More than half of US consumers say they sometimes or often bypass company websites while researching with AI tools, yet fewer than one in five always trust an AI product recommendation without doing further research, according to survey data Kinsta published on September 22, 2026. The poll of 1,008 people, commissioned by the managed WordPress hosting provider, describes a buying journey split in two: chatbots draw up the shortlist, and a brand's own pages are left to confirm it.

In Short

Kinsta, a company that hosts websites, asked 1,008 people in the US how they use AI chatbots when looking into products, and most said they use them but do not fully believe what they are told. That matters for any business with a website or an advertising budget, because a large share of those people still go to the company's own site to double-check an AI suggestion before deciding. You can think of the company site as moving from the place where shoppers first find a brand to the place where they check whether the AI got it right.

What was measured

Propeller Insights ran the national survey for Kinsta, which put the results on its site on September 22, 2026, alongside a press release datelined Los Angeles the same day. According to Kinsta, the margin of error is plus or minus 3 percentage points at a 95% confidence level. Neither the report page nor the release gives fieldwork dates, a weighting scheme, a panel source or the wording of individual questions, and several results are reported for subgroups whose size is not stated. The 3 point margin applies to the full sample; for any smaller base, it is wider.

The findings are organised under four headings - discovery, trust in AI, websites as verification, and reviews and authenticity - and every figure is self-reported. None of it is observed browsing behaviour, a distinction that becomes important once the survey is set beside panel and server-log data gathered elsewhere.

AI takes the first pass

According to Kinsta, 59.2% of respondents use AI tools often or occasionally for research. Some 53.8% sometimes or often skip websites entirely during that research, and 61.2% believe AI will eventually replace websites as the primary way people discover products.

The 53.8% figure is the one Kinsta leads with, and its base is not clear. The report places it directly after the 59.2% result and ties it to that research, which reads as a share of AI users. The press release presents it as a share of all consumers. Measured only among the 59.2% who use AI tools, 53.8% would correspond to roughly 31.8% of the full sample. Measured across all 1,008 respondents, it would imply that the great majority of regular AI users skip sites at least some of the time. The published material does not say which reading is correct.

Kinsta also asked respondents who often skip websites and rely on AI answers why they do so. According to the company, 52.2% said AI is faster than searching multiple websites, 45.2% said it makes follow-up questions easier, 42% pointed to too many ads or pop-ups on websites, and 37.6% said traditional search results feel cluttered or SEO-driven. The four figures add up to 177%, so respondents were evidently able to choose more than one reason.

Two of those four reasons are complaints about the open web's own economics. Among the respondents most inclined to bypass websites, four in ten cite advertising and pop-ups as a reason to stay inside a chatbot - a finding with direct relevance for ad-supported publishers whose page layouts are paid for by exactly those formats. The complaint about cluttered results arrives while Google's own generated summaries absorb clicks that once went to publishers. Ahrefs research published on February 4, 2026 put the click-through reduction for top-ranking pages at 58% when AI Overviewsappear, and a randomised field experiment involving 1,065 desktop Chrome users produced the first causal estimate: a 39.8% fall in outbound organic clicks, alongside a 34.5% rise in zero-click searches.

Trust has not followed usage

Adoption, on these numbers, has not produced confidence. According to Kinsta, 58% trust content less once they suspect it was AI-generated, 52.1% trust traditional search results more than AI-generated answers, and 90.7% are concerned about AI-generated misinformation. Only 18.8% always trust an AI product recommendation without doing additional research.

The 52.1% result deserves a caveat of its own. With a margin of plus or minus 3 points, the true figure could sit anywhere between 49.1% and 55.1%, which makes it statistically indistinguishable from an even split. The press release describes consumers trusting traditional search more than AI; the data support a narrower reading, that the population is roughly divided.

Respondents also believe synthetic material is everywhere. Some 86.2% think they encounter AI-generated content often or occasionally online, and 77.3% are very or somewhat confident they can identify it. The survey did not test that confidence. It is a self-assessment by people who, elsewhere in the same questionnaire, admit to doubting the reviews they read.

The trust figures fall within the range other research has produced, though question wording varies too much for direct comparison. A Yelp and Morning Consult study of 2,202 US adults, covered by PPC Land in April, found 65% had used an AI search tool in the preceding six months but only 15% trusted the information a lot, while 63% double-checked results against other sources. Equativ's survey of 4,000 people across North America and five European countries recorded 21% trusting AI completely and 60% cross-checking in October 2025.

