How independent gyms get found by ChatGPT instead of paying ClassPass

A new mover to Tokyo opens ChatGPT. They type: 'boutique pilates near Shibuya station that takes drop-ins, under ¥4,000, with classes after 7pm.' Five years ago that question went to Google, then to ClassPass, then to a 30-minute scroll through Mindbody listings. Today it goes to one AI assistant, which returns three recommendations with hours, drop-in prices, and a one-line description of each studio's vibe. The user picks one and walks in.

If your studio is one of the three, you got a member who is yours — no commission, no platform fee, no member-data lock-in. If you're not, you've already lost that customer to whichever studio fed the AI assistant the right structured data. The irony is that the studios winning these queries today are not the ones with the biggest ClassPass spend or the slickest Instagram. They're the ones whose websites are readable to a machine.

This is the same shift we covered in the middleman tax is dying, specialized to fitness. The fitness category is uniquely exposed because the platforms — ClassPass, Mindbody, GymPass/Wellhub — extract some of the highest blended commission rates in any local-services vertical, and because boutique fitness customers are exactly the demographic most likely to use AI assistants for discovery.

How fitness discovery is actually shifting

The 2024-2026 window has produced a stack of agentic-commerce launches that, taken together, make AI-driven local discovery a real channel rather than a demo. McKinsey's October 2025 *Superagency in the workplace* report projects up to $1 trillion of US B2C retail revenue could be orchestrated through agentic commerce by 2030, with $3-5 trillion globally. Microsoft launched Copilot Shopping in April 2025 with OpenTable, Kayak, Klarna, Expedia, and Instacart as launch partners — and the local-services flow has expanded since.

Stripe's product lead Will Gaybrick framed the shift bluntly at the December 2025 Agentic Commerce Suite launch: "The browser is no longer the surface. The agent is." That sentence captures what's hard for fitness operators to absorb — the customer is not landing on your homepage and reading your About page. The agent is reading your homepage on the customer's behalf, and the customer is reading the agent's summary.

Alipay AI Pay clears more than 120 million weekly agent-mediated transactions in China today. OpenAI's Buy-in-ChatGPT shipped September 2025 with Etsy as the first integrated merchant. Google's Universal Commerce Protocol launched January 2026 with Walmart, Home Depot, Wayfair, and Urban Outfitters. None of those are fitness brands yet — but the rails being built are general-purpose, and local services are the next adjacency.

On the consumer side, the queries are getting more specific. The typical AI-assistant fitness query is far more constrained than its Google equivalent: a Google query says 'yoga Shibuya'; a ChatGPT query says 'beginner-friendly Vinyasa in Shibuya, weekday mornings before 9am, under ¥3,500, English-speaking instructor.' That precision is good news for studios with rich, machine-readable schedules — and very bad news for studios whose website has class times trapped inside a JavaScript widget the AI can't parse.

The fitness middleman tax, by the numbers

Fitness has not one middleman but three layered ones, each taking a cut of a different demand stream. Independent operators routinely pay all three.

ClassPass. Founded 2013, acquired by Mindbody in October 2021. ClassPass does not publish a public rate card; payouts are negotiated per-studio. The most cited industry estimate, from a 2019 *New York Times* feature and corroborated in Athletic Business coverage, is that studios receive roughly 40-50% of their normal drop-in rate for a ClassPass-booked class — meaning the platform's discounting plus commission captures 50-60% of the price gap. Studios also surrender ownership of the customer relationship.

Mindbody. The dominant studio software platform, used by an estimated 75,000+ fitness, beauty, and wellness businesses globally per Mindbody's own published data. Mindbody's pricing starts at $99/month per location on the Starter tier and rises through Accelerate, Ultimate, and Ultimate Plus tiers — Ultimate Plus is quoted privately and runs into the high hundreds per location per month. On top of subscription, Mindbody charges payment-processing fees on every booking and add-on fees for SMS reminders, branded apps, and the ClassPass integration it now owns.

GymPass / Wellhub. The corporate-benefits middleman, valued at $2.4 billion in its 2023 funding round per TechCrunch. GymPass (rebranded to Wellhub in 2024) routes corporate-benefits members to studios at a fraction of the studio's drop-in rate, with the gap captured by the platform and its corporate clients. Wellhub does not publish studio payouts. Operator interviews place the per-visit payout in the ¥800-1,500 range in Japan for a class with a normal ¥3,500-4,500 drop-in price.

Stack the three. A boutique studio in Tokyo with a ¥3,800 drop-in price might receive ¥1,800 from a ClassPass visit, ¥1,000 from a Wellhub visit, and pay $99-300 per month to Mindbody before any of those bookings happen. The proxy showing platform-aggregator scale is travel: Booking Holdings reported $26.9 billion in 2025 revenue, almost entirely from commissions. Fitness's three-layer middleman extracts a smaller absolute number but at a comparable percentage of revenue from the studios that depend on them.

