Nobody Has Measured Which Gyms AI Names — the Vendor Selling the Fix Says So Itself. What AI Filters For Is Whether You Can Get There.

No published study measures which gyms AI names, and the vendor selling the fix is the one that admits it

Start with the sentence every other page in this category skips. There is no published primary study of AI recommendations at the per-fitness-studio scale. That is not our reading of the literature — it is OpenLens's, and OpenLens is an AI-visibility vendor whose product exists to solve the problem this absence would otherwise prove. A vendor conceding that the evidence for its own category was never gathered is the most credible sentence in this market. The entire evidence base, item by item:

• No per-domain citation-share study exists for fitness. OpenLens says qualitatively that ClassPass and Mindbody dominate for class-based fitness, and Yelp and Google Business Profile for general gyms — with no shares behind either claim. OpenLens sells AI visibility: that is a vendor's characterization of a market it sells into, not a measurement of it.
• The number circulating everywhere here — "26% of relocating consumers now use ChatGPT, Google AI Overviews, Perplexity, and DeepSeek when picking a gym, studio, or personal trainer" — comes from OpenLens (May 4 2026) and discloses no sample: who was asked, how many, and where are all absent. It is a vendor's figure with no published sample, which is the only way it can honestly be quoted, including here.
• Keepme, a fitness AEO vendor, says its research covered 901 fitness operators across 27 countries — the largest fitness-specific sample anyone claims, and not US-only, which matters if you run one box in Ohio. Scale is not applicability.
• No US study of chain-versus-independent share in AI answers exists for fitness. None. A vendor quoting you the percentage of AI's gym recommendations that goes to the chains is quoting a number nobody has measured.

AI is not choosing the best gym. It is filtering for the gym a person can actually get to three times a week

Look at what people type. "Gym open at 5am before work." "Gym with childcare near [area]." "How much is a gym membership in [city], no contract?" Not one is a question about quality. Every one asks whether this person can realistically show up — on the days they promised themselves they would, for long enough that the membership survives February.

So the assistant is not ranking gyms. It is filtering them, on attendance — and a filter needs a value to filter on. Opening at 5am is a value. Thirty days' notice to cancel is a value. "A premier fitness experience in the heart of the city" is not a value; it is an adjective, and an adjective is not an input. The peer-reviewed GEO paper (Princeton, Georgia Tech and IIT Delhi, ACM SIGKDD 2024, 10,000 queries) found statistics lifted a page's visibility in generated answers by 41% and quotations by 28%, while keyword stuffing produced no gain (arXiv). Adjectives are keyword stuffing with better manners. Below is each attendance filter and the sentence that survives it.

• "Open 5:00am to 10:00pm Monday to Friday, 7:00am to 6:00pm weekends. Staffed from 6:00am." The 5am lifter and the 9pm shift worker are two different filters; "open early" passes neither. Pair it with where you are relative to the office and the house — those are two different products, so say which you are.
• "Childcare runs 8:30am to 12:00pm and 4:00pm to 7:00pm, ages six months to ten years." A parent's entire decision can turn on that Tuesday-morning window.
• "Month-to-month. No joining fee. Thirty days' written notice to cancel." Then the price, in text — a range if you will not publish a figure. "No contract" is among the most-typed phrases in this category and the least-published.
• "Your first session is a free 45-minute intro with a coach. Classes cap at twelve." "Not intimidating" is what the member types; that sentence is what makes it true.
• "Eight squat racks, four platforms, two turf lanes." It answers the 6pm question a photo gallery cannot.
• "Every session is scaled. Coming back from a knee injury? Tell the coach at check-in and they will substitute the movement." The most common qualifying question in strength and CrossFit, and almost nobody answers it in text.
• All of it in crawlable HTML — not in a booking widget, a PDF timetable, or only the app. A schedule living inside an embedded Mindbody iframe is one the assistant answering at 11pm cannot read, and cannot name you from.

