Top 10 AI Agents forGyms and Studios in 2026

Every ranking of fitness AI you will read this year is built on numbers nobody outside the vendor has ever checked. Open five of them and the same shapes recur: a platform claims it resolves 60% of conversations, another promises 30% business growth, a third reports a 185% increase in tour bookings from a single unnamed deployment. None of those is a measurement. Each is a marketing assertion about the seller, published by the seller, with no denominator, no control group, no date range and, in most cases, no definition of the verb being measured.

In most industries that would be a soft criticism. In this one it is not soft, because the claim shape has already been tested by a regulator with subpoena power and it failed. In January 2025 the Securities and Exchange Commission entered an order finding that a publicly traded voice-AI company had made materially false and misleading statements about precisely this kind of figure — the company had told investors its product delivered “over 94% accuracy even in noisy environments” while, on the Commission's findings, the units "lacked the capability to take orders on their own and required substantial human involvement," with human workers abroad processing the vast majority of the work. Presto Automation consented to that order without admitting or denying the findings, and the remedy was a cease-and-desist order with no civil penalty. The findings run against Presto only; the order's unnamed "Supplier A" is a separate company against which the SEC made no findings whatsoever.

We raise it here for one reason. As of 23 August 2026, a fitness-software vendor's live pricing page publishes a resolution rate and a customer-satisfaction score with no denominator, no definition of "resolves," no sample and no date range. That is the same claim shape. It is not the same company, not the same industry, and no regulator has made any finding about that vendor whatsoever — we are scrupulous about that distinction and you should be too. The point is that the shape of the claim has already been found wanting once, which is why we do not score on it.

This piece sits inside a wider set of vertical rankings built on the same method; the cross-industry version, with the same refusals applied to general-purpose tools, is our 2026 ranking of AI agents across every category. The operations-side companion for this vertical — workflows, staffing and scheduling rather than vendor selection — is our guide to AI agents for fitness and gym operations.

One more thing before the table, and it is the reason this article is structured the way it is. In most verticals the binding constraint on an AI agent is an integration limit or a data-protection rule. In fitness it is neither. It is the law governing how a member cancels a membership — the most heavily litigated area in the industry — and an AI retention agent operates precisely inside it. Get the vendor choice wrong and you have wasted money. Get the cancellation flow wrong in California and the contract itself can be void, with treble damages and attorney fees on top. Those are not the same size of mistake.

We scored eight attributes, every one of which you can re-check yourself in under an hour.Date checked: 23 August 2026. Where a fact could not be confirmed from a primary source or the vendor's own live page, the cell reads "not publicly disclosed" — which is a scoring input, not a gap in our research.

• Documented public integrations — named on the vendor's own integrations or marketplace page, with a category label, not implied by a logo wall.
• Whether the product is genuinely agent-shaped: a named agent, the specific actions it takes without a human, and a published escalation rule.
• Whether the platform's API allows an agent to write — book, cancel, amend — at the tier you are actually on, or only to read.
• Pricing transparency: a published number, a published number with gated tiers, or a lead form. We never estimate a price on a buyer's behalf.
• Ownership and corporate status from public record — a portfolio page, a press release, a site footer, a legal entity name.
• Whether the vendor publishes a SOC 2 Type II attestation, an ISO 27001 certificate in its own legal entity's name, a DPA, or a PCI DSS Attestation of Compliance.
• Data portability and exit terms — whether the vendor commits publicly to letting you leave with your member, attendance and payment records.
• Whether any independent evaluation of the product exists. In this vertical, for every product, it does not.

What we refused to score, and why. Accuracy, resolution rate, containment, deflection, retention lift, churn reduction, revenue growth, hours saved, satisfaction. Every published figure of that kind in this market is the vendor describing itself. We name several of them below specifically so you can recognise them when a salesperson hands them to you.

On independent evidence, we hold a precise line. No independent benchmark of any commercial AI agent for gyms or studios exists — we searched on 23 August 2026 and located no third-party head-to-head evaluation, academic, trade or consumer. But it would be false to say no independent evidence exists in this market at all. It does; it measures members rather than software.

