A woman in Scottsdale opens her phone at 9pm with a cracked molar and no regular dentist. She doesn't scroll a map. She doesn't open ten blue links. She asks ChatGPT who she should see, reads the two names it gives her, and books the first one that has Saturday hours. That entire decision took ninety seconds, and your practice was either in it or it wasn't. There was no second page to lose on. There was a list of two, and you made it or you didn't.
This is already how a growing share of local decisions get made. And the uncomfortable part for anyone who spent a decade building a reputation the old way is that the assistants making these calls no longer agree with each other about who deserves the recommendation.
The platforms are pulling apart
For a while, AI visibility behaved like one thing. If ChatGPT knew you, Gemini probably knew you, and Perplexity likely did too. You could think about "being visible to AI" as a single status you either had or didn't.
That's over. Through 2026 the major assistants have diverged hard in how they decide who to name. Perplexity has leaned into structured, source-cited local answers that look more and more like a ranked shortlist. Gemini has drawn tighter to the signals living inside Google's own ecosystem. ChatGPT has sharpened its geographic precision, so "near me" means something more exact than it did a year ago. Claude has grown noticeably more conservative, preferring to cite established third-party directories over a business's own website.
The practical consequence lands on you directly. A family attorney in Tampa can be the confident pick on one assistant and completely absent on another, for the identical query, on the same afternoon. Nothing about the firm changed between the two answers. The platforms simply weigh their evidence differently now, and a firm that clears one platform's bar can sit just under another's.
Why an assistant trusts one name and not another
Here's the mechanism worth actually understanding, because it explains almost everything above.
An AI model does not take your word for who you are. It has no way to know whether "top-rated cosmetic dentist in Scottsdale" is true or just something you wrote about yourself. So it does what a careful person does when they can't verify a claim directly: it looks for agreement. It checks whether the same facts about you — your name, your specialty, your location, your standing — show up the same way in more than one independent place. When a model finds that agreement, its confidence climbs, and confidence is what lets it say your name out loud as a recommendation. When it finds contradiction instead — one source calls you a general dentist, another an orthodontist, a third has an old address, a fourth a different phone number — the model does the safe thing. It hedges. And a hedging model reaches for someone else it feels surer about.
That's the quiet engine behind the divergence. Each platform draws its "independent places" from a slightly different pool and sets its own threshold for how much agreement it needs before it commits. Claude's growing preference for established directories is exactly this instinct turned up: it would rather cite a source it treats as vetted than a self-published page it can't corroborate. Same underlying logic as every other assistant. Different tolerance for how much confirmation it demands before it'll put a name in front of a real person about to spend real money.
What "established directory" actually means to a model
Take Martindale-Hubbell, the attorney directory that's been rating lawyers since the 1800s. To a person, it might read as one more listing site. To a language model, it's something more useful: a source with a long institutional history, a peer-review rating tradition, and structured, consistent records that have been reproduced and referenced across the legal web for generations. When a model encounters a firm's details inside Martindale-Hubbell, it isn't just reading another page. It's finding a version of the firm's identity that has the texture of something vetted by an outside authority rather than asserted by the firm itself.
That's why a presence there can carry disproportionate weight in whether an assistant feels safe naming a firm. Not because the directory is magic, and not because filling in a profile is a trick. It's that the model is looking for corroboration from sources it reads as independent and stable, and a century-old directory reads exactly that way. The same principle explains why a doctor's consistent, matching presence across the medical directories carries similar weight, and why a scattered, contradictory presence quietly undermines it. Healthgrades exists in that same category of source a model tends to lean on. The lesson isn't "go fill out these profiles." It's understanding what the model is actually hungry for — confirmation it can trust — and recognizing that where you show up, and whether you show up consistently, is the raw material it's feeding on.
The trap of optimizing for the platform you happen to check
Most professionals who think about this at all check one assistant. Usually the one they personally use. A real estate agent in Denver asks ChatGPT "best listing agent in Wash Park," sees her name, and relaxes.
She shouldn't. Her next seller might ask Perplexity, which is building its answers from directory data she's neglected. Or Gemini, which is reading signals tied to her Google presence she hasn't touched in a year. Or Copilot inside a work laptop. Or Grok. Each of those assistants is running its own evidence check against its own pool of sources, and being the confident answer on one tells you almost nothing about the other six. Single-platform confidence is the most dangerous kind, because it feels like safety while leaving most of the doors unwatched.
The divergence is precisely what makes this expensive. When the platforms moved together, checking one was a decent proxy for all. Now the correlation has broken, and the only honest picture is the cross-platform one. You can read more about how each assistant assembles its recommendations in our breakdown of how ChatGPT, Claude, Gemini, and Perplexity recommend businesses, and about why the early movers are building a real competitive moat while most of their competitors haven't noticed the game changed.
The shift you'll feel before you read about it
The hardest part of a platform reweighting is that it doesn't announce itself. No email arrives telling a periodontist in Charlotte that Gemini has adjusted how heavily it leans on a certain class of signal. He just starts appearing less often. If he isn't measuring, the first thing he notices is a slower schedule three months later, and by then he's guessing at causes.
This is why steady measurement matters more than any single optimization. When you track your citation rate across all seven assistants on a fixed cadence, a platform shift shows up as a number moving in your own data — a rate that drops or climbs without any change on your end. That unexplained movement is the early warning. Our guide to AI visibility KPIs and benchmarks walks through what those measurements look like in practice. The businesses that stay recommendable through all this churn aren't the ones who guessed which platform would win. They're the ones who saw the divergence in their own numbers early enough to move.
Where this actually stands
The RankCommander AI Visibility Index tracks this across thousands of real assistant answers in local professional verticals, and the pattern it keeps surfacing is the same one this whole piece circles: the gap between businesses on the same block is widening, and it's widening fastest on the platforms owners check least. What separates the professionals who stay recommendable from the ones quietly falling out of the answers isn't a longer to-do list. It's whether they can see their own position clearly across every platform at once, before the divergence turns into a decision made without them.
You've spent years earning the reputation that should make you the obvious answer. Right now, in your city, an assistant is deciding whether to say your name or a competitor's to someone who is ready to book — and if you've never checked, you genuinely don't know which way it's going. That's not a slow erosion you can address next quarter. It's a patient walking into someone else's office tonight. Run your free scan and see, across all seven assistants, exactly where you stand and who's showing up in your place. Find out now, while there's still room to change the answer.