The class schedule changed on Monday. The answer didn't.

Fitness has a problem no other local vertical has at this frequency: almost everything a prospective member wants to know is something the gym changed recently. Schedules move every season, intro offers run six weeks, childcare hours shift when a staffer leaves, prices go up in January. Each is a fact an assistant may have learned three months ago and will state, fluently, as current.

SOCi's 2026 Local Visibility Index, built on roughly 350,000 locations by a company selling local-marketing software — so a finding that AI gets your data wrong sells local-marketing software — measured business-data accuracy in AI answers at about 68% for ChatGPT and Perplexity, against 100% for Gemini, which is grounded in Google Maps. There is no fitness cut of it, but a gym carries more perishable facts than almost any location in that sample.

The prospect finds out at your front desk. Someone arriving with a joining fee that was waived until March, or turning up for a 6:30 class that moved to 7:00, has been misled — by the assistant, but in your name, in your lobby, in the first ninety seconds of knowing you. And this audience has no trust to spare: in a Yelp and Ipsos survey of 2,202 US adults in late February 2026, 65% had used AI search in the previous six months but only 15% trusted it strongly.

The fix is dull and structural: one canonical schedule page, one canonical pricing page, both in text, both updated the day the thing changes, and no contradicting copy left alive on a landing page from last spring. Delete the expired offer rather than unlinking it. Freshness looks like a live signal — Omnibound, a GEO vendor reporting its own data (a finding that fresh content wins sells content refreshes), reports 76.4% of ChatGPT's most-cited pages were updated within 30 days — though whether that holds for a studio in your suburb is unmeasured. Updating the price page costs twenty minutes either way.

The sources AI reaches for when someone asks for a gym, ranked — and every rank below is an untested hypothesis

The caveat first: nobody has counted how often an AI answer about a gym cites Google, ClassPass, Instagram or Reddit, because no per-domain citation-share study exists for fitness. The ranking below is assembled from local-business-wide measurements with no fitness cut — a hypothesis to test on your own market, not a source list to buy. One line is not a hypothesis. The engine reads text, not your booking system: your childcare window and your cancellation terms are facts to a model only because some page states them, and the page that can state them first is yours. Platforms corroborate that statement or, once your schedule moves, contradict it.

The nearest primary evidence is ours, and it is from other verticals. In the three markets we have measured — dental, med spa and plastic surgery in Dallas, July 2026, across ChatGPT, Gemini and Perplexity — 73% to 79% of every source cited behind those recommendations was a business's own website rather than a directory or platform (the Dallas benchmark). Gyms and studios were not in that sample, so treat it as direction, not as your number. Dallas, July 2026 — one market, one day, and a primary measurement: the raw answers are stored, the method is published, and the run repeats against any city, including yours.

