Exploding Topics conducted original research into consumer attitudes to AI-generated online content.

Our survey of 1,000+ web users surfaced some fascinating insights.

We found that trust in AI is low, but that this hasn’t prevented an increased reliance on the technology.

Fast facts

• 42.1% of web users have experienced inaccurate or misleading content in AI Overviews
• Only 18.6% always or usually click through to the sources of AI Overviews
• 21.6% of people think AI has made Google searches worse, but 28.9% have seen an improvement
• Users are almost evenly split on whether AI-generated content makes the internet better or worse overall
• But more than half are less likely to engage with content marked as AI-generated
• Just 1 in 5 people want to see more AI-generated content online
• Three-quarters of respondents are worried about the environmental impact of AI

Most people have issues with AI Overviews

AI is changing the way we browse the internet. But what does all this mean for end-users?

We asked respondents to ignore any examples from social media, and concentrate on their own experience. Even so, 71.15% had personally experienced at least one significant mistake in an AI Overview.

The biggest theme was “inaccurate or misleading content”, experienced by 42.1% of search users. 35.82% have found AI Overviews to be “missing important context”, while 31.5% indicated “biased or one-sided answers”.

16.78% of people have even experienced unsafe or harmful advice from an AI Overview.

Females (34.44%) were significantly more likely than males (21.03%) to say they had not seen any significant mistakes in AI Overviews.

The results are concerning. More than 1 in 10 Google searches trigger an AI Overview, with that ratio more than doubling from January to March 2025.

There have been plenty of viral examples of inaccurate or confusing results.

We gave respondents the chance to describe their specific personal experiences with AI Overviews. There were some notable recurring themes. The crux of the matter is the quality of the information being surfaced. “Incorrect”, “wrong”, and “inaccurate” were all mentioned numerous times.

“Sometimes” was also mentioned a lot, reflecting that consistency is one of the biggest issues faced by AI Overviews.

Our AI sentiment analysis of nearly 400 user responses uncovered a troubling 4:1 negative-to-positive sentiment ratio. Download the full report to discover the specific issues driving user dissatisfaction.

Trust in AI Overviews is weak

Given that the majority of users have encountered at least one significant mistake in an AI Overview, it is not surprising that overall trust in the search tool is low.

Only 8.5% of respondents always trust AI Overviews.

More than 1 in 5 (21.05%) say that they never trust them.

By far the most significant attitude is that users only “sometimes trust” AI Overviews. 61.17% of participants chose this response, meaning around 82% of people are at least somewhat skeptical.

Older people are the most skeptical of AI Overviews by a significant margin. Only 4.3% of respondents over the age of 60 always trust AI Overviews, and 30.94% never trust them.

But trust does not decrease on a linear scale in line with age. In fact, the next-most cautious age group is 18-29-year-olds, only 5.56% of whom always trust AI Overviews.

Conversely, people aged 30-44 are most likely to trust AI Overviews. This age group contains the highest proportion of respondents who will “always” do so, and the lowest who say that they “never” trust.

Low trust in AI Overviews, but limited fact-checking

So 3 in 5 people only sometimes trust AI Overviews, and 1 in 5 people never trust them.

Yet despite that, across all age groups, more than 40% rarely or never click through from AI Overviews to the source material.

Just 7.71% report always following the links provided by AI Overviews, and only 10.97% usually do so.

How do we reconcile that? Claire Broadley, lead editor at Exploding Topics and an SEO and content marketing professional of 15+ years, believes that users are balancing reliability with convenience:

“AI Overviews (and AI Mode) represent a move towards convenience. The danger is that some AI Overviews may be ‘good enough’, and some may be harmful.

“Given these results, it’s clear we can’t rely on people to read our content and check. Businesses will have to be on board with optimizing for AI search visibility, or they risk leaving it to Gemini to join the dots.”

The extent to which content professionals can rely on their audience clicking through from AI Overviews also appears to depend on the household income of the target market. High-earners showed a significantly higher propensity to visit the source material.

56% of respondents with a household income between $175,000 and $199,999 “always” or “usually” clicked on links provided in AI Overviews. 42.1% of respondents earning $200,000 or more did the same, way above the overall average.

Among those with a household income of $10,000 to $99,999, 46.53% “rarely” or “never” clicked on links. Only 14% did so “always” or “usually”.

People who routinely click the links to source material are far more likely to trust AI Overviews.

Among those who say they “always” follow the links in AI Overviews, 62.82% also say that they “always” trust those Overviews.

Conversely, among those who “never” follow links, only 1.96% report “always” trusting AI Overviews.

Get My Free Site Audit 👉

Most users would keep AI Overviews

It’s clear that audiences have a highly complex relationship with AI Overviews. They don’t completely trust them, but nor do they routinely fact-check them — and on balance, they would not get rid of them.

70.62% of people believe that Google search is either the same as or better than it was before the launch of AI Overviews.

And given the choice to enable or disable AI Overviews, only 36.6% would turn them off. 43.03% would turn them on, with a little over 1 in 5 people undecided.

This backs up what we heard earlier from Claire. Users know that the information they are receiving is less reliable, but the convenience trade-off is generally considered worth it.

Despite being the most inclined to fact-check, users with higher household incomes are also more likely to say they would enable AI Overviews if given a toggle option. Only 17.54% of users with household income of over $200,000 would turn the Overviews off.

“AI slop”: Are users really bothered?

We’ve seen a broad range of attitudes to the role of AI in web searches. But what about when the actual content online (and not just the search results) is AI-generated?

The derogatory term “AI slop” is used to refer to low-quality content flooding online spaces. Anecdotally, AI has produced unworkable knitting and crochet patterns, and entirely fictional posts in the Reddit “AITA” forum.

