Ask an AI engine about your company by name and it will probably describe you well. Ask it the question your best prospect would actually type, and you may not appear at all.
Those are two different tests. Most companies have only ever taken the easy one.
The five minutes that ruin a good afternoon
Here is how marketing leaders usually discover this.
Someone types their company name into ChatGPT and asks what the company does. The answer comes back accurate, fluent, and flattering. Reassuring.
Then, out of curiosity, they ask a different question. The one a real buyer would ask, with no company name in it. Something like: “Who are the best commercial insurance brokers in central Arkansas?” or “What should I look for in a cardiology practice near me?”
Three companies come back. None of them are theirs.
How big is the gap?
Bigger than almost anyone expects.
In a study of 112 companies, researcher Amit Prakash Sharma ran 2,240 separate questions through ChatGPT and Perplexity. The results split cleanly along the line described above.
Asked about by name, ChatGPT recognized the companies 99.4% of the time. Nearly perfect.
Asked a discovery question, the kind with no brand name in it, the same companies appeared 3.32% of the time.
Perplexity showed the same pattern: 94.3% recognition by name, 8.29% in discovery.
Worth noting how those figures were produced, because it matters. They come from thousands of queries, not from someone checking once and drawing a conclusion. AI answers vary a great deal from one run to the next, which is a subject worth its own article, and we wrote one.
Why does the gap exist?
Because the two questions test completely different things.
Consider the difference between asking an acquaintance “have you heard of Smith and Company?” and asking that same person “who should I hire?”
The first question tests whether they know you exist. Almost everyone you have ever met can pass it.
The second asks them to put their own credibility on the line by naming a short list. Very few people pass that one, and the ones who do got there because other people they trust have spoken well of you.
AI engines work the same way. A question with your name in it is a lookup. A question without your name in it is a judgment call: out of everyone who could possibly be named, who belongs on a list of three or four?
That judgment does not come from your website. Your website says you are excellent. So does every competitor’s. The engine cannot break that tie using self-description, so it leans on what independent sources appear to agree on.
Which means the gap between recognition and recommendation is not a technical gap. It is a reputation gap, and it appears at the exact moment someone is deciding.
Why this got expensive recently
For years, being left out of an answer was survivable, because the answer was a list of links and people clicked several of them.
That is changing, and there is now evidence for how much.
Mehrzad Khosravi and Hema Yoganarasimhan, a professor of marketing at the University of Washington, compared traffic to English Wikipedia articles appearing beneath Google’s AI-generated summaries against the same articles in other languages, where those summaries had not rolled out. They examined more than 161,000 matched pairs.
The English pages lost roughly 15% of their daily traffic, with the steepest declines on pages an AI could summarize completely.
The lesson is not about Wikipedia. It is that the answer is becoming the destination. When someone gets a satisfying response without clicking, the short list inside that response is the entire competitive set. There is no page two to fight your way onto.
Being absent from an answer used to cost you a click. Now it can cost you the consideration.
This is not a website problem
Most companies will read this and route it to whoever handles SEO. That is a mistake worth being specific about.
If an AI engine leaves you out of a short list of recommended providers, you have a share of voice problem happening at the moment of decision. That is brand awareness, measured against competitors rather than against a checklist.
If the engine includes you but describes you vaguely, or in a way that makes you sound interchangeable with two other firms, you have a positioning problem.
Neither gets solved by a developer ticket. Both are what a marketing team is for.
Where this hits hardest
The pattern is clear: the more someone needs help deciding, the more likely they are to ask an AI engine to help them decide. The exposed categories are the significant, infrequent, heavily researched purchases.
Higher education, where prospective students ask which programs fit their schedule and goals. Banking, where small business owners ask which institution suits a business like theirs. Healthcare, where patients ask about service lines and local options. Legal, where people ask what to consider before hiring and who handles a particular kind of matter nearby.
These are the decisions where a recommendation carries the most weight, because the cost of choosing wrong is high.
The two-question test
You can find out where you stand in about five minutes, with no tool or subscription.
Ask any AI engine about your company by name, and note whether the description is one you would have approved. Then ask the question your best prospect would ask, with no company name, using the words a normal person would use rather than the words in your marketing materials. Then ask that same question two or three more ways.
Look at four things. Do you appear at all? How are you described? Which competitors show up? And which websites is the engine citing as sources?
That last one is the most useful and the one most people skip. The cited sources are a map of where your reputation is actually being built, and it is rarely where you assumed.
One honest caveat. A handful of questions will tell you whether you have a problem. It will not tell you how big it is, because AI answers are inconsistent enough that a small sample can mislead you badly in either direction.
The question has changed
For twenty years the question was where do we rank.
The question now is whether we are in the answer, and whether we like how we are described.
Most companies have never checked. The ones that have are usually surprised, and it is almost always the same surprise: the engine knew exactly who they were and still did not think to mention them.
Knowing you exist is not the same as recommending you. It never was.
Want to know how AI engines describe your brand and your competitors?
In a Brand Awareness Consultation for AI Engines, we run your category’s real buying questions across ChatGPT, Gemini, Perplexity, and Google’s AI search, repeatedly and over time, then show you where you appear, how you are described, which competitors surface, and which sources are shaping the answer.
You get a clear read on the gap and the first improvements worth making.
Sources
Sharma, Amit Prakash. “The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries.” arXiv. Based on an M.Tech thesis at the Indian Institute of Technology Patna. https://arxiv.org/abs/2601.00912
Khosravi, Mehrzad, and Hema Yoganarasimhan. “Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia.” arXiv. Yoganarasimhan is Professor of Marketing and Michael G. Foster Faculty Fellow at the University of Washington Foster School of Business. https://arxiv.org/abs/2602.18455