23% of AI-referred shoppers leave for a rival after a bad site visit, Contentsquare finds

Contentsquare published survey data on September 2, 2026 finding that only 3% of shoppers would go ahead with a purchase when an AI assistant points them at a brand and the website they land on disappoints them. Twenty-three per cent said they would switch to a competitor, 57% said they would carry on researching, and 16% said they would drop the purchase altogether. The digital analytics company, headquartered in Paris and New York, drew the finding from a survey of 2,000 consumers in the United States and France. PPC Land reported the release on the same day.

The fieldwork ran through Pollfish, the survey sampling platform, across 2,000 consumers in the United States and France. Contentsquare says the questionnaire covered AI use in online shopping, brand discovery and consideration, purchase decisions, agentic commerce and digital experiences, and that every published figure reflects stratified and weighted results across the combined two-country sample. The company offered PPC Land an early look under embargo on August 31, 2026, with publication set for September 2.

The measurement at the centre of the release is the handoff. It asks what a consumer does when an AI recommendation arrives first and the website that follows fails to support it. That is a narrower question than most AI-search surveys ask, and it is the one that sits closest to revenue. Almost every other study published this year measures whether a brand appears in an AI answer. This one starts after that has already happened.

The four outcomes after a poor AI-referred visit

Contentsquare grouped respondents into four reactions to a disappointing site experience that followed an AI recommendation. The shares reported in the September 2, 2026 release are set out below.

Those four numbers total 99%. The release does not account for the missing point, and rounding is the most plausible explanation, but Contentsquare does not itemise it. The figure that travelled furthest is the 23%: close to one in four AI-referred shoppers going to a rival because of what happened after the click rather than because of anything the assistant said. The 3% carries the same message from the other side. Almost nobody pushes through a bad experience just because an AI recommended the brand.

The 21% that matters more than the headline

About one in five respondents, 21%, said an AI shopping assistant gave them information that differed from what they later found on the brand's own website. A further 20% said the assistant did not give them enough detail to buy with confidence. Contentsquare presents both as experience problems. They read more like data problems: a price, a specification, a shipping window or a returns policy stated in the answer and contradicted on the page.

Timing is what makes that expensive. The contradiction surfaces at the point of highest purchase intent, when the shopper has already decided to look. Brands do not control what an assistant says about them. They do control what the destination page says, and this survey attaches a number to the cost of a mismatch between the two.

Independent measurement cited alongside the release points the same way. An audit published on August 18, 2026 by the research firm Empirank examined 1,257 testable factual claims across 182 businesses recommended by a search-enabled OpenAI configuration, and found 20 of those businesses, or 11.0 per cent, carrying at least one claim the evidence contradicted, with a further 479 claims the researchers could not resolve either way. A separate study of 4,776 audited food venues in Bali found ChatGPT and Gemini missing 85.6% of them entirely. Earlier research put 47.1% of marketers encountering AI inaccuracies several times each week.

Where AI now sits in the shopping journey

Thirty-six per cent of respondents told Contentsquare that AI has replaced traditional search engines for at least some shopping activity. Within that behaviour, 53% use AI to find the best price or deal, 52% use it to compare products and 29% use it to discover new brands.

The generational pattern runs against the usual assumption. Millennials, defined in this survey as consumers aged 30 to 44, lead adoption at 74% likely to use AI when shopping online, ahead of Gen Z at 70% for ages 18 to 29 and Gen X at 65% for ages 45 to 60. Millennials also lead on brand discovery at 35% and on price and deal hunting at 57%. The release reports no figures at all for consumers over 60, so the survey says nothing about that group.

On money, Contentsquare reports that 57% of consumers say AI recommendations have already moved real spending. The breakdown that follows is expressed as a share of all consumers rather than of that group: 21% spent between 50 and 150 dollars on AI-influenced purchases, 14% spent between 150 and 500 dollars, and 5% spent 500 dollars or more. Those bands sum to 40%, seventeen points short of the 57% headline. A band below 50 dollars would close the gap, but the release does not say one exists.

The trust figures are comparisons, not absolutes

Two-thirds of respondents, 67%, said they would trust an AI recommendation over that of a social media creator or influencer when evaluating an unfamiliar brand, and 47% said AI has changed their mind about which brand to buy. Both measurements rank AI against a named alternative. Neither establishes how much confidence consumers place in AI answers on their own terms.

Other 2026 research cited around the release points the other way on the absolute question. Yelp and Morning Consult surveyed 2,202 United States adults and found 15% trusting AI search platforms a lot, even though 65% had used one in the previous six months and 63% double-checked results elsewhere. Reddit's Path to Purchase research found half of United States shoppers verifying AI product recommendations on the platform before completing a purchase. RTB House reported in August that leading AI assistants now outrank TikTok, Instagram, Facebook, newspapers and influencers on shopping trust, while 42% of United States respondents said the same tools lengthen the time they need to settle on a purchase. Reddit's own comparison found influencer reviews trailing peer posts by roughly two to one in United States buying decisions, which makes the influencer benchmark a low bar to clear in the first place.

What the survey does not establish

The limits deserve stating plainly, because the headline will be quoted without them. Contentsquare describes the results as global, but a sample drawn from the United States and France is not global. The release does not give the split between the two markets, the fieldwork dates, the margin of error or the weighting variables, and it does not separate United States responses from French ones anywhere in the published data. That removes any way of checking whether the headline numbers hold across both markets or are driven by one. The company has not published the questionnaire or the response-level data either.

The survey also measures stated intent rather than observed behaviour. Asking someone what they would do after a disappointing website is not the same as watching them do it, and the direction of that gap cannot be worked out from what has been released.

