RTB House published a consumer study on 10 August 2026 titled "Who's Buying? Consumer Trust in the Age of Agentic AI", and the number it led with runs against the standard story about AI and the buying funnel. Asked to react to the statement that AI tools increase the time it takes to make a final decision because they provide more information, new choices and a larger set of products or brands to consider, 42% of United States respondents agreed. The study covered 1,840 people in the United States, the United Kingdom, France and Japan, with fieldwork run during June and July 2026 in partnership with the survey sampling platform Cint.
The second finding is about trust. On RTB House's scoring, leading AI assistants now sit above TikTok, Instagram, Facebook, newspapers and influencers as shopping information sources among United States respondents. Both results were reported by PPC Land on 10 August, in coverage that also documents several places where the report's own figures fail to agree with each other. Those criticisms carry as much weight as the findings, and this article covers both.
What RTB House measured, and how
The RTB House consumer trust study is a survey of stated attitudes, and its shape decides what the numbers can carry. RTB House, a performance marketing company founded in 2012, said fieldwork ran across June and July 2026 with Cint as the sampling partner. The sample of 1,840 respondents is split between a United States group and a single combined non-United States group drawn from the United Kingdom, France and Japan. Per-country sample sizes are not published, so the non-United States figures cannot be separated into British, French and Japanese components. Three markets with different retail structures are reported as one unit throughout the document.
The scoring rules matter for reading any number in the report. Trust figures count respondents who selected "mostly" or "very much". Agreement figures combine "somewhat agree" with "strongly agree". All percentages in the published report are rounded to whole points. None of that is unusual, but it means a two point difference between two bars is close to noise, and the report itself does not publish confidence intervals.
The trust hierarchy, and the flip outside the United States
RTB House reported that Google AI Overviews records 43% trust among United States respondents, with ChatGPT level at 43%. Outside the United States the order reverses: ChatGPT leads at 41%, ahead of Google AI Overviews at 35%. The second tier is a long way back. Claude registers 23% in the United States and 20% among non-United States respondents, and Grok sits at 21% and 14% respectively.
The table below carries the AI shopping trust scores the report published for each named source, in both sample groups.
| Information source | US trust | Non-US trust |
|---|---|---|
| Friends and family | 59% | 56% |
| Google AI Overviews | 43% | 35% |
| ChatGPT | 43% | 41% |
| Claude | 23% | 20% |
| Grok | 21% | 14% |
RTB House also expressed the result as point gaps against established media. Among United States respondents, the leading AI tools sit 20 points above influencers, 16 above newspapers, 15 above major media, 11 above both Instagram and Facebook, and 8 above TikTok. The non-United States gaps are wider almost everywhere: 25 points over influencers, 22 over Facebook, 18 over TikTok, 17 over Instagram, 16 over major media and 12 over newspapers.
One source still outranks the assistants. RTB House reported that 59% of United States consumers and 56% of non-United States consumers trust friends and family, against 44% for the leading AI tool. PPC Land noted that the report does not reconcile that 44% with the 43% recorded in its own trust chart, and leaves the one point difference unexplained.
Trust in AI neutrality is still a minority position
The ranking is relative, and RTB House published an absolute measure alongside it that reads very differently. Across the United States sample, 42% agree that AI provides trusted, unbiased information, falling to 37% outside the United States. Millennials in the United States are the most likely group to agree, at 48%. In other words, a clear majority in both samples declined to call AI output unbiased, while still ranking it above social platforms and influencers.
Analysis, not from the source: those two results are compatible once you separate ranking from endorsement. A respondent can consider an assistant the least bad of a set of sources without considering it neutral. For media planning that distinction is the whole point, because a channel that people consult but verify behaves differently from a channel people accept.
The mechanism: a wider consideration set, not a shorter funnel
The central claim rests on one statement respondents were asked to react to: "AI tools increase the time it takes me to make a final decision because they provide more information, new choices, and a larger set of products or brands to consider." In the United States, 29% somewhat agree and 13% strongly agree, which produces the 42% headline. Among non-United States respondents, 26% somewhat agree and 6% strongly agree, summing to 32% in the report's chart, though the accompanying press materials describe the non-United States figure as 33%.
Generational agreement in the United States is highest among Gen Z at 48%, followed by baby boomers at 42%, Gen X at 40% and Millennials at 40%. Outside the United States the spread is flatter: 35% of Millennials, 34% of baby boomers, 33% of Gen Z and 32% of Gen X.
The report gave the behaviour a mechanical anchor. RTB House reported that United States shoppers most commonly visit a site four to six times before completing a purchase, and framed AI as expanding the consideration phase rather than shortening the path to purchase, by bringing more alternatives to a shopper's attention. That is the part worth sitting with. The industry assumption has been that an answering surface compresses research into one turn. The survey response says the opposite happens at the shopper's end: more candidates enter the set, so the set takes longer to resolve.
Discovery skews older in the United States
The brand discovery numbers invert the usual framing. Across the United States sample, 59% agree that AI is good at surfacing brands they did not previously know, against 52% outside the United States. Broken down by generation in the United States, agreement rises with age: baby boomers record 62% and Gen X 61%, both ahead of Gen Z at 52%. The non-United States pattern runs the other way, with Gen Z leading at 55%, boomers at 50% and Gen X at 49%.
RTB House also reported that 63% of United States respondents have used AI to turn a shopping need into at least a general list of products. That is a usage figure rather than a purchase figure, and the report does not attach a completion rate to it.
Where the report does not hold up, according to PPC Land
PPC Land raised specific data quality problems inside the 14-page document, and they are the reason this study should be quoted with the caveats attached rather than lifted as a clean stat. The full PPC Land analysis is here.