Raptive's research from July 2025 is the closest match to Kinsta's 58%. It measured a near-halving of reader trust among 3,000 US adults where content was suspected of being AI-generated, alongside a 14% drop in purchase consideration for products in adjacent advertising.

There are grounds for the caution. An audit by the research firm Empirank, published on August 18, checked 1,257 testable claims about 182 businesses recommended by a search-enabled OpenAI configuration and found that 20 of them, or 11.0%, carried at least one claim the evidence contradicted, with a further 479 claims left unresolved. Misinformation in AI answers takes several forms, from stale records to hallucination, the fluent output that has no basis in fact.

The site visit as a checkpoint

Kinsta's central argument rests on what happens after the AI answer. According to the company, 44.7% of respondents always or most of the time visit a company's website after receiving an AI recommendation, and a further 25% do so occasionally. Together, that is 69.7% who visit at least some of the time. The report does not break down the remaining 30.3%, so it is not possible to say how many never check a brand's own site at all.

A related result gives that visit its weight: 61.1% associate the standard of a company's site with the standard of its products or services. For a brand, that turns the page a sceptical visitor lands on into a proxy for the thing being sold.

Independent measurement points the same way, though it also shows how poorly the checking visit is recorded. Similarweb clickstream research published on June 21, based on six months of US desktop panel data, found that brands recommended by ChatGPT were 2.5 times more likely to receive a site visit within seven days, and that 55.9% of AI-influenced traffic arrived through search rather than as a referral from the assistant. In analytics terms, much of the checking Kinsta describes is likely to be credited to branded search, not to the chatbot that prompted it.

What happens when the destination disappoints has also been measured. Contentsquare's survey of 2,000 consumers in the United States and France, released on September 2, found that only 3% would complete a purchase if the website following an AI recommendation let them down, while 23% would switch brands and 16% would abandon the purchase. Some 21% had come across AI details that contradicted a brand's own page.

The brand site is also not the only place verification happens. Reddit's Path to Purchase research, presented on June 22and based on 13,956 US and 6,485 UK adults surveyed by Attest, found half of US shoppers checking AI recommendations on Reddit before buying. Kinsta's questionnaire, as published, asked only about company websites, so it cannot show how the two routes compare.

Speed, uptime and security

Kinsta's release goes further than its data. It argues that slow pages and broken experiences can make a company feel less credible, and that a site's speed, uptime and security are what confirm or ease the scepticism of visitors referred by AI. The published survey contains no question about any of those three attributes. The 61.1% result concerns a general judgement about the site, and the connection to infrastructure is Kinsta's interpretation - one that happens to describe the product it sells.

Behavioural evidence on performance does exist, though it measures purchases rather than opinions. Shopify analysed Core Web Vitals across its merchant base over 28 days in January and February 2026 and found conversion about 3.5% lower for every additional 100 milliseconds taken to load a store's main content. Those are platform-reported, correlational figures.

Reviews, bots and synthetic content

The authenticity questions produced some of the survey's highest numbers. According to Kinsta, 70.9% have suspected a product review was fake or AI-generated, 59% have stopped engaging with content because they suspected it was AI-generated, and 43.3% say knowing bots are active online makes them trust websites less.

Regulators and platforms have acted on the review problem, if not on consumer perception of it. The US Federal Trade Commission finalised a rule on August 14, 2024 prohibiting fake reviews and undisclosed paid testimonials; it took effect in October 2024 with civil penalties of up to $51,744 per violation. Google added a search guideline in July 2026requiring disclosure of incentivised reviews, enforceable by manual action. Coordinated fake praise is a form of astroturfing, and a result in which seven in ten consumers say they have suspected a review suggests the doubt now attaches to reviews generally, genuine ones included.

The bot question carries a commercial echo. On June 9, Kinsta added Bot Protection to all of its hosting plans at no extra cost, and on June 18 it published an analysis of more than 10 billion HTTP requests across its infrastructure. That analysis found a single crawler sending 3.75 million requests to WordPress shopping cart URLs in 24 hours. Cloudflare Radar data for the week ending June 5, 2026 put bots at 57.4% of HTML web traffic. The survey's 43.3% measures sentiment rather than traffic, but it comes from a company with a product aimed at both.

What Kinsta says

Matt Reid, chief marketing officer at Kinsta, framed the findings as a sequence. "People are using AI to narrow their options, then coming to websites to make sure the recommendation checks out," he said in the release. "Websites aren't becoming less important because of AI. If anything they've become the crucial last line of defense against bad information."