• 1. Your own schedule and pricing pages — the only source here where "5:00am", "thirty days' written notice" and "ages six months to ten years" can originate rather than be echoed. Yext's October 2025 analysis of 6.8 million citations across 1.6 million AI answers put Gemini at 52.15% own-site citations and ChatGPT at 48.73% third-party listings; Yext sells listings management, so a finding that listings decide everything sells listings management — and its own numbers still land more than half of Gemini's citations on brands' own sites. Our Dallas run points the same way in adjacent verticals: 73% to 79% of cited sources were businesses' own sites across dental, med spa and plastic surgery (the Dallas benchmark; our primary measurement — one market, one day, no fitness cut). Do this: get the timetable and the price into HTML. Every source below repeats these pages or contradicts them.
• 2. Google Business Profile and Google Maps. SOCi (about 350,000 locations, 2026; sells local-marketing software) measured Gemini's business-data accuracy at 100%, because it is grounded in Google Maps, against about 68% for ChatGPT and Perplexity; BrightLocal (20 queries, 10 verticals, July 2025; sells local SEO tools) found Google Business Profile strongly preferred by Google's own models. Do this: pick the exact category — "gym", "pilates studio", "boxing gym" and "personal trainer" are four different answers — and get the hours right.
• 3. ClassPass and Mindbody. OpenLens says these dominate AI answers for class-based fitness, publishes no shares behind it, and states in the same breath that no per-domain study exists. An AI-visibility vendor's unquantified characterization — and the most-repeated line in this market. Do this: make the listings carry the same schedule, prices and class names your site does, then measure whether the assistant names your studio or the platform.
• 4. Instagram and social profiles. Semrush, which sells AI-visibility tracking, found Google AI Mode cites 11.4 sources per answer and reaches for Facebook, Instagram and Yelp (126 million prompts, January to April 2026). Fitness lives on Instagram, and the engine reading it finds a bio with no address, no hours and a Linktree. Do this: address, hours and timetable link in the bio, in text.
• 5. Reddit, Yelp and "best gyms in [city]" roundups. Semrush, which sells AI-visibility tracking, found Wikipedia, YouTube, Google properties, Reddit and Amazon account for 38% of everything AI cites (46 million citations, March to August 2025); Yelp took 512,680 citations across 28 million AI answers in Q4 2025 (Foundation Inc. with AirOps, a content agency and an AI-search vendor; no fitness cut). Do this: you cannot edit your city's subreddit, and it is where "which box is good with beginners" gets answered. Read it; do not astroturf it.
• 6. Chain and category brand sites. Unmeasured — no US study of chain-versus-independent share in AI answers exists for fitness. Default brand recognition is a plausible mechanism and an unproven one. Do this: stop competing on "best gym in [city]" and own the question the chain page does not answer.

Does AI just name the big chains? Nobody in the US has measured it, and we are not going to guess

Every independent studio assumes the assistant reaches for the recognizable name whenever a query is generic. It is a plausible mechanism: a model that has read a national brand's name ten thousand times has more to lean on than one that has read yours in eleven Google reviews. But plausible is not measured. If someone quotes you a percentage, ask for the sample.

What is measured, across local business generally: getting named at all is rare for everyone. SOCi's 2026 index of roughly 350,000 locations found ChatGPT recommends a given location in 1.2% of answers and Gemini in 11%, against 35.9% for Google's local three-pack. SOCi sells local-marketing software, so the gap it reports is a gap it profits from reporting. Carry the shape, not the number: the AI answer is a far narrower door than the map pack.

And there is a prior question — whether the engine speaks at all. Whitespark ran 540 queries across 3 cities and 6 industries and found AI Overviews on 68% of local business queries, but on only 15% of purely local "near me" searches, against 97% of hybrids. Fitness was not in that sample, so carry it as a hypothesis. The implication is sharp: "gym near me" is where AI is least likely to say anything, and "how much is a gym membership in [city] with no contract" is the hybrid where it says the most. Which is where the specific question becomes an independent gym's real advantage, and possibly its only one — nobody asks a chain-shaped question about a knee. That is not a strategy anyone can sell you; it is a page you write.

Run these seven prompts a prospective member actually types, then count how many name your gym

Fifteen minutes, no tool, no vendor, no login — and it is the only number in this article that will be about your gym. Open ChatGPT, Gemini and Perplexity, use a fresh chat per prompt, paste the seven below with your city in the brackets, and write down every gym named and every platform named. That is 21 answers. What matters is not only how often your name appears, but how often the answer names a gym you know is a worse fit for that question than you are.

• "Beginner-friendly gym near me in [city] that isn't intimidating."
• "Gym in [city] open at 5am before work."
• "Best CrossFit box in [city] for someone coming back from a knee injury."
• "Yoga studio in [city] for a complete beginner."
• "Gym with childcare near [neighborhood]."
• "How much is a gym membership in [city], no contract?"
• "Women-only gym near me in [city]."
• "Is [your gym name] in [city] any good, and what do they offer?" — the eighth one, which shows what the engines already believe.

