group that one unusual page doesn't decide it. The sample size calculator shows how the margin of error shrinks as answers grow.

Related

Sources

More from the blog

Get your first answers today.

Set-up takes a few minutes, and the free plan covers one website or app.

Start free

Free plan, no card needed.

en