The Trial Converted Itself. Nobody Had to Press Anything.
Most AI products are sold on a trial that bills unless the customer stops it. That design is a negative option feature, and it pulls in a disclosure standard, a consent standard, and a cancellation standard that operate independently of each other. Teams tend to fix the copy. The cancellation path is what actually gets enforced.
What Makes a Plan a Negative Option
A negative option feature is any offer where the seller treats the customer's failure to act as consent to be charged. The Restore Online Shoppers' Confidence Act (ROSCA) has prohibited charging a consumer through a negative option online without clear disclosure, express informed consent, and a simple cancellation mechanism since 2010; the FTC's more recent negative option rulemaking — widely known as the click-to-cancel framework — restated and sharpened those three obligations and has itself been the subject of litigation over its scope and effective dates.
That procedural churn is a reason to be careful about citing a specific rule provision, and not a reason to relax. The three underlying duties predate the rulemaking, sit in a statute the FTC enforces directly, and are mirrored in state automatic renewal laws that were never affected by any federal court challenge. A business waiting for the federal picture to settle is still fully exposed under the state statutes and under ROSCA itself.
Why AI Pricing Models Aggravate This
Conventional SaaS bills on a date. AI products increasingly bill on consumption — credits, tokens, generations, seats that auto-scale, or a trial capped by usage rather than by days. Each of those designs makes the charge harder for the customer to anticipate, which raises the disclosure burden rather than lowering it. "Your plan will top up when your balance falls below 10%" is a recurring charge with no predictable date, and the customer cannot set a reminder against it.
Usage-triggered auto top-up
A credit balance that re-bills at a threshold rather than on a calendar date. The customer cannot diary the charge, so the pre-purchase disclosure and the in-product balance warning are doing all the work.
Trial that converts to annual
A short trial converting directly to a twelve-month term is the highest-friction version of this pattern — the gap between what the customer expected to risk and what they were charged is the largest, and it is the fact pattern regulators and chargeback teams see most often.
Seat counts that auto-expand
Plans that add billable seats when a workspace grows shift the charge trigger to an action by someone other than the account holder. Whoever consented to the recurring charge is not who caused the increase.
Cancellation gated behind a retention flow
Interstitials, discount offers, and 'tell us why' surveys are permissible in principle, but each additional required step widens the gap between sign-up effort and cancellation effort, which is the comparison a regulator actually runs.
Cancellation that only pauses
A control labeled 'cancel' that suspends rather than terminates, or that terminates at the end of a term the customer did not know they had entered, is a disclosure failure presented as a feature.
The State Layer Runs Alongside
California's Automatic Renewal Law, New York's General Business Law provisions on automatic renewal, and comparable statutes in a widening set of states impose their own requirements: clear and conspicuous presentation of the renewal terms, affirmative consent, an accessible online cancellation mechanism for online sign-ups, and in some cases a reminder notice before a longer term renews. These are enforced by state attorneys general and, in some states, through private litigation — which means the practical enforcement risk for a small AI SaaS is more likely to arrive from a state consumer protection unit or a plaintiff's firm than from the FTC. The federal framework and the state statutes are cumulative, and compliance work should be scoped to the strictest applicable standard rather than the federal one.
Auto-Renewal Compliance Checklist for AI SaaS
- ☐State that charges recur, the amount, and the cadence in the checkout view itself — not only in linked terms
- ☐Give the exact date or the exact usage threshold that triggers the first charge after a trial
- ☐Place the disclosure immediately adjacent to the consent control, above the fold on mobile
- ☐Describe how to cancel in the same disclosure, in plain language
- ☐Separate the recurring-billing consent from acceptance of general terms and privacy policy
- ☐Store a timestamped record of what disclosure text was shown at the moment of consent
- ☐Version that disclosure text so an old customer's record resolves to the copy they actually saw
- ☐Avoid pre-checked boxes and consent inferred from continuing past a screen
- ☐Provide an in-product cancel control reachable in no more steps than sign-up required
- ☐Do not require a phone call, a support ticket, or a reply-and-wait email for an online sign-up
- ☐Make retention offers skippable in a single action, never a required step
- ☐Confirm cancellation in writing, stating the effective date and whether access continues
- ☐Send renewal reminders before long-term renewals where state law requires them
- ☐Instrument the cancel flow and review its completion rate as a compliance signal, not just a churn metric
- ☐Re-run the whole flow after every pricing or onboarding change — this is where drift enters
- ☐Track chargeback and dispute reason codes for 'unrecognized recurring charge' as an early warning
See what your checkout actually shows
Disclosure defects are almost always rendering defects — the term is in the code and not on the screen the customer saw. RatedWithAI scans the pages your buyers land on and reports what is actually presented, not what the template intended.
Scan Your Product for Free →Frequently Asked Questions
The FTC's click-to-cancel rule has been challenged in court. Do we still need to do this?
Yes. The three duties — clear disclosure, express informed consent, simple cancellation — come from ROSCA, which is a statute and was not disturbed by challenges to the rulemaking. The state automatic renewal laws are entirely independent of the federal rule's status. Treating the litigation as a reprieve mistakes the rule for the obligation.
Can we require a customer to email us to cancel if we respond quickly?
The standard is not how fast you respond; it is whether cancellation is as simple as sign-up and available in the same medium. If the customer subscribed self-serve online, a cancellation path that depends on your response introduces a step the sign-up never had. Response speed does not close that gap, and it is not something the customer can verify before choosing to subscribe.
Does a discount offer in the cancel flow count as an obstacle?
Presenting a save offer is not itself prohibited. The problem is structural: if the offer cannot be dismissed in one action, or if it is presented more than once, or if the actual cancel control is visually subordinate to the accept-offer control, the flow has added required effort to cancellation that sign-up never demanded. Design the save offer so declining is a single, equally prominent action.
We are a small team with a handful of paying customers. Is this a real risk?
The dollar exposure scales with customer count, but the structural defects do not — a bundled consent checkbox and a support-ticket-only cancel path are equally wrong at ten customers and ten thousand. Fixing them at ten is a copy change and a settings page. Fixing them later means reconstructing what disclosure every historical customer was shown, which is usually impossible because nobody versioned it.