The best of it is Sperandei, Vieira and Reis, published in the Athens Journal of Sports, which monitored the records of 5,242 members of a fitness centre between January 2005 and June 2014 for twelve months after cancellation of membership or until re-enrolment, fitting a Cox proportional hazard regression. Its central finding, verbatim: "The general survival curve shows that 38% of members who drop out will return to activities within 12 months. Of those who return, more than half return within the first month."

Three figures we will not print as fact, with their origins. These circulate constantly in fitness sales decks and each one falls apart on contact with its source.

We also refuse, on sight, any phone-abandonment benchmark sold alongside a voice agent — the "7% of calls abandoned" and "85% never call back" genre. Those figures have no published methodology and they are precisely what a fitness voice-agent vendor will hand you first.

This sector consolidated harder than any other vertical we cover, and the largest move completed five months ago. An article that describes Mindbody, ClassPass, EGYM and Netpulse as four separate companies is describing 2024. Getting this wrong is the fastest way to be dismissed by the operator you are writing for, so here is the public record.

From Playlist's own newsroom, in dated order: Playlist launched on 4 June 2025; announced an agreement to merge with EGYM alongside $785 million in new equity investments on 15 January 2026; completed that merger on 31 March 2026; launched Kite, an enterprise product for multi-brand franchises, on 21 April 2026; appointed Palantir's CFO to its board on 17 June 2026; and launched AI Concierge, described as a 24/7 front-desk assistant for fitness and wellness operators, on 8 July 2026. Playlist's brand list, from its own homepage, is Mindbody, Booker, Kite, ClassPass and EGYM. Fritz Lanman is CEO and co-founder; EGYM founder Philipp Roesch-Schlanderer is co-founder.

Third-party outlets reported a $7.5 billion valuation for the combined company. That figure comes from press coverage linked on the vendor's own page rather than from any filing we read, so we report it as reported and prefer the $785 million equity figure, which appears in Playlist's own headline.

The second consolidation story is ABC Fitness. Thoma Bravo's own portfolio page records ABC Fitness Solutions as headquartered in Sherwood, Arkansas, founded 1981, invested in 2018, with Bill Davis as CEO — and it remains listed there as of 23 August 2026. ABC absorbed GymSales (announced 22 July 2020), Trainerize (announced 15 September 2020), Fitness BI (announced 8 April 2021) and Glofox (completed 25 August 2022), which is now sold as ABC Glofox. Reuters reported in August 2024 that Thoma Bravo was preparing a sale of ABC Fitness; we found no evidence that any sale completed, and ABC still appears on the live portfolio page. We say what the page says and no more.

The third is quieter. Daxko's product navigation lists Daxko Operations, Core, Club Automation, FitnessForce, Exercise, Zen Planner, SugarWOD, CSI, Motionsoft, Engage, Engage Pro, FrontDesk, AI Websites, AI Digital Intelligence, Payments, GroupEx Pro, Playerspace, Elevate and Revenue Recovery Services. That is nine or more formerly independent gym-software brands under one roof. An operator comparing "Zen Planner versus Club Automation" is comparing two products from the same vendor. We could not establish Daxko's ownership from any page we retrieved, so we name no owner for it — and neither should any article you read.

Fourth, Xplor Technologies holds Mariana Tek, Xplor Gym, Legend by Xplor and Xplor Recreation — the last of which now presents as "nextRec (Xplor Recreation)," a rename worth knowing because older comparisons use the previous name. Xplor's about and newsroom paths returned 404 to us on 23 August 2026, so we establish no ownership above Xplor itself.

Every cell below is either citable to a vendor's own live page or reads "not publicly disclosed." Checked 23 August 2026. Note the compliance column: it is uniform, and that uniformity is the single most important finding in this section.

Also checked and trading, but not ranked in the ten:Zen Planner (a Daxko brand, Studio from $99 a month, Website $99, Engage $249, branded app $39), Clubworx (Australia; AUD $0, $109, $149 and $169 monthly, with classic automation and no agent claim), TeamUp (TeamUp Sports, Inc.; priced by active customers, $189 a month for 101–200, with an "AI Business Advisor" copilot), Momence (Basic free, Pro $60, Custom $199 a month, with per-country processing fees published), Arketa (Individual $49 a month published, Studio tiers gated, with an AI-powered email builder), and WellnessLiving (published pricing page; its homepage returned 403 to us, so treat it with care and re-verify before shortlisting).

