AI visibility is whether AI assistants like ChatGPT, Gemini, and Perplexity mention your business, describe it accurately, and recommend it when a buyer asks who to hire. It moves independently of your search rankings and traffic, and neither can tell you where you stand. You can check it yourself, across three engines, in about ten minutes, for free.
Somewhere today, a buyer asked an AI assistant the question you built your business to answer: who should I call about this? The assistant replied with a handful of names. Either yours was one of them or it was not, and nothing in your analytics recorded which.
This is already normal behavior, not a preview of it. Adobe surveyed more than 5,000 US consumers in March 2025: 39 percent had used generative AI for online shopping, 55 percent of those users to run research and 47 percent to get product recommendations. Every one of those conversations named some businesses and passed over the rest.
At Probably Genius we measure exactly this: we ask five AI engines the questions real buyers ask and record who gets mentioned, how accurately each business is described, and who gets recommended. Then we do the work that changes the answer.
If you did everything right by the old rules, decent rankings, steady traffic, strong reviews, it is reasonable to assume you are covered. It is also, increasingly often, wrong. The answers your buyers now read are assembled by a different process, judged by different criteria, and delivered somewhere your analytics has never been.
AI Visibility, Defined (Without the Acronyms)
Strip away the industry noise and AI visibility is three questions, asked in order. When a buyer asks an AI assistant about your category, does your business get mentioned at all? When it comes up, does the machine describe it accurately: the real services, the real people, the real specialty? And when the buyer asks “who should I hire for this,” does the machine recommend you?
Mentioned, accurate, recommended. Each test is harder than the one before it, and each one can be passed or failed independently of your Google rankings.
You will meet this idea wearing a costume of acronyms: GEO, AEO, LLMO, AI SEO. Do not wait for the industry to settle on one. The acronyms describe kinds of work. AI visibility is the outcome the work is for, and it is the only term in the pile a business owner actually needs.
Here is the frame that lands best with owners who built their firms on referrals: AI has become a referral partner. It gets asked “who do you know who can handle this” every day, in conversations you never see, by people who trust its answer, and it refers entirely from the record it can find and verify. The question AI visibility asks is whether that partner has ever heard of you, and whether what it has heard is true.
A Different Scoreboard: Why Rankings and Traffic Don’t Transfer
Search rankings measure the position of a page in a list. AI visibility measures the presence of a business in an answer. A results page holds ten doors and lets your buyer choose one. An assistant reads behind the doors, decides who it believes, and speaks an answer with a handful of names in it, or none.
Our co-founder Jacquie Baker put the mechanical difference in one line, in an essay published back in December 2025: “AI doesn’t rank pages. It understands entities.” An entity, in plain terms, is the machine’s file on you: who this business is, who runs it, what it actually does, where it operates, and whether every place that file draws from agrees.
Before an engine decides whether to recommend you, it decides whether it is sure who you are. Pages are how the file gets assembled. The file is what gets recommended. That is why a beautifully ranked page with a scattered identity keeps losing to a plainer page with a coherent one, and why the foundation of this work is entity identity rather than another round of keywords.
The scoring differs too. Rankings still respond to link authority and on-page relevance. Citation behaves differently: when Semrush studied 11,882 prompts across ChatGPT Search, Google AI Mode, and Perplexity in January 2026, the traits most correlated with being cited were clear summarization (+32.83 percent) and strong E-E-A-T signals (+30.64 percent). Verifiable, quotable, consistently described. A different game, scored by a different referee.
And the stakes concentrate. Page one of Google holds ten spots plus ads, so twelve businesses share the buyer’s attention. An AI answer might hold two or three names, and for a growing share of buyers that answer ends the search. The scoreboard did not just change its rules. It got smaller.
Why Your Traffic Stats Can’t See It
Google Analytics measures events on your property: visits, pages, conversions. An AI conversation about your category happens on someone else’s property, start to finish. A buyer asks, the assistant answers, a shortlist forms, and your website was never touched, so your analytics files the entire event under nothing.
Owners find this the hardest part to accept, so it is worth saying flatly: the most consequential moments of your new-client acquisition can now happen in a room your instruments do not reach.
The direction of travel was visible before assistants went mainstream. The SparkToro and Datos clickstream study found that by 2024, 58.5 percent of US Google searches already ended without a click to the open web; for every 1,000 US searches, only 360 clicks reached a site Google does not own. Buyers were learning to take answers instead of visiting sources. Conversational AI completes the shift.