The release also quotes two executives from outside the company, without describing their relationship to Kinsta. Cody Landefeld, co-founder and partner at Mode Effect, said: "Our clients tell us that people don't take AI-generated content or reviews at face value anymore." He added: "When someone lands on a website after getting a recommendation from AI, they're already asking themselves whether they can trust what they're looking at."

Jeremy Fremont, director of business development at Multidots, put the division of labour more bluntly. "AI has changed how people discover a brand, but buying still remains largely the same," he said. "Consumers still make that call once they are on the website, after they've had a chance to check things out for themselves."

Kinsta says it serves more than 230,000 customers across 128 countries. A hosting company has an obvious interest in the proposition that websites still matter, and that their performance determines credibility. Neither interest invalidates the survey, but both shape which findings the release chose to lead with.

Where the release and the report part ways

Set side by side, Kinsta's report page, its press release and the promotional email circulated on its behalf to PPC Land on September 22 differ at several points beyond the unclear base of the 53.8% figure. The release says most consumers now skip company websites, when the report's measure counts those who do so sometimes as well as often. It also says 18.8% trust AI product recommendations without additional research; the report specifies those who always do, a narrower group, and does not publish how many sometimes act on AI suggestions unchecked. The promotional email describes the 44.7% as visiting a company's site before buying, a purchase condition the report page does not attach. And the release headline says consumers still rely on websites over AI, although the closest measure concerns trust in search results rather than websites, and a 61.2% majority of the same respondents expect AI to replace websites as the main route to product discovery.

Stated habits against measured traffic

Surveys of attitude and logs of behaviour do not always agree, and the Kinsta figures invite the comparison. Datos's first-quarter 2026 State of Search report, published in April, found AI tools collectively accounting for less than 2% of total desktop web visits in the United States and Europe. A Semrush analysis of ten US industrial categories, covered by PPC Land this month, put AI assistants and AI Mode together at 0.48% of sessions between January and July 2026. Microsoft Clarity data from more than 1,200 publisher and news sites, reported in December 2025, showed AI referrals below 1% of traffic but converting to sign-ups at 1.66%, against 0.15% for organic search.

These are not contradictions. A share of visits and a share of people who sometimes do something are different measures, and a respondent who skips websites during some AI sessions may still generate most of their traffic through search and direct visits. The gap in scale is nonetheless a reminder that the 53.8% describes a habit, not a volume. What has changed is the point in the journey at which many of those visits now arrive.

Why this matters for marketers

For advertisers and publishers, the survey's significance lies less in its headline figures than in the sequence it describes. If assistants handle discovery and brand sites handle confirmation, the pages that count are the ones a sceptical visitor reaches after an assistant has already named the brand. When that visit arrives as a branded search, paid and organic search teams may be credited for demand generated inside systems they cannot observe.

The trust results matter to brands experimenting with AI-generated copy and reviews, a gap the IAB's own research in January 2026 put in sharp terms: 82% of advertising executives believed Gen Z and Millennial consumers felt positively about AI-generated ads, against 45% of those consumers who actually did.

The limits of the evidence are plain. The sample is American, so it says nothing directly about European markets. The research was commissioned by a company that sells the infrastructure its conclusions favour, and it omits the fieldwork dates and question wording that would allow independent scrutiny. Read with those caveats, the survey is directional evidence from an interested party - and it is consistent with a growing body of independent work suggesting that AI has shortened the path to a shortlist without shortening the path to trust.

Timeline

Summary

Who: Kinsta, a managed WordPress hosting provider, commissioned Propeller Insights to survey 1,008 US consumers. The release quotes Matt Reid, Kinsta's chief marketing officer, along with Cody Landefeld of Mode Effect and Jeremy Fremont of Multidots.

What: The survey found 59.2% of respondents use AI tools often or occasionally for research and 53.8% sometimes or often skip websites during that research, while only 18.8% always trust an AI product recommendation without further research. Some 44.7% visit a company's site always or most of the time after an AI recommendation, and 61.1% associate the standard of a company's site with that of its products or services.

When: Kinsta published the report and a press release on September 22, 2026. Fieldwork dates were not disclosed.

Where: The sample covered consumers in the United States. The release was datelined Los Angeles, and the report sits on Kinsta's website.

Why: Consumers are routing early product research through AI assistants without extending full trust to them, which shifts part of a brand site's job from discovery to verification. For marketers, that affects how AI-influenced visits are attributed, how ad-heavy pages are perceived and how suspicion of synthetic content weighs on credibility. The data are self-reported and come from a vendor whose products align with its conclusions.