The poll was conducted online between 5 and 7 February 2026, among 2,180 American adults. The sample breakdown reported by Quad is 370 Gen Z, 715 Millennials, 560 Gen X, and 535 Boomers. The full release is on PR Newswire.
One thing to note before going further. This is a commercial survey from a marketing services company, run with a polling partner. It is not peer-reviewed research. The questions were designed by parties with an interest in the answer. That does not make the result wrong. It does mean the result should be read as one well-conducted poll, commissioned with a point of view, and not as settled consumer science.
What the numbers actually say
The headline figure is the cleanest piece of the data. Three quarters of respondents say sponsored placement inside an AI shopping recommendation would reduce their trust in both the AI agent and the brand that paid for the placement. The two numbers are the same. That symmetry matters. Consumers are not separating the platform’s behaviour from the brand’s behaviour. If the system can be bought, both sides take the reputational hit.
Around the headline figure sit a set of supporting numbers. Seventy-three per cent of respondents report that algorithm-driven pricing makes it hard to know whether they are getting a good deal. Seventy-three per cent feel uneasy about how AI might use personal shopping data. Fifty-four per cent find the idea of giving an AI access to their shopping history unappealing. Thirty-nine per cent say they trust AI agents to make everyday purchases on their behalf. Thirty-four per cent say they are comfortable with AI making larger purchases.
Read together, the picture is consistent. Consumers are interested in AI shopping for narrow, defensive reasons. Quad’s release reports that 51 per cent would use AI shopping tools to reduce the risk of making a bad purchase. Two in three use it to spot pricing inconsistencies. Three in five use it to stay on budget. These are uses driven by uncertainty and price pressure, not by enthusiasm for delegating taste.
The OpenAI question
The article title we were given frames this as a finding that bears directly on OpenAI’s recent moves toward monetising shopping inside ChatGPT. We want to be careful here. The Quad and Harris Poll survey does not name OpenAI. It is a general survey about agentic AI shopping. Tying it specifically to OpenAI is interpretation, not data.
What is reportable is that OpenAI has moved decisively — and now publicly — into exactly the territory this survey measures. On 16 January 2026, OpenAI officially confirmed it would begin integrating advertisements into ChatGPT. The rollout went live on 9 February 2026, three days after this survey closed. That timing is worth sitting with: the 2,180 Americans who answered these questions were polled in the final days before ChatGPT ads launched. Their responses describe a pre-launch expectation; we will need subsequent waves to know whether the reality matches the stated concern.
The confirmed format, as OpenAI described it: contextual text ads, placed at the bottom of chat responses, clearly labelled “Sponsored” and physically separated from the conversational answer. OpenAI has stated that ads will not influence ChatGPT’s responses or recommendations, and that paid tiers — Plus, Pro, Business, Enterprise, and Edu — remain ad-free. The free tier and the new ChatGPT Go tier ($8/month) carry ads. The exact mechanics of how paid placement, sponsored results, or revenue-sharing with merchants will evolve beyond this initial format remain to be confirmed. Search Engine Journal and The Information have covered the rollout most closely for those who want the underlying detail.
The Quad finding is relevant to that direction of travel in a structural sense. If three quarters of the addressable market say sponsored AI recommendations would erode trust in both the platform and the paying brand, the platform faces a design problem. Recommendation surfaces work because users believe the recommendations are reasoned. The moment a user can no longer tell whether a suggestion is the best fit or the highest-bidding fit, the surface loses the property that made it useful in the first place.
The critical unknown is whether OpenAI’s chosen format — ads below the answer, labelled and separated — is enough to preserve that perception of independence. The survey measures a general stated preference, not a response to the specific design OpenAI has deployed. That distinction matters.
Why this is different from search
The obvious comparison is to web search, which has been a paid-placement business for two decades and remains the most valuable advertising property on the open internet. The argument from that comparison is that consumers say they hate sponsored results in surveys and then continue clicking on them in practice. The data is well known. So why would AI shopping be different.
The structural answer is that search returns a list. The user does the selection. Sponsored links sit next to organic links and the user reads both. A conversational AI agent does not return a list in the same way. It produces a recommendation, often a single one, in language that resembles considered judgement. The user is not scanning ten blue links and picking one. The user is being told what to buy.
That changes what sponsorship means. In search, sponsorship is a labelled tile inside a larger page that the user navigates. In an AI agent, sponsorship is woven into the sentence that does the recommending. The Quad survey suggests that consumers, asked directly, recognise that difference. Whether they would behave on that recognition once the products are live is a separate question. Stated preference and revealed preference often diverge.
OpenAI’s current format — ads below the answer rather than inside it — is an attempt to hold that structural line. Whether it holds in practice, or whether the boundary erodes as the commercial model matures, is what the next few quarters will reveal.
What brands and platforms appear to be hearing
The Quad release frames the implication as an opportunity for physical retail. Eighty-one per cent of respondents told the survey that a good in-store experience makes them more confident trying the same brand online. Eighty-one per cent agree that it is easier for brands to misrepresent product quality online than in store. Seventy-one per cent say personalised online pricing makes them want to shop in stores, where everyone pays the same price.
That framing serves Quad’s business, which sits on the print and physical-marketing side of the retail stack. It is worth flagging that bias. But the data points themselves are interesting independent of who collected them. They describe a consumer who is willing to use AI for narrow, instrumental tasks while reserving final trust for surfaces they can verify physically.
Libby Rodney, The Harris Poll’s Chief Strategy Officer, is quoted in the release framing the implication for retail this way: “The real competitive question for physical retail isn’t whether a human can out-personalize an algorithm. It’s whether a store can create an experience compelling enough to earn the visit.” That is a defensible reading of the data. It is also, again, the reading most useful to the company paying for the survey.
What to watch
The interesting test will not be the next survey. It will be the first quarter in which AI shopping agents are clearly running paid placement at scale, alongside disclosure that is or is not visible to the user. Three things will be worth watching. Whether sponsored recommendations are labelled in a way the user actually registers. Whether trust scores in subsequent waves of this kind of polling move in the direction the current survey predicts. And whether platforms that resist sponsored placement, or restrict it to clearly demarcated surfaces, gain measurable share against those that do not.
For those tracking the rollout, Search Engine Journal has covered the ad launch in detail; The Information has broken several of the mechanics stories. On the consumer research side, subsequent waves of polling from Harris or comparable firms will be the cleaner signal than any single launch-period data point.
The Quad and Harris Poll survey is one data point. It is a useful one. It should not be read as proving that sponsored AI recommendations will fail. It should be read as evidence that the platforms are now testing, in public and at scale, how much trust their users were extending on the assumption that the recommendation was disinterested.