Not ranked, because they are no longer independently purchasable: Glofox (now ABC Glofox), Trainerize (now ABC Trainerize), GymSales (now ABC GymSales), Fitness BI (absorbed by ABC), Netpulse (joined EGYM, now inside Playlist), ClassPass and EGYM themselves (Playlist brands). Wellhub, formerly Gympass, is trading but is a corporate-wellness membership network rather than an operations agent a gym buys, and its site returned 403 to us, so we describe it no further.

The honest finding first, because it should shape how you read every entry: of everything in this category, one vendor ships a documented roster of named autonomous agents. The rest ship copilots, scoring models, or trigger-based CRM automation with the word "AI" applied to it.That is not a criticism of the rest — several of them are among the most transparent software companies we have looked at in any vertical, and one of the best of them makes no AI claim whatsoever. It is a warning about what you are actually buying when a page says "AI-powered."

As of 23 August 2026 there is no FTC trade regulation rule governing how a gym must let a member cancel — and that fact is far more dangerous than it sounds, because the obligation to make cancellation simple did not go anywhere. Both halves have to be said, in that order, or an operator draws exactly the wrong conclusion.

Half one: the rule is vacated. The FTC's 2024 Rule Concerning Recurring Subscriptions and Other Negative Option Programs — the "click to cancel" rule — was vacated in its entirety, nationwide, by the Eighth Circuit in Custom Communications, Inc. v. Federal Trade Commission, 142 F.4th 1060 (8th Cir. 2025), filed 8 July 2025. The status word is vacated. Not stayed, not enjoined, not on appeal.

Concluding that the Commission failed to follow procedural requirements under § 22 of the Federal Trade Commission Act ("FTC Act"), 15 U.S.C. § 57b-3(b)(1), we grant the petitions for review and vacate the Rule.

— Custom Communications, Inc. v. FTC, 142 F.4th 1060 (8th Cir. 2025), from the body of the opinion

Two framing points matter and are routinely garbled. First, the rule fell on a proceduralground — the Commission failed to issue a preliminary regulatory analysis after an ALJ found the annual economic impact exceeded $100 million — and the court expressly declined to reach the substance, writing that "because we hold the Commission's rulemaking process was procedurally insufficient and Petitioners demonstrated prejudicial error, we need not address Petitioners' other substantive challenges to the Rule." The court did not endorse anything. It said the opposite in terms: "While we certainly do not endorse the use of unfair and deceptive practices in negative option marketing, the procedural deficiencies of the Commission's rulemaking process are fatal here." Second, the vacatur was total because party-specific relief was not feasible given the rule's breadth.

A detail worth knowing, because it is on the face of a federal appellate opinion and is a matter of public record: the Health and Fitness Association— the trade body formerly named IHRSA — appears in the caption as an amicus on behalf of the petitioners, alongside the American Property Casualty Insurance Association, the Consumer Credit Industry Association, the International Franchise Association, the National Association of Spa Franchises and the Service Contract Industry Council. Amicus participation is not wrongdoing and we draw no further inference. It is simply useful to know whose interests the sector's principal data publisher represented in that litigation, given that the same body's attrition statistics circulate through every vendor deck in this market.

What the FTC did next, and why it does not reach you. The Commission conformed the CFR at 91 FR 6507, effective 12 February 2026, restoring the pre-2024 text and, per its own footnote, "changing the full name of the rule back to 'Use of Prenotification Negative Option Plans.'" That restored rule is 16 CFR Part 425, and its very first operative words settle the question: it applies "in connection with the sale, offering for sale, or distribution of goods and merchandise."

Status precision on what comes next: an ANPRM is not a rule. The FTC published an Advance Notice of Proposed Rulemaking on prenotification negative option plans on 13 March 2026; comments closed 13 April 2026. An advance notice is the firststep of a section 18 rulemaking — not a proposed rule and not a final rule. It matters that the Commission issued one, because the Fifth Circuit vacated the CARS Rule on the ground that the FTC "violated its own regulations when it failed to issue an ANPRM." Issuing one here is the agency starting over from the beginning. We make no prediction about a date, because none is supportable from the record.

Half two — and this is the half that gets an operator sued. Simple online cancellation is still legally required. Three layers do the work the vacated FTC rule would have done, and every one of them is stricter than the rule was.

