AI chatbots now shape purchase decisions before your brand tracker ever gets a chance to ask a question. Most research programs have no idea what ChatGPT, Gemini, or Perplexity are telling prospective customers about them. Here's what it takes to bring AI perception into a real research program.
Most brand trackers still only survey humans. That's worth sitting with, because a growing share of the opinions forming about your brand right now aren't coming from a person who saw your ad or heard about you from a friend. They're coming from an AI model, answering a question your customer typed into ChatGPT, Gemini, or Perplexity before your sales team ever heard from them. Your tracker has no idea what that model said. It never asked.
The Purchase Journey Now Runs Through an AI Interlocutor
The numbers moved fast on this one. Consumers are using AI chatbots at every identifiable stage of a purchase decision: 51% during early discovery, 57% to narrow down a shortlist, 53% to compare options they're already considering. About 60% say chatbots frequently influence what they end up buying, and 77% of people who use AI in the process say it helps them decide faster. Not a niche behavior. It's becoming the default first stop for category research that used to start with a search engine or a sales rep.
That shift matters for brand health research specifically, not just marketing. If a real share of your prospective customers are forming an impression of your brand from a model's answer before they ever fill out a survey, that answer is now part of the attitude formation your tracker exists to measure. Ignoring it doesn't make it stop happening. It just means your tracker is measuring what people believe after a step it never observed.
Brand Trackers Weren't Built to See This
A standard brand tracker asks a sample of humans a consistent set of questions on a regular cadence: awareness, consideration, favorability, NPS. It's a well-understood instrument, built for a world where the inputs shaping those answers were ads, word of mouth, press, and direct experience. None of it was designed to catch what happens when a customer asks a model to compare you against a competitor and gets a confident, specific answer back.
The blind spot is bigger than most research teams assume. A 2025 Gartner study found that 63% of B2B brands had zero presence in AI search engine responses for their primary keyword category. Zero presence doesn't mean neutral. It means the model answered the question anyway, using whatever competitor or category information it did have, and the absent brand simply wasn't part of the conversation. A tracker measuring unaided awareness among humans has nothing to say about that kind of absence. It wasn't built to catch it.
That's not a small gap. Brands missing from AI answers may be missing the phase where a purchase already got narrowed down without them.
This Is a Research Question, Not Just an SEO One
The response so far has come almost entirely from marketing and SEO teams building "AI visibility" tools that track how often a brand gets mentioned inside model responses. Fine work. But it's a different question from the one brand health research answers. Counting mentions tells you if you showed up. It doesn't tell you what the model actually said about you, how that framing compares to what real respondents believe, or whether the AI's characterization is quietly drifting from your intended positioning. Those are perceptual questions. They belong in a research program, not a citation dashboard.
Treating this as marketing's problem alone misses something research teams already know how to do: compare a synthetic "respondent," the model, against a human sample on the same questions, with the same rigor you'd apply to any other perception gap. Where does the model's read on your brand line up with what real customers say? Where does it diverge, and does that gap run heavier in categories where AI-assisted discovery is already normal?
What Adding This to a Research Program Actually Looks Like
The practical version doesn't require infrastructure most teams lack. Treat a handful of frontier models as an additional, low-cost respondent group inside the existing tracker cadence. Run the same core brand questions, category consideration sets, and competitive comparisons through ChatGPT, Gemini, and Perplexity on the same schedule as the human wave. Log the responses. Code them the way you'd code any qualitative input: sentiment, category framing, specific claims made. Then set that next to what human respondents say in the same period.
The value isn't in one snapshot. It's in the drift. Models get retrained, get connected to fresher web data, and their read on a category can shift without warning. A research program checking quarterly catches a positioning problem while it's still small. One that never checks hears about it from sales, several deals too late.
Brand research has always tried to answer one question honestly: what do people actually believe about us, and why? An AI model forming that belief on a customer's behalf, before a human researcher gets a chance to ask anything, is a strange thing to leave outside the tracker. The teams measuring it now aren't chasing a trend. They're just refusing to let the newest respondent in the room go unasked. See how Quali-Fi approaches integrated brand health research across human and AI-driven signals ->