The platforms do not fill the gap for you, either: there is no console where ChatGPT reports how many times your firm was mentioned this month, no impression count for the conversations your buyers are having.
When AI does send a visitor to your site, the numbers are small but the visitor is serious. In Adobe’s analysis of more than 1 trillion visits to US retail sites, traffic arriving from generative AI sources showed 8 percent higher engagement and a 23 percent lower bounce rate than other sources, and it had grown 1,200 percent between July 2024 and February 2025. Those visitors researched first and arrived close to decided. They are the visible edge of a much larger invisible process.
Your traffic stats are not lying to you. They are answering a question that matters less every quarter, while the question that matters more goes unmeasured. Flat traffic and healthy rankings can coexist with total absence from the answer layer. For most professional firms we assess, they do.
The Ten-Minute Check: The Recommend-Test
The good news: AI visibility is directly observable, today, for free. We call the method the recommend-test. Write down five questions a real buyer in your category would ask, in full sentences, the way people actually talk to an assistant. Ask them across at least three engines. Log who was named, what was said about them, and whether you appeared at all.
One trap to avoid: “tell me about [your firm]” is the branded question, and nearly any engine will answer it politely. It only tests whether you are known. The buyer question, “who should I hire to handle X,” tests whether you are chosen, and a business can pass the first for years while failing the second every single day.
Then hold the result the way a careful professional holds any small sample: as a sample. The engines disagree with each other, and with themselves, week to week. One answer is an anecdote, and a scary anecdote is still an anecdote.
This is why we run the measurement as structured panels, the same buyer questions asked across five engines on a schedule, using our published multi-model methodology, and why we report results as rates, never as a fixed rank. Anyone who tells you exactly where you “rank in ChatGPT” is selling you a precision the technology does not offer.
One distinction will save you money when you start shopping this category: measured versus ready. Measured is whether the engines say your name today, and a growing pile of dashboards will happily chart that for a monthly fee. Ready is whether an engine that goes looking finds an identity it can verify, proof it can corroborate, and answers it can quote. A dashboard can tell you that you are absent. It cannot make you recommendable.
Readiness takes more than a probe to measure. Our 109-Point AI Visibility Diagnostic scores a business across eight zones, because “you didn’t get named” is a symptom with several possible causes. Scores run 0 to 109. Ninety and up means AI treats you as a primary source. Under 35 means you are effectively invisible. Most firms are not narrowly losing the answer slot. They have not yet built the record that would put them in contention for it.
What Moves AI Visibility
The engines walk every business up the same ladder: findable, readable, trusted, recommended. Most firms are findable and readable. They stall at trusted. Three kinds of work move you up.
Start with identity, because everything else compounds on top of it. The machine’s file on you has to be coherent: one name, one clear description of what you do, the same people and the same specialty confirmed everywhere the engines look, stated in structured data as well as prose. It is unglamorous work, and it is the difference between an engine confidently recommending a specific, verifiable firm and an engine declining to gamble on a maybe.
Then claim your expertise, on the record, in public. The pattern we see constantly: the experts hardest for a machine to find are frequently the best ones, because they spent twenty years doing the work while thinner operators spent those years publishing about it. The war stories, the judgment calls, the numbers all live in your head and your phone calls, which means to an AI engine they do not exist. We call that trap the Black Box Effect.
The way out is not becoming a marketer. You are the genius in this arrangement. The missing piece was never your expertise; it was the translation of that expertise into a record a machine can read, verify, and repeat. That work can be done in your voice, under your name.
Finally, publish answers only you could publish. The engines are drowning in generic content produced at industrial speed, and they are visibly learning to route around it in favor of sources with something original to say. Jacquie’s February 2026 essay put it bluntly: “The machines are filtering out the machines. AI cites humans.”
The Semrush numbers above are the same lesson in statistics. Your unreasonable client story, your actual pricing logic, your contrarian take that turned out right: that is what a machine quotes, because that is what it cannot get anywhere else.
None of this requires you to dance for an algorithm, and it certainly does not require TikTok. It requires your genius, documented, verified, and made legible to the systems your buyers now consult first. The people AI will never replace are the ones AI should be recommending. Getting the machines to see that about you is translation work, and translation is a solvable problem.
Want to Learn More?
Probably Genius was built by Jacquie Baker and Christopher Shaw, who spent more than 40 combined years and two agencies translating what makes an expert the best in the room for the system that decides who gets chosen. The audience changed from investors to AI. The skill did not.