ROSCA is the federal backstop, and unlike Part 425 it plainly covers gyms. 15 U.S.C. § 8403 makes it unlawful to charge a consumer "for any goods or servicessold in a transaction effected on the Internet through a negative option feature" unless the seller discloses all material terms clearly and conspicuously before obtaining billing information, obtains express informed consent before charging, and "provides simple mechanisms for a consumer to stop recurring charges." Three checkable steps follow. ROSCA reaches services; Part 425 reaches goods and merchandise. ROSCA reaches internet transactions, so any studio selling through a website, an app or a booking widget is inside it. And "simple" is undefined by rule text, so it is enforced case by case as an unfair-or-deceptive-practice matter — which is exactly the hook an AI retention agent can trip. We do not print a penalty figure, because the amounts adjust annually and we did not verify the 2026 number.

State health-club statutes are the law that actually governs a gym, they are unaffected by the Eighth Circuit, and they are far more specific than anything the FTC proposed. We read five directly on 23 August 2026. California caps a health-studio contract at three years and $4,400 total inclusive of initiation fees, caps any termination notice period at 30 days, requires that a contract be cancellable "in person, via email from an email address on file with the health studio, or via first-class mail," relieves a member who moves more than 25 miles (with a predetermined fee of no more than $100, or $50 once more than half the contract life has run), and then supplies the enforcement that makes all of it real: a contract that does not comply "shall be void and unenforceable as contrary to public policy," waivers are themselves void, and "judgment may be entered for three times the amount at which the actual damages are assessed plus reasonable attorney fees."

New York already has click-to-cancel, and it survived the Eighth Circuit untouched because it is a state statute.GBL § 624(4), as last amended 28 February 2025, requires that a club "shall accept notice of cancellation of a membership through methods including, but not limited to, website, electronic mail, telephone, mail, or in person," and that a club which lets a buyer contract through a website "shall accept a notice of cancellation of such contract through such website." It caps terms at 36 months and annual payments at $3,600, requires bonding filed with the Secretary of State scaling from $50,000 to $150,000 by contract length with additions by location count, requires a posted sign no smaller than nine by fourteen inches, and gives renewal-cancellation windows of 15 business days after an annual renewal takes effect and three business days after a monthly one.

Texas is a licensing regime, which surprises operators who expand there. A person may not operate a health spa or sell a membership without a health spa operator's certificate of registration; the application must be accompanied by sample contracts, proof of security and a $100 fee, and it is renewable annually. Security scales by prepaid membership dollars per location from $20,000 to $50,000. Terms are capped at three years. A contract is void if it does not comply, or if the seller did not hold a certificate at the time of contract. And note who enforces: the Secretary of State is a filing officer only and, in its own words, "does not have authority to regulate the business practices of a health spa" — enforcement runs through the Attorney General, a district or county attorney, or a private action by the member.

Florida is where AI-generated marketing copy itself becomes the offence. Fla. Stat. § 501.017(5) provides that no health studio may, orally or in writing, represent that a contract for future services is for a lifetime or a perpetual membership, "or use any words or combination of words which may tend to give a prospective buyer the impression that a contract or membership entitles the buyer to services or the use of facilities for an indefinite term." That is not a keyword ban. It is a ban on an impression. A generative sales agent composing its own offer copy at eleven at night, unsupervised, is a first-degree-misdemeanour risk under § 501.019(1). This is the clearest reason in the whole article why an agent that composes membership copy needs fixed-template output rather than free-form generation.

And Ohio is the trap for an automated cooling-off calculation. Ohio has no statute called a health club law; gyms are regulated as prepaid entertainment contracts under the Consumer Sales Practices Act, alongside dance studios and martial-arts schools. The cooling-off clock runs to midnight of the third business day after the date on which the first service under the contract is available — not the signing date — and stretches to the seventh business day if the facility or service was not available at signing. A rule engine that computes the window from a signed-at timestamp is simply wrong in Ohio, and the mistake is a Consumer Sales Practices Act exposure rather than a customer-service complaint.

An AI retention agent that tries to save a cancelling member is operating at the exact point where four bodies of law bite simultaneously — and one of them, unusually, writes the software specification for you. This is not a hypothetical exercise. At least one vendor in this vertical ships a named retention agent whose documented job includes engaging directly when a member chooses to cancel. That is a legitimate product. It is also the one product in this category where the statute tells you precisely how it must behave.