None of this is illegal or even unfair. The platforms genuinely brought studios customers in 2014-2022 when independent fitness was unbundling from big-box gyms. But the arbitrage window between 'platforms are the only discovery channel' and 'platforms are one channel among many' is exactly what AI agents are starting to open.

What ChatGPT and Claude actually read

An AI assistant answering 'boutique pilates near Shibuya station that takes drop-ins' does roughly four things in sequence: it issues real-time search queries via its crawler (OAI-SearchBot, Claude-SearchBot, PerplexityBot), it reads the returned pages as raw HTML, it extracts structured data, and it synthesizes an answer with citations. Each of those four steps has a specific failure mode for studios.

Step 1 — robots.txt. If your robots.txt blocks `OAI-SearchBot`, `Claude-SearchBot`, or `PerplexityBot`, you do not exist in the answer. Many studios accidentally block these because their developer copy-pasted a generic 'block AI bots' robots.txt template intended to block training crawlers (`GPTBot`, `ClaudeBot`). The training bots and the search bots are different. OpenAI documents the distinction here, Anthropic here, and Perplexity here. Allow the search bots even if you block the training ones.

Step 2 — server-side rendering. Google's own JavaScript SEO documentation confirms that not all crawlers run JavaScript. AI search crawlers do not document JS execution and should be assumed to read raw HTML only. If your class schedule lives inside a Mindbody iframe or a JavaScript-rendered React widget, the schedule is invisible to the AI. This is the single most common reason boutique studio sites fail audits.

Step 3 — Schema.org structured data. This is where studios can pull ahead of national chains by being more specific. Schema.org SportsActivityLocation is the right top-level type for a fitness studio. Where applicable, use the more specific subtype: `HealthClub` for full-service gyms, `ExerciseGym` for gym-only spaces. Required fields: `name`, full `PostalAddress`, `geo` at 5+ decimal places, `openingHoursSpecification` (per-day, including holidays), `amenityFeature` (showers, lockers, mat rental), `priceRange`, and `image`.

Step 4 — Event schema for individual classes. This is the highest-leverage move and the one almost no boutique studio does. Each scheduled class should be a Schema.org Event with `eventSchedule` (recurring `Schedule` with `byDay`, `startTime`, `endTime`), `instructor` as a Schema.org Person with the trainer's credentials (RYT-200, NASM-CPT, BASI-trained), `location` linking back to the SportsActivityLocation, and `offers` with the drop-in price. With this schema, an AI answering 'morning Vinyasa in Chiyoda' can find your class by name, time, and instructor.

Step 5 — Person schema for trainers. A trainer with verifiable credentials (`hasCredential` linked to the issuing body) is an authority signal AI assistants weight heavily. Yoga Alliance RYT-200 and RYT-500, NASM-CPT, ACE-CPT, ACSM-CEP, and BASI Pilates are all standardized credentials with public registries — link to the registry.

Step 6 — llms.txt. Add an llms.txt at your root following the Howard 2024 spec at llmstxt.org that links to your class schedule, trainer bios, intro-offer page, and FAQ. Cost: 5 minutes. Downside if it's never used: zero.

The KDD 2024 GEO findings, applied to fitness

The most rigorous public research on what lifts AI citation is the KDD 2024 paper *GEO: Generative Engine Optimization* by Aggarwal et al. (arXiv 2311.09735). The team tested seven content rewrites against Perplexity.ai across 10,000 queries. The results, and how they translate to a studio's content:

• Direct quotations from named sources: +41% citation lift (the highest of all tested patterns). For a studio: quote your founder, your head instructor, or a credentialed expert with an attributed name. Example: 'Yuki Tanaka, our head Vinyasa instructor and a Yoga Alliance E-RYT 500, runs the 7am sequence as a slow-build flow rather than a power class.' That single sentence — quote, named source, credential — is more AI-citable than five paragraphs of generic 'our classes are welcoming' copy.
• Statistics with cited sources: +31% lift overall, +37% on Perplexity specifically. Hard, verifiable studio numbers belong on the page. 'Average class size 12.' 'Studio capacity 18 mats.' 'Drop-in rate ¥3,800.' 'Member retention at 12 months: 64%.' Each gets cited if sourced; each is invisible if hidden in marketing fluff.
• Inline citations to authoritative sources: +27% overall, +115% on currently low-ranked pages. This is the most democratic finding in the paper — sites that are not yet visible in AI answers gain the most by adding citations. For a studio that means linking to the Yoga Alliance registry when you mention RYT-200, to NASM when you mention NASM-CPT, to ACSM for ACSM-CEP.
• Keyword density: zero lift. Stuffing 'yoga Shibuya' into your H1, H2, H3, and meta description does not move AI citation rates. Worse, keyword stuffing tested ~10% worse than baseline.
• Authoritative tone alone: zero lift. Writing in a confident voice without underlying citations or quotes does nothing. The AI is not impressed by your prose; it is anchored to verifiable signal.

