Research
9 July 2026
A payment brand's customers now get their trust question answered by a machine, and that machine can get the brand's facts wrong, dig up old trouble, and hand the customer to a competitor, all while the brand is not watching. WhiteLobby checked how 93 payment brands appear across Google and the AI answers from ChatGPT, Gemini and Perplexity, in 47 markets, over two weeks. The AI layer also disagreed with what Google showed for the same brand the same day, and none of it surfaces in the rank-tracking most brands rely on. This is the new front line of brand protection. Below is the method, the data, and what it looks like across the industry.
| Payment brands checked | 93 |
| Home markets and languages | 47 |
| Live Google result pages read | 734 |
| Ranked results logged | 45,920 |
| Google AI Overviews read across the 93 brands | 352 |
| Sources cited inside those AI answers, each checked against the live page | 5,243 |
| Brands holding position one for their own name | 68 |
| Plus a deep dive, full chat answers on 5 brands across ChatGPT, Gemini and Perplexity | 60 |
WhiteLobby started from the real branded keywords people type for these 93 payment brands, not queries we invented, and ran them at full top-100 depth in each brand's home market and language. That produced 734 live result pages and 45,920 ranked results, which we stored and read rather than sampled. Nineteen further captures failed on a request-parameter error and returned nothing, so we treated those as data gaps, not findings, and left them out of every count.
That is the AI layer for all 93 brands. We pulled the 352 Google AI Overviews those queries triggered and went through all 5,243 sources the overviews cite. Then, as a deep dive on five brands chosen to span the categories and home markets, from e-wallet to open banking to prepaid, we opened the actual ChatGPT, Gemini and Perplexity from inside each brand's own country and asked the four questions a nervous customer asks. What the brand is, whether it is safe, what its reviews say, and whether it is a trustworthy payment provider. Sixty answers in all, checked line by line against the live record. For domains carrying a brand's name we resolved the network, opened the page, and traced where the logins and links lead.
Every number here we checked against the live source the same week. That is the part of this research another firm cannot reprint, because the work was in the reading and the checking, not the export.
We counted the domains across all 45,920 results, and the ones that appear most across a brand's search results are not the brands' own. In order of how often they surface.
Narrow it to the reputational cluster, the is-it-safe, reviews and complaints queries and their local-language forms, and an independent review or complaint platform sits in the top 10 for more than two-thirds of them. By our count, 190 of 277.
We will state the part most vendors will not. This is not, by itself, a failure. A person asking for reviews wants other people's opinions, so the engine shows other people. Google has worked this way for a decade and the answer engines copied it. What it raises is the quieter question of what those third parties say about a brand, and whether any of it is wrong, or years stale, and left standing. Knowing that answer is the professional move. Chasing first place on "is your brand legit" is the amateur one.
We checked every claim the engines made against the live page. The answers went wrong in ways that cost a brand, and in more than one direction.
Wrong facts. On brands whose live Trustpilot score sat in the low twos, one engine called a 2.2 "good to great," and another called a 2.7 "relatively high, around 3.5 to 4." A point-and-a-half error on a public rating would be caught in any analyst review. An engine makes it to a very large audience and no one in the building hears it. For the record, ChatGPT did not overstate a single score in these captures. The two clear errors came from the other two.
Old trouble, resurfaced. A years-old regulator event reappeared in one engine's answer, with the regulator's page linked, while another engine asked the identical question the same day never raised it. The brand does not decide when its history comes back up. The engine does.
A competitor, named. Several answers closed by recommending a rival by name. For a business that earns per transaction, this is the line that costs money. A trace of doubt in the question turns the answer into a comparison the engine assembled, and the brand did not choose it.
The engines did not even agree with each other. Same brands, same day, same questions, different verdicts on our own screen.
It goes past the chatbots. One prepaid brand, French market. Google ranks a national police fraud-warning page in the top three for its trust queries, while the same day one AI engine calls the brand reliable and never mentions fraud. Read only the rankings and you see a fight. Read only one chatbot and you see calm water. Both are wrong about what the customer actually meets, because the customer might hit either. WhiteLobby reads them together, which is the whole point of doing it this way. For the record, the fraud there abuses the product rather than the brand, and the brand publishes its own warnings about it, which is exactly why the strict engine can cite them.
Hundreds of domains carrying these brands' names surface in the top 20 without being the official site. We classified 380 of them by hard signal, resolved network, live page, and where the logins and links go.
Hand-reviewing those 68 leaves about a dozen genuine impersonation candidates, several on a single shared host, pointing at one another. Even those we decline to label copycats, and we will defend that. From the outside, a mirror site is indistinguishable from a brand's own alternate domain, and only the brand knows which ones it runs. So what WhiteLobby hands a brand is a shortlist to confirm or disown, per brand, that clears in ten minutes. We would rather lose the dramatic headline than publish a number we cannot stand behind.
For one brand's safety query, its own pages supplied 13 of the 23 citations in Google's AI Overview, a majority, because the brand had published real, specific safety content and the engine reached for it. Across 93 payment providers it was the clearest case of the answer layer carrying a brand's own words back to the customer when those words exist and are precise. That is the fix, in one example, and the starting point of an AI visibility audit. The engines are not closed to a brand. They reward the brand that publishes the answer first, and most simply have not.
Across the 93-brand Google layer and the five-brand answer-engine deep dive, three risks stood out. The AI answer can work against a brand, with wrong facts, old events, and competitor steers. It disagrees with Google about the same brand on the same day. And the domains wearing a brand's name cannot be judged from the outside, only shortlisted for the brand to settle.
None of that shows on a rank tracker, and none of it in a single chatbot read once. It shows only when Google and the AI answers are read together, brand by brand, in the market and language the customer actually uses.
We measure and manage how a brand appears across search and AI, in industries where reputation is contested. For one brand we map every market and every engine, Google and AI side by side, every claim checked against the live record, and every look-alike domain resolved to a confirm-or-disown list. We will walk you through a real one before you commit to anything.
Request your Brand Answer Map or email hello@whitelobby.comYes. In WhiteLobby's captures the engines misstated brands' live review scores, resurfaced years-old regulatory events, and recommended named competitors as alternatives.
Not always. We saw them describe the same brand in opposite terms on the same day, for example a national police fraud-warning ranking on Google while an AI engine called the brand reliable.
Not consistently. Across five brands, two of three engines overstated live Trustpilot scores by up to a point and a half. In these captures ChatGPT did not overstate any score.
WhiteLobby classified 380 name-bearing domains. Most are brand-owned or unrelated namesakes. After hand-reviewing, roughly a dozen present as the brand on infrastructure it does not appear to run, a shortlist for the brand to confirm, not a copycat count.