The controlling text is California Business and Professions Code § 17602(e), part of the strictest automatic-renewal law in the United States, as amended by AB 2863 (Stats. 2024, Ch. 515), effective 1 January 2025 and applying to contracts entered into, amended or extended on or after 1 July 2025. California expressly permits a save offer and then states exactly what it must do.

For purposes of subdivisions (c) and (d), providing a discount offer or other consumer benefit or informing a consumer of the effect of the cancellation shall not be considered an obstruction or delay, provided that the consumer remains able to cancel or terminate… If a consumer conveys a request to cancel by telephone, the business may present the consumer with a discounted offer, retention benefit, or information regarding the effect of cancellation, provided that the business first clearly and conspicuously informs the consumer that they may complete the cancellation process at any time by stating that they want to "cancel" or words to that effect… If a consumer conveys a request to cancel by an online system, the business may display a discounted offer… provided that the business simultaneously displays a prominently located and continuously and proximately displayed direct link or button entitled "click to cancel," or words to that effect.

— Cal. Bus. & Prof. Code § 17602(e)(1) and (e)(2)

That converts into three engineering requirements, and it is rare to be able to derive a specification this directly from statutory text.

1. 1.The disclosure fires before the offer, deterministically. It cannot be something the model decides to say. It is a fixed utterance emitted by the call flow before the retention branch is reachable at all — because if the model chooses when to say it, the model can choose not to.
2. 2.“Cancel” is an interrupt, not an intent to be classified. A generative agent that scores “cancel” as one candidate intent among several and continues persuading has, on that call, obstructed a cancellation. The word must terminate the retention branch unconditionally, in code, upstream of the model.
3. 3.One offer. § 17602(e) authorises presenting a discounted offer or retention benefit — singular. A model that improvises a second and third offer after a member has said cancel is outside the safe harbour the statute created, and the safe harbour is the only reason the first offer was lawful.

The rest of California's automatic-renewal law surrounds that safe harbour with obligations an agent can also trip.Cancellation must be available "in the same medium that the consumer used in the transaction." A business that lets a consumer accept online must let them terminate "exclusively online, at will, and without engaging any further steps that obstruct or delay," via a prominently located direct link or button or a pre-formatted termination email. Phone cancellation must actually be answered — the business "shall answer calls promptly during normal business hours and shall not obstruct or delay," and a voicemail requesting cancellation must be processed or returned within one business day. Annual auto-renewing agreements need an annual reminder. Initial terms of a year or longer need renewal notice at least 15 and not more than 45 days before renewal. A fee change needs notice no less than 7 and no more than 30 days beforehand. Verification of affirmative consent must be kept at least three years, or one year after termination, whichever is longer. And misrepresenting "expressly or by implication, any material fact related to the transaction" is independently unlawful. The statute's exemptions cover utilities, insurers, alarm companies, banks and credit unions — there is no fitness exemption.

Now the New York layer, which is the one most likely to be violated by accident.GBL § 626 declares a list of unfair and deceptive trade practices, and § 626(8) makes it unlawful for a seller to "misrepresent in any manner by the seller or his assignee the buyer's right to cancel under this article."

Whether any given vendor implements the § 17602(e) sequence correctly is not something we could verify, and we will not assert that any of them does or does not.What we can tell you is exactly what to ask for in a demo, and it takes four minutes: ask the vendor to run a live cancellation call in front of you. Watch for whether the "you can cancel at any time by saying cancel" disclosure is emitted before any retention content, every single time, including when you interrupt. Then say "cancel" mid-offer and watch whether the agent stops immediately or makes one more attempt. Then ask, in writing, whether that disclosure is a code path or a prompt instruction. Those three answers tell you more about the product than any deck will.

And do not conclude that click-to-cancel is dead.It is dead as a federal trade regulation rule. It is live California statutory text — the phrase "click to cancel" appears literally in § 17602(e)(2) — and live New York statutory text at GBL § 624(4)(c)–(d). The obligation moved down a level of government, where it is both stricter and enforced by attorneys general and private plaintiffs rather than by an agency mid-rulemaking.