The commercial interest is not concealed, and it belongs in the reading. Contentsquare sells experience analytics: session replay, heatmaps, zone-based analysis and frustration scoring across web, mobile and app channels, on a platform the company says more than 1.3 million websites use. Research concluding that post-click experience decides conversion outcomes maps onto what the company sells. That does not make the data wrong. It does mean the question was chosen by a party with a position in the answer.

Jean-Christophe Pitie, chief marketing officer at Contentsquare, framed the release around budget. "As companies pour money into AEO and GEO to make their brands visible to AI, the next question is: what happens after the AI sends the customer your way?" he said. He returned to the consistency point in a second statement: "Brands have spent years trying to create a single source of truth for customers; AI just created another one."

The evidence problem behind AI visibility budgets

The spending question Pitie raises has stayed open through 2026. Adobe research covering more than 500 marketers found 98% without a confident AI search roadmap. Fractl surveyed 343 United States marketing decision-makers and found 81% still saying SEO when they discuss AI search work internally, with only 19% having adopted the GEO label at all. Vendor tooling has moved faster than the vocabulary: HubSpot shipped an answer engine optimisation product on April 14, 2026 and disclosed at the same time that organic traffic for its own customers had fallen 27% year over year.

The evidence for the techniques themselves is contested. A critical survey of 45 studies posted to arXiv on July 15, 2026 concluded that no reviewed generative engine optimisation technique produces a stable cross-platform effect on discoverability or downstream traffic, and that one rewrite scenario cut a page's retrieval by 16%. Read against that, the Contentsquare release argues for a measurement boundary rather than a new channel. Visibility inside an AI answer is an input. The conversion event sits several steps later, on infrastructure the brand still owns and can still fix.

There is a retention angle too. Adobe research published in August found 72% of shoppers deleting retail apps after a single use, and Optimove's holiday data found 53% of United States consumers intending to buy only from retailers they used the previous year. A first impression formed inside an assistant and then contradicted on arrival is a weak starting point for either outcome.

What this means for Thai marketers

Nothing in the Contentsquare release covers Thailand. The sample is American and French, no Southeast Asian market appears in the published data, and the percentages should not be repeated as though they described Thai shoppers. What carries across is the mechanism, and the mechanism is not country-specific. An assistant states something about a brand, the shopper checks the site, and the two either agree or they do not.

For a Thai e-commerce team the practical consequence is that generative engine optimisation stops being a standalone project. If an assistant quotes a price the site contradicts, or a shipping window the site does not honour, the visibility spend has bought a shopper who now trusts the brand less than before the click. Thai retail adds its own version of the problem, because product truth is often spread across a website, a LINE account, a Lazada listing and a Shopee listing that were last reconciled on different days. An assistant reading any one of those can produce an answer the website will not confirm.

Measurement is the second consequence. Most Thai analytics setups report organic and paid traffic without difficulty, but AI referrals arrive with thin or missing referrer data, which makes the post-click question hard to answer with the reports a team already has. Before arguing about AI search visibility budgets, it is worth checking whether the current setup can isolate those sessions at all and compare their conversion rate with everything else. For teams running catalogues, the discipline that keeps feeds accurate for e-commerce marketing is the same discipline that keeps an assistant's answer defensible.

What to check on your own site

The survey prescribes no actions, so what follows is analysis rather than a finding. Four checks come directly out of what it measured.

• Ask an assistant about your own products and compare the answer with the live page field by field: price, stock, specification, shipping time, returns window. The 21% figure is the price of a mismatch, and most mismatches are stale content rather than invention.
• Find where product truth actually lives. If a price changes in one system but not in the feed, the marketplace listing or the help page, an assistant will eventually read the version nobody updated.
• Look at what an AI-referred visitor lands on. A shopper who arrives already recommended sits further down the funnel than a search visitor, and a homepage or a broad category page makes them start again.
• Separate AI-referred sessions in analytics where possible, and compare their conversion rate and exit behaviour against organic search. Without that split, the 23% has no local equivalent anyone can measure.

Contentsquare found that 3% of consumers would complete a purchase when an AI recommendation is followed by a disappointing website, while 23% would switch brands, 57% would keep researching and 16% would abandon the purchase. The company published the survey on September 2, 2026 from a sample of 2,000 consumers in the United States and France, fielded through Pollfish.

Does the 23% figure apply to Thai shoppers?

Not directly, because the sample covers the United States and France only and Contentsquare published no Thai or Southeast Asian data. The mechanism behind the number, a mismatch between what an assistant says and what the site shows, is not specific to any market, but the percentage itself should not be quoted as a Thai figure.

How reliable is this survey?

It is self-reported survey data commissioned by a vendor that sells the tooling its conclusion points toward, and the release omits fieldwork dates, the split between the two markets, the margin of error and the weighting variables. Two of its published breakdowns also fail to sum correctly, at 99% across the four post-click outcomes and at 40% against a 57% headline for AI-influenced spending.

What is the difference between AI visibility and what this survey measured?

Visibility measures whether an assistant mentions or cites a brand, while this survey measures what happens after the resulting click, on the brand's own site. The distinction matters because the first is largely outside a brand's control and the second is not.

What should a marketing team do first?

Compare an assistant's answer about your products against your live pages before committing anything further to visibility work. The 21% consistency finding is the cheapest problem on the list to fix and the one most directly under a brand's control.

Where this leaves the AI search question

The survey does not settle whether AEO and GEO budgets are worth what they cost, and it was never designed to. What it does is move the argument to ground a marketing team can actually stand on, because the conversion step after an AI referral runs on a website the brand owns and can measure. The sequence this data suggests for the coming quarter is to fix the consistency problem first and buy the visibility second. If a second pair of eyes on that sequencing would help for a Thai catalogue, get in touch.