What gym owners say when we show them this, including the objections that cut against us

Sort the misses by filter, not by mood: a gym named on the 5am prompt and absent on childcare has a page to write, while "we did badly" has nothing to act on. Read the eighth answer for facts that are simply wrong — the class that no longer runs, the price from two years ago, the promo that ended. And know the limit: one pass is one Tuesday afternoon, while the moments deciding a gym's year are the first week of January and the Sunday night in March. The objections below come up with almost every owner, and more than one is right.

• "ClassPass already gives me trial volume." Often true, and a different question. ClassPass sends you people who already opened ClassPass; the prompts above get answered before anyone opens anything. The parent typing about childcare hours at 9pm has not chosen a platform, they have chosen an assistant. And the honest limit: OpenLens, an AI-visibility vendor, claims ClassPass and Mindbody dominate class-based fitness answers but publishes no shares and says no per-domain study exists. Nobody — them, us, or the agency emailing you — can quantify how much of that answer ClassPass holds. What is not in doubt is the direction of supply: a platform can only recite a schedule, a price and a childcare window that somebody published. If the only readable copy lives inside the platform, the platform is the only name the answer can safely give.
• "Every number in this space comes from a company selling AI visibility." Almost all of it, and it pays to know who profits from which finding: OpenLens sells AI visibility; Keepme sells fitness AEO; SOCi, Yext, Semrush, BrightLocal, Chatmeter and Omnibound sell software to the businesses they measured. RecoSignal measures too — and stores the raw engine answers and publishes the method, so its numbers can be checked instead of believed. Which is also why the fifteen-minute check above costs nothing: it is the one number in this market nobody is selling you.
• "Our members come from word of mouth and people walking past." Still largely true, and the assistant has entered the referral rather than replaced it. BrightLocal's 2026 survey of 1,002 US consumers (BrightLocal sells local SEO and reputation tools) found 45% now ask AI to recommend a local business, up from 6% in 2025; GatherUp, which sells review software, found 57% of more than 1,000 US consumers had asked ChatGPT, Claude or Gemini to find a local business. The friend still recommends; then the person checks with an assistant.
• "The tour and the free week are what actually close people." Right, and they only close people who walked in. In a January 2025 survey of 1,059 US consumers, Chatmeter — which sells reputation and location software — found 64% want a free trial or day pass, 44% take a tour and 47% read reviews when choosing a gym. Three conversion tools, none of which runs on someone the filter removed three steps earlier.
• "Isn't this just SEO with a new name?" No, and the overlap is shrinking: Ahrefs found citations to the organic top 10 fell from 76% to 38%, so a first-page Google position no longer buys a seat in the answer (AI visibility versus local SEO).

The only number that means anything for your gym is the one you generate yourself

The two figures a gym owner would most want to buy — how often an assistant names an independent studio, and what share of fitness answers goes to the chains and the platforms — have never been measured by anyone in the United States. We will not paper over that with a borrowed statistic, because the statistic being borrowed across this niche discloses no sample. The fitness AI-visibility industry is selling a service against an evidence base that does not exist yet. That is the state of the field, and you should hear it before you sign anything.

What survives is small, dull and yours: the hours in text, the price in text, the contract terms, the childcare window, the class cap, the one sentence about the knee. Those are the values the filter reads. A gym that publishes them has entered the filter it is currently excluded from; a gym that publishes adjectives has handed the filter nothing, and keeps losing to gyms that are, on any honest reading, worse. See how to improve your AI visibility for the ordered playbook, and how to get recommended by ChatGPT for one engine's mechanics.

RecoSignal's free AI Visibility Snapshot puts member-style questions to ChatGPT, Gemini and Perplexity and reports which assistants named your gym, which gyms they named instead, and which sources each answer leaned on. It is a measurement, not a promise. Nobody can guarantee an assistant will name your gym, there is no #1 position inside ChatGPT to buy, and a vendor offering either is selling a certainty that does not exist.