AI Is Forming a View of Your Company. Most People Can't Tell How.

Our research tracks a wide range of public perceptions of AI. Three questions from this year's waves focus on AI and companies: how often people use it to learn about companies, what they want to find, and whether they know how the answer gets made. The answers fit together into one picture. People already turn to AI for answers about companies, mostly to answer commercial questions, and most can't say how those answers are built. In banking, for example, AI rated every bank lower than the public did and drew mostly on sources the banks didn't write.

That makes AI a stakeholder in its own right, and one worth measuring on the same terms as people. The question for a communications team is where AI's view of the company departs from the public's, and on which drivers.

Each finding says less on its own. If few people used AI to research companies, the fact that most can't describe how it works would be a curiosity. Our earlier waves show many people do use it, and for which questions. Our banking study, as one example, shows what those answers can contain. Together they turn a finding about understanding into a measure of exposure.

Only 28% can describe how an AI answer gets made

Our Q3 2026 Stakeholder Intelligence research, fielded between June and August 2026, asked the Informed General Public (IGP) how an AI assistant answers a question about a company. Only 28% chose the technically accurate description: the assistant creates a new answer by predicting what fits the question, based on patterns in information it has learned.

The rest hold a different picture, or none:

31% think it searches the internet and summarizes what it finds

10% think it's whatever the AI company or advertisers want people to see

9% think it's picked from answers written in advance

3% think a person writes it before the user sees it

19% aren't sure

That suggests most people reading an AI answer have no clear way to judge where it came from or how far to trust it.

People already rely on AI, mostly for commercial questions

The Q3 result matters because of what the two earlier waves found.

ChatGPT and Gemini rated seven banks on our standard drivers and factors. Company-owned material made up about 9% of the sources each platform cited. Every bank scored lower with AI than with the public, by about 19 to more than 51 points.

The gaps were widest on Conduct and on Products & Services, the drivers the public weighted most in that study. On Conduct, AI scored the banks 26.1 against the public's 68.0. On Products & Services, it scored them 42.0 against 71.0.

Read against the Q3 wave, the source mix stands out. The 31% who picture AI searching and summarizing likely assume an answer reflects a fair read of what's published. In banking, about nine in ten of the sources behind that read came from someone other than the bank. Banking draws heavy regulatory and complaint coverage, which is a likely contributor on Conduct.

Read against the Q2 wave, the gaps fall in a telling order. Products & Services, where the most-asked questions sit, showed a 29-point gap. Conduct showed 42. The topics people ask about most and the driver where AI departs most from the public can be two different lists.

These results come from US banking. We'd expect a similar source mix elsewhere, since AI builds its view from what's published. Whether a similar score gap follows has to be measured company by company.

Work on the widest gap first, on the drivers that matter most

Measure where AI's view of your company departs from the public's, then put effort into the widest gap on the drivers your stakeholders weight most. Start there, ahead of the questions people ask most often.

If your sector looks like banking, that gap is likely to sit on Conduct, and updated product pages won't reach it. A practical first pass:

Ask ChatGPT and Gemini the questions a skeptical customer would ask about how you handle complaints and customer data.

Where an answer cites sources, log which third-party sources dominate.

Give that list to the team that owns your public account of those topics, so it can see where your own account is missing.

Update prices and product details in parallel. They're what people ask about most, and they sit on pages you control.

A quarter later, ask again and check whether the cited sources have changed. Answers can lag behind your changes, since assistants also draw on what they learned in training, so a shift in sources is the earlier signal.

How RepTrak helps

A first pass like this shows what AI says about you. It leaves three questions open: how AI's view compares with your stakeholders', which of the gaps matter most, and whether they're closing.

Our AI as a Stakeholder research answers all three. It rates your company through AI platforms on the same seven drivers and 23 factors we use with people, so AI's view sits beside the public's driver by driver, as it did for the seven banks in our American Banker study. It shows which sources sit behind each rating, which tells you whether a gap traces to your own pages or to coverage someone else wrote. Importance weights from your human stakeholders then tell you which gaps to work on first. Measured again in the next wave, it shows whether the sources and the scores have moved.

Our advisory team turns each wave into insights on what changed and why, recommendations on what to do about it, and specific actions naming the channel, the audience, and the message. That puts the source list in front of the team that can act on it, with a view on what to publish and what to correct.

See where AI's view of your company departs from the public's, and which sources it's drawing on.