If your agent dials or texts a member, the TCPA governs, an AI voice is an "artificial" voice for its purposes, and the 2026 revocation rule is fully in force. None of that is softening, and any source telling you the TCPA got easier for AI callers has inverted the law.

FCC Declaratory Ruling 24-17, adopted 2 February 2024 and released 8 February 2024, holds that AI-generated voices fall inside the TCPA's restrictions on artificial or prerecorded voices. Paragraph 6 is unusually direct: the ruling "makes clear that the TCPA does not allow for any carve out of technologies that purport to provide the equivalent of a live agent, thus preventing unscrupulous businesses from attempting to exploit any perceived ambiguity in our TCPA rules." The same paragraph records that no commenter opposed those conclusions. Paragraph 9 sets the duties: prior express consent to initiate such calls absent an emergency purpose or exemption; identification and disclosure of the entity responsible for the call; and, wherever the message includes or introduces an advertisement or constitutes telemarketing, specified opt-out methods. Under 47 CFR § 64.1200(b), the identification must use the name under which the entity is registered to conduct business, must state a callback number that is not the dialer's, and, for telemarketing, must provide an automated interactive opt-out mechanism within two seconds of the opt-out notification.

The trap. Insurance Marketing Coalition v. FCC (11th Cir., 24 January 2025) vacated only Part III.D of the FCC's 2023 order — the one-to-one consent provision — and the court recorded that the 2012 Order was not at issue. Prior express written consent for marketing survives, as does the artificial-voice classification. One narrow consent provision was vacated; the underlying regime stands, and an AI voice agent sits squarely inside it.

Now the part that is new for 2026 and that most fitness content has not caught up with. The revocation rule at 47 CFR § 64.1200(a)(10) applies today, and compliance with the neighbouring provisions was required as of 11 April 2025. There is exactly one waiver, and it is narrow. The Consumer and Governmental Affairs Bureau granted a limited waiver in Order DA 25-312, adopted and released 7 April 2025, covering only the revoke-all element — the duty to treat a revocation given in response to one type of message as cutting off calls and texts on unrelated topics. That single element was originally waived to 11 April 2026 and was further extended by the Bureau on 6 January 2026 to 31 January 2027.

A called party may revoke prior express consent, including prior express written consent, to receive calls or text messages… by using any reasonable method to clearly express a desire not to receive further calls or text messages from the caller or sender. Any revocation request made… using the words "stop," "quit," "end," "revoke," "opt out," "cancel," or "unsubscribe" sent in reply to an incoming text message… constitutes a reasonable means per se to revoke consent… If a reply to an incoming text message uses words other than [those], the caller must treat that reply text as a valid revocation request if a reasonable person would understand those words to have conveyed a request to revoke consent… All requests to revoke… must be honored within a reasonable time not to exceed ten business days from receipt of such request. Callers or senders of text messages… may not designate an exclusive means to request revocation of consent.

— 47 CFR § 64.1200(a)(10), current eCFR text read 2026-08-23

Note the natural-language obligation, because it is the part that turns a model metric into a legal exposure. The rule does not stop at seven magic words. It requires the caller to honour any reply "if a reasonable person would understand those words to have conveyed a request to revoke consent." That is a classification task with statutory consequences, performed by a model, on inbound consumer text — which is to say it is exactly the place where a classifier's false-negative rate stops being a dashboard metric and becomes a liability. Instrument it as such: sample and human-review the replies your classifier scored as not a revocation, not the ones it caught.

Two adjacent provisions complete the picture. Section 64.1200(a)(11) provides that revocation by a channel not on the list — a voicemail or an email to a number or address intended to reach the caller — creates a rebuttable presumption of revocation once the consumer produces evidence of the request, resolved on a totality-of-circumstances analysis. And § 64.1200(a)(12) is the one narrow safe harbour: a singleconfirmation text of the revocation is permitted, provided it merely confirms the request, includes no marketing or promotional information, and is the only additional message sent. Sent within five minutes of receipt it is presumed within the consumer's prior express consent; sent later, the sender must show the delay was reasonable.

That five-minute window is a latency budget on a compliance path, not a UX preference. If your revocation handling runs on a nightly batch, the presumption in § 64.1200(a)(12) does not protect the confirmation message you send the next morning. This is a small architectural decision with a legal consequence attached, and it is the kind of thing that gets designed correctly at build time or fixed expensively later — the general pattern we work through in our AI agent creation practice.