Translate the +115% finding for low-ranked pages into operator language: a small studio in Setagaya that ranks page 3 on Google for 'pilates Setagaya' is the studio with the most to gain from adding inline citations. The paper's authors, lead by Pranjal Aggarwal, write that "GEO methods can democratize visibility — domains with low search authority benefit disproportionately from citation patterns." That is unusually good news for boutique fitness, where Google is dominated by chains and aggregators.

What AI discovery cannot fix

The honest counter-thesis matters. AI discoverability is not a substitute for running a good studio. If your classes are inconsistent, your space is dirty, or your schedule changes weekly without warning, the AI will eventually surface the negative reviews and route customers away. AI assistants triangulate between your structured data, your Google Business Profile reviews, and third-party mentions — three sources of truth, not one.

AI also cannot fix a category mismatch. If a customer asks for 'beginner-friendly Vinyasa' and your studio teaches advanced Ashtanga, no schema markup makes you the right answer. The structural data has to be truthful, not aspirational.

ClassPass and Wellhub also still have a real role: they drive discovery for new movers to a city who don't yet know any studio names. AI assistants are best at narrowing a known set of options; ClassPass is best at exposing a customer to options they would never have searched for. The realistic 2026-2028 model is hybrid — keep ClassPass for top-of-funnel sampling, but convert AI-routed direct bookings into your highest-margin acquisition channel.

And the broader caveat from the middleman tax piece holds: every prior 'disintermediation' wave produced a new aggregator. If one AI agent wins category share, agent-placement fees become the new commission. The current arbitrage window may be 5-10 years; it is not permanent.

The Monday playbook

Five concrete actions a gym or studio owner can take this week, ranked by leverage. The first three are DIY. The last two are the structured work most owners hire out — see our pricing page for the audit and build packages.

• 1. Audit your robots.txt today. Open `yourdomain.com/robots.txt` in a browser. If it blocks `OAI-SearchBot`, `Claude-SearchBot`, or `PerplexityBot`, fix it this week. Allow them explicitly. This is the single highest-leverage 10-minute task in the entire playbook.
• 2. Run the discoverability test. Open ChatGPT, Claude, and Perplexity. Ask each, in fresh sessions: 'best [your category] in [your neighborhood]', 'beginner-friendly [your specialty] near me with morning classes', and '[your specialty] near [nearest train station] under [your price].' Record whether you appear and who's cited. This is your baseline. Re-test in 30 days.
• 3. Claim and complete your Google Business Profile. Free, takes an hour at business.google.com. Add full hours, services (each class type as a separate service), photos, and the website link. AI assistants triangulate against GBP heavily.
• 4. Get SportsActivityLocation + Event schema deployed. This is the work most owners cannot DIY because the schema must be syntactically correct, the subtype must match the studio type, and class Events must reference the SportsActivityLocation by `@id`. Validate at search.google.com/test/rich-results. If this is the work you want help with, start here.
• 5. Add Person schema for every credentialed trainer, with `hasCredential` linked to the issuing body. Yoga Alliance, NASM, ACE, ACSM, and BASI all maintain public registries. Link to them. This is the authority signal that separates 'yoga studio in Chiyoda' from 'yoga studio in Chiyoda where I'd actually want to take a class.'

The bet for boutique fitness

AI shopping agents are not a magic wand for fitness. Most members in 2026 still discover studios through Instagram, Google, friends, and yes, ClassPass. The shift to AI-driven discovery will take years, possibly decades, possibly never if AI consolidation gives us a renamed Google.

But the work to be AI-friendly is also the work to be trustworthy on the open web. SportsActivityLocation schema helps Google. Event schema with named instructors helps Google. Server-side rendering helps Google. Verifiable trainer credentials help potential members. Allowing AI search bots to crawl your site costs nothing if the AI shift fails — and costs you everything if it succeeds and your competitor's schema is cleaner than yours.

The fitness middleman tax extracted roughly half a class's price for a decade because there was no other way for a boutique studio to be found at scale. That is starting to change. The studios that win the next five years are the ones whose websites speak the language the new agents read — and the ones who understand that owning the customer relationship is worth far more than ranking inside ClassPass's search.