Many of the pages surfaced by web searches have also been crafted with the assistance of AI. The percentage of AI-generated content on Medium is as high as 37.03%.

However, overall user attitudes to artificially generated content are broadly balanced. 39.84% of people believe that AI-created content at least slightly improves the quality of the internet.

That’s more than the 36.94% of people who think AI-generated content has made the internet worse.

On the other hand, people who believe that AI has made the internet worse tend to hold stronger convictions. The majority of those who have seen an improvement think it has been “slight”, whereas detractors are more likely to say things have “greatly” worsened.

Moreover, 50.3% of people would be less likely to engage with content marked as AI-generated. Only 18.51% would be more likely to engage.

Women in particular show less interest in engaging with AI content. 55.57% of women would be less likely to engage with content labeled as AI-generated, compared with 42.54% of men.

Regionally speaking, the Mid-Atlantic appears most receptive to engaging with AI-generated content. Only 34.03% would be put off by an AI label, while 34.72% would actually be more likely to engage.

Conversely, respondents from the West North Central have the least time for overtly AI-generated media. Only 6.38% would be more likely to engage with AI-labeled content, and 57.45% would be less likely.

The AI content Rubicon: This far, but no further

There is no clear consensus on whether AI content is currently a net positive or negative. However, the data takes clearer shape when it comes to desires for the future.

Only 21.78% of people want to see more AI-generated content online. Of those, just 10.79% want to see “much more”.

Meanwhile, more than 1 in 4 would like to see the amount of AI-generated content stay “about the same”.

And despite apparent ambivalence over the current impact of AI on the quality of the internet, 48.12% of users want to see “less” or “much less” AI content moving forward.

In other words, 74.06% of internet users would like to see either a pause or reversal in the amount of AI-generated content online.

There is a significant gender divide on that front. Whereas 15.89% of men wish to see “much more” AI-generated content, only 7.32% of women think the same.

At the other end, 42.54% of men would like to see “less” or “much less” AI-generated content in the future. That figure is almost 10 percentage points higher for women (51.91%).

Additionally, we can see a similar age split to the one we observed in the tendency to follow the links in AI Overviews.

Those aged 30-44 (who checked the sources of AI Overviews least often) are also more likely to want more AI content on the internet in the future. Those aged 60+, who showed the highest levels of skepticism with AI Overviews, are correspondingly more keen on scaling back AI content.

84.89% of this older age group wanted the same amount or less AI content in future.

As with the AI Overviews results, respondents aged 18-29 actually showed up as the next-most AI-skeptic age group.

Only 8.33% favored “much more” AI content in the future, with a further 7.78% wanting “more”. Exactly 25% wanted “about the same”, 23.33% favored less, and 31.67% expressed a preference for “much less”.

And those who want more AI are also the most inclined to trust it. Over 90% of people who expressed a wish for “much more” AI content also said that they sometimes or always trusted AI Overviews.

72.48% of those who want “much more” AI also believe that Google has felt better since AI Overviews were introduced. Just 3.67% think it has felt worse.

Environmental fears around AI

AI Overviews and AI-generated web content both have an environmental impact. One estimate says that generating text takes about 30x more energy than extracting it from a source.

A significant majority of web users are concerned by the environmental consequences of AI. 74.46% are at least a little worried.

More than a third (34.46%) of respondents say that the environmental impact of AI worries them “a lot”.

There is also a highly curious pattern whereby those who want to see more AI are also the most conscious of its environmental impact.

Among those who said they wanted “much more” AI-generated content, 70.64% said that the environmental consequences worried them “a lot”.

Likewise, among those who “always trust” AI Overviews, 68.6% worry a lot about the environmental impact — well above the overall average.

Despite their overall AI skepticism, older people are the most likely to reject environmental concerns.

On average, only 14.21% of internet users aged 18-44 are “not at all” worried about AI’s effect on the environment. Among users aged 45 and over, 19.73% had no concerns, a figure which rises above 20% in the 60+ age group.

What the AI trust gap means for online content

For the most part, internet users are aware of the pitfalls attached to increased AI in search and web content.

They know that it comes with a risk of inaccurate or misleading results. They know that it cannot be wholly trusted. They are even significantly concerned by the environmental impact.

Yet despite all of this, the trade-off for practicality and convenience means that there is only limited appetite for a reversal of the AI developments we have already seen:

• Most people would keep AI Overviews if given the choice, and “about the same” is the most popular answer when it comes to the future levels of AI content online.
• Those who oppose the proliferation of AI Overviews and AI-generated content will often do so in strong terms. But it would be wrong to mistake this as the prevailing view.
• That being said, users’ embrace of AI is qualified and tentative. Most people don’t want to see the internet taken up with more AI-generated content in the future, and anything labeled as AI will face an uphill battle for trust and engagement.

For content marketers, it is clearly necessary to adapt to the world of AI, which is not going anywhere soon. But at the same time, it is vital to recognize and harness the added authority that comes from a human author, and to ensure that all content — regardless of its provenance — is accurate, trustworthy, and valuable to the audience it is designed to serve.

Methodology

The survey comprised 1,115 respondents. Of those, 1,027 said they were aware of an increase in AI-generated content, with the remainder being filtered out of the survey.

Respondents who moved beyond the screener question were asked a further 10 questions about AI and its impact. We also gathered demographic data.

There were 570 female respondents, 392 male respondents, and 10 non-binary respondents. 13 preferred to describe their gender identity in another way, and 25 preferred not to say.

Respondents came from adults throughout the USA across a wide range of ages. Median household income was $50,000-$74,999.