One deliberate omission. Several states impose their own consent rules with private rights of action. We did not read any of those statutes for this article, so we state that generically and give no state-specific figure, threshold or requirement. If you operate in multiple states, that is a question for your counsel rather than for a listicle — including this one.

This is an illustrative worked scenario, not a client outcome. We say that explicitly because an article that refuses unsourceable industry statistics and then asserts an unverifiable client result has destroyed its own standing. Every figure below is either from a federal statistical series or from a statute, both cited above. Nothing here describes a real engagement.

On win-back cadence, the only real evidence points to a narrow window. The peer-reviewed survival analysis discussed earlier found that 38% of members who drop out return within twelve months and that more than half of those who return do so within the first month — one Brazilian club, 2005 to 2014, not a US benchmark and not about software. Even hedged that heavily, it argues for concentrating a win-back sequence in the first four to six weeks after a lapse rather than running a year-long re-marketing cadence against people who were going to come back anyway, or who were never coming back at all. It also means any vendor attributing a twelve-month win-back rate to its own product owes you an answer about the counterfactual.

In this vertical the first failure is almost never a wrong class time. It is a sentence about cancellation rights that nobody reviewed. The table below is ordered roughly by how expensive the failure is, not by how likely it is.

On prompt injection, we frame it the way the evidence supports and no further: it is unsolved. Any agent that reads member-supplied text — an inbound email, a DM, a contact-form submission, a review you are replying to — is exposed to instructions embedded in that text. There is no filter that reliably prevents it, and vendors who imply otherwise are overselling. The defensible control is blast-radius reduction: no session that reads untrusted content holds a credential that can write to your billing system, change a membership or send an outbound message without a human in the path. Log every tool call an agent makes, and alert on any action initiated inside a session that read external content.

Every row below is drawn from a statute or a rule quoted earlier in this article rather than from our judgement — which is what makes it printable, and what makes it something you can hand to a vendor. The one row that rests on professional judgement rather than law is labelled as such.

These are our published bands and we do not vary them by vertical. They cover a build with us; they are not vendor licence costs, which sit on top and which, for six of the ten products above, are not publicly disclosed at all.

Senior-led work is $150–$225 per hour. Retainers run $2,500–$9,500 per month. Every engagement carries a 30-day post-launch warranty, and full source-code and IP ownership transfers to you. We return a fixed-price phased proposal within 5 business days of a discovery call. We are a senior-led, Black-owned agency based in Los Angeles, reachable at +1 (424) 272-5601.

For most independent operators the honest first move is not a build. It is the audit — because the question that decides everything downstream is whether your current platform tier lets an agent write. On at least one platform in the table above, published tiering means an agent that needs to book, cancel or amend anything requires the top subscription tier, and read-only API access buys you an agent that can only answer questions. Finding that out in week two of an integration is an expensive way to learn it. Get it in writing first.

Nine signals, each of which you can check yourself, most of them in under ten minutes. None requires a technical evaluation and none depends on a number only the vendor can see.

1. 1.A published performance percentage with no denominator. “Resolves 60% of conversations,” “grow 30%,” “185% increase in tour bookings.” Ask what counts as resolved, over what period, across how many accounts, measured by whom. If the answer is not written down, the number is not evidence.
2. 2.Two vendors each claiming about 40% of the US market. Both cannot be right. At least one of those claims is wrong, and a vendor that publishes an unmethodologised market-share figure will publish unmethodologised performance figures too.
3. 3.A compliance page citing certificates that belong to the hosting provider. Confirm the certificate names the vendor's own legal entity, not its cloud provider. This has been found on live vendor pages in adjacent verticals more than once.
4. 4.No answer to the PCI question. Ask: who is the merchant of record, at which PCI DSS validation level are you assessed, and will you provide your current Attestation of Compliance? No vendor in this vertical answers that on a public page. Every one of them should answer it in an email.
5. 5.A “revenue leak” or “cost of inaction” calculator presented as evidence. It takes your own guesses as inputs and returns a loss figure. That is a lead-capture instrument, and its output tells you nothing about the product.
6. 6.A retention agent whose “say cancel to stop” disclosure is a prompt instruction rather than a code path. Ask directly. If the model decides when to say it, the model can decide not to.
7. 7.Cancellation-rights language generated by the model. In New York, misrepresenting the buyer's right to cancel is a declared unfair and deceptive practice that reaches the seller “or his assignee.” Require fixed templates that you own and version.
8. 8.No published data-export commitment. One vendor in this category publishes an unconditional export right including on your way out, at no cost, and names the counter-practice explicitly. Hold the others to that standard and read the exit clause before signature, not after.
9. 9.An integration list you cannot verify on the vendor's own documentation. A logo wall is not an integration list. A named partner with a category label and a documentation URL is.