First, the generational chart on the decision-time question appears to carry swapped labels. The panel labelled United States shows totals of 33% for Gen Z, 35% for Millennials, 32% for Gen X and 34% for boomers, while the panel labelled non-United States shows 48%, 40%, 40% and 42%. Those are the narrative's United States numbers sitting under the non-United States label. PPC Land judged the chart labelling to be the more likely error because the narrative text agrees with the report's own headline figures, and noted that RTB House does not flag the inconsistency anywhere.
Second, a line under the same section states that 42% of Gen Z shop fashion exclusively on mobile, nearly three times the rate of boomers. PPC Land reported that no supporting chart, question wording or cross-tabulation for that claim appears anywhere else in the 14-page document. A figure with no visible question behind it is a figure nobody can check.
Third, the report puts free or simple returns at 28% overall as a comfort condition for agentic purchases, then describes generational results for the same item ranging from roughly 30% among Millennials to 35% among baby boomers. A range sitting entirely above the stated overall figure cannot come from the same question without further explanation, and the report gives none.
Fourth, the press materials issued with the report state that 46% of United States Millennials have already purchased from a brand they discovered through AI this year. PPC Land reported that the figure appears in the release headline bullets but not in the published 14-page report, which contains no supporting chart or question wording for it.
Finally, the commercial position behind the framing is public. RTB House sells retargeting and on-site engagement technology, and the report's guidance to marketers is that sustained engagement and retargeting matter more as AI expands the research window. A longer consideration window is, mechanically, more retargeting inventory. That does not make the survey wrong. It does mean the finding most useful to the publisher is also the finding the publisher led with, which is a reason to check the question wording rather than the headline.
How to read vendor research without swallowing it
Analysis, not from the source. Four habits cover most of the risk in a report like this one.
- Read the question, not the percentage. The 42% here answers one long compound statement about more information, new choices and more products. Agreement with a compound statement cannot be attributed to any single clause in it.
- Check whether stated attitude is being reported as behaviour. PPC Land made this point directly: the study measures stated attitudes, not observed behaviour, and a willingness expressed in a survey is not a transaction.
- Look for the number in the report body before quoting a number from the press release. Two of the criticisms above exist only because someone opened the actual document.
- Ask what outcome benefits the publisher, then check whether the questions were built to produce it.
What this means for Thai marketers
The study contains no Thai data. Thailand was not surveyed, and Japan sits inside the combined non-United States aggregate rather than being reported on its own, so there is no Asian market figure in this report that can be quoted for Thailand at all. Anyone presenting these percentages as Thai consumer behaviour would be inventing them.
What can transfer is the mechanism, tested locally rather than assumed. If assistants expand the set of brands a shopper considers, the practical consequences show up in measurement settings long before they show up in revenue: longer gaps between first touch and conversion, more sessions per converting user, and a rising share of conversions where the last click is brand search or direct. Those are all things a Thai advertiser can look at in existing data this week without waiting for a Thai survey. Work on visibility inside AI answers belongs alongside that measurement work, which is what generative engine optimisation is for, and the trust ranking against social platforms is an argument for auditing what paid social is actually being asked to do in the plan rather than cutting it on the strength of a vendor survey.
What to check in your own account
Analysis, not from the source. If a wider consideration set is real in your category, these are the settings where it would surface.
- Conversion window length in Google Ads and GA4. A four to six visit path with a longer research phase can fall outside a short lookback, which quietly moves credit to the last channel.
- Remarketing list membership duration. Lists tuned for a fast decision expire before a slow one finishes.
- Assisted conversion reporting. If assists are growing faster than last-click conversions for upper funnel campaigns, the consideration phase is stretching in your data too.
- Branded search volume trend against unbranded. Discovery inside an assistant often resolves as a branded query later, which is a pattern you can watch without any new tooling.
FAQ
Does this study say anything about Thai shoppers?
No. The survey covered the United States, the United Kingdom, France and Japan only, and Thailand was not included. Japan is folded into a combined non-United States group, so even the Asian responses in the sample are not reported separately. Any Thailand read on this is reasoning by analogy and should be labelled that way.
Is the 42% figure reliable enough to put in a client deck?
It is quotable with its caveats attached, and misleading without them. The 42% comes from a single compound statement combining more information, new choices and more products to consider, it measures stated attitude rather than observed behaviour, and PPC Land documented separate labelling and sourcing problems elsewhere in the same document. Quote it as a survey response from vendor-published research, with the sample described.
Does this mean AI search is shortening the funnel or lengthening it?
On this data, lengthening. RTB House framed AI as expanding the consideration phase by putting more alternatives in front of shoppers, and reported that United States shoppers most commonly visit a site four to six times before buying. That is one vendor survey of stated attitudes, so it is evidence rather than proof, and it points the opposite way to the common assumption that an answering surface compresses research.
Do I need to change anything today because of this?
Nothing in the study requires an immediate change. The useful response is measurement rather than strategy: check whether your conversion windows, remarketing durations and assist reporting would even detect a longer consideration phase if one were happening in your account. If they would not, that is worth fixing regardless of whether this particular survey holds up.
Why does the trust ranking put AI above social platforms but only 42% call it unbiased?
Because those are different questions. The ranking is relative, comparing named sources against each other, while the neutrality question is absolute. RTB House reported 42% of United States respondents agreeing that AI provides trusted, unbiased information, against 37% outside the United States, alongside AI tools outranking influencers by 20 points in the United States. People can rate a source highest in a set and still decline to call it neutral.
If the consideration phase in your category is getting longer, the first thing worth knowing is whether your measurement setup can see it. That is a short audit of windows, lists and reporting, and it costs nothing to run before deciding anything larger.