One more, and it comes from a vendor in this roster giving advice against its own short-term interest, which is why it is worth quoting: "We strongly advise every operator to insist on an API for every piece of gym software they integrate into their operations — and why we also recommend an audit of your current tools' data capabilities. No API? Ask the provider when this will be available. Ask specifically: 'When is an API scheduled on the roadmap for delivery?'" That is exactly right, and the follow-up question this article adds is: and does my tier include write access, or only read?

Stated plainly, because the refusal is the product. An article that hides its gaps is asking you to trust it on the parts you cannot check.

• No independent benchmark of any product in this table exists. We searched on 23 August 2026 and located no third-party, head-to-head evaluation of fitness AI agents — academic, trade press or consumer group. Nothing here is ranked on accuracy, resolution rate, containment, satisfaction or ROI.
• No public SOC 2 Type II attestation was located on any fitness-software vendor's site. Not one. Every /security, /trust-center or /legal/security path we tried returned 404, resolved to the homepage, or was blocked to a plain fetcher.
• No public trust centre, no published DPA, no data-residency or retention commitment, and no PCI DSS Attestation of Compliance was located for any vendor in this vertical — despite every product storing card credentials and running recurring billing. That is a statement about what is publicly findable, not a statement that any vendor lacks these things.
• Several vendors' sites could not be fully retrieved. Mindbody's trust, security, privacy and integrations paths returned 404; ABC's security, trust-centre and partner-marketplace paths resolved to its homepage; Daxko's security, trust-centre, FrontDesk, AI and API-documentation paths returned 404; Xplor's about and newsroom paths returned 404; Wellhub and WellnessLiving returned 403. Statements about those pages' contents are therefore absent, not negative.
• Ownership could not be established from public record for Daxko, Xplor Technologies, Keepme, Gymdesk, Momence, Clubworx, WellnessLiving or Arketa. We name no private-equity owner for any of them, and neither should any source you read.
• Two vendors each claim about 40% of the US market. At least one is wrong. We say so rather than picking.
• The Ohio and Texas statutory texts here were read via statutory mirrors, because the state hosts refused a plain fetcher or served a JavaScript shell. The Texas Secretary of State's health spa FAQ, and all California, New York and Florida texts, came from state servers directly. Any direct quotation of an Ohio or Texas section should be re-checked in a browser before you rely on it.
• The FTC's next move on negative-option rulemaking is genuinely unknown. An advance notice closed for comment in April 2026 and nothing further is verified. Any date-certain prediction you read about a new federal click-to-cancel rule is speculation.
• No state or federal AI-specific statute governing fitness businesses was located. We do not import the Colorado AI Act, the EU AI Act or any general AI statute into this article as though it regulates gyms.
• We did not read any state mini-TCPA statute for this article, so no state-specific consent figure, threshold or requirement appears anywhere in it.
• HIPAA does not generally apply to a commercial gym, which is normally not a covered entity, and we have deliberately not imported healthcare framing. A medically integrated or physician-referred programme, or a facility administering a health plan benefit, may be different — that is a fact-specific question this article did not investigate.
• Mindbody's AI Concierge is six weeks old at publication. Nothing about how it behaves in production can be known yet, by anyone.
• Prompt injection is unsolved. Any agent that reads member-supplied text is exposed, and the control is blast-radius reduction rather than prevention.
• Vendor customer counts, location counts and funding figures are as published by the vendor. We did not independently audit any of them and we do not restate them as market data.
• This article is not legal advice. Every statutory provision cited here should be confirmed with counsel licensed in the states where you hold members, particularly before you deploy anything that touches a cancellation.