Auto-Renewal and Cancellation Research: Testing Whether Subscribers Actually Understood

The federal statute governing online subscriptions requires three things - clear and conspicuous disclosure of material terms, express informed consent, and a simple mechanism to stop recurring charges - and it defines none of them. Those are not legal questions with lookup answers. They are empirical questions about what your subscribers understood, and the only way to answer them is to ask people who just went through your flow.

This guide covers the current state of the law after a significant court decision, how to design a subscription comprehension study, and a cancellation symmetry audit you can run this week.

A note on scope: this is about whether your disclosures worked. It is not about why customers leave - for that, see cancel-flow exit interviews and churn surveys. The two get confused constantly because they happen on the same screen. One asks "why are you going?" The other asks "did you ever understand what you agreed to?"

Where the law actually stands

This area moved twice in eighteen months, and a lot of published guidance is out of date.

The Restore Online Shoppers' Confidence Act (ROSCA), 15 U.S.C. 8403, is the operative federal statute and has been since 2010. It makes it unlawful to charge a consumer for goods or services sold online through a negative option feature unless the seller:

1. provides text that clearly and conspicuously discloses all material terms of the transaction before obtaining the consumer's billing information;
2. obtains the consumer's express informed consent before charging; and
3. provides simple mechanisms for a consumer to stop recurring charges.

The FTC's 2024 Negative Option Rule - widely known as the Click-to-Cancel rule - attempted to put detailed specifications behind those three phrases. On 8 July 2025 the Eighth Circuit vacated the rule in its entirety on procedural grounds, finding the Commission had failed to conduct a required preliminary regulatory analysis. The rule is not in force. On 11 March 2026 the FTC announced an Advance Notice of Proposed Rulemaking to begin the process again, with comments due 13 April 2026.

Two practical consequences follow, and teams routinely get both wrong:

• ROSCA did not go away. Neither did Section 5 of the FTC Act, and neither did state automatic renewal laws, which in several states are stricter than the vacated federal rule. The obligations survived; only the federal specification of them disappeared.
• The vacated rule is still the best available research brief. It no longer carries the force of law, but it remains the most detailed public statement of what the FTC believes adequate disclosure, consent, and cancellation look like - written by the agency that will evaluate your flow under ROSCA anyway. Treating a vacated rule as a compliance checklist would be a mistake. Treating it as a hypothesis list for your research is smart.

The vacated rule as a testable specification

The rule's definition of "clear and conspicuous" enumerated requirements that read almost like a study protocol. The disclosure had to be easily noticeable and easily understandable by ordinary consumers; a visual disclosure had to stand out by size, contrast, location and duration; in an interactive electronic medium the disclosure had to be unavoidable; it could not be contradicted or mitigated by, or inconsistent with, anything else in the communication; and where a practice targeted a specific audience, "ordinary consumers" included members of that group.

Every one of those is measurable. "Unavoidable" is a question you settle by asking people what they saw. "Not contradicted by anything else in the communication" is a question about the whole screen, which is why testing the disclosure text in isolation tells you almost nothing.

The state layer you cannot ignore

California's Automatic Renewal Law, amended by AB 2863, is the most demanding widely-applicable standard and is worth designing to regardless of where you are based. Its current requirements include presenting the renewal terms clearly and conspicuously in visual proximity to the request for consent, obtaining express affirmative consent to the renewal terms specifically, providing a retainable acknowledgment, and maintaining verification of consent for at least three years (or one year after termination, whichever is longer).

Two provisions matter especially for research design:

• The law prohibits including any information in the contract that interferes with, detracts from, contradicts, or otherwise undermines the ability of consumers to give affirmative consent. That is a claim about interference effects, which is a comprehension question.
• On cancellation, a business may still present a discount offer or retention benefit, provided it prominently and proximately displays a button that immediately effectuates cancellation on the same page. So a save offer is permitted; a save offer that functions as a detour is the risk. Whether yours reads as one or the other is exactly what a study tells you.

The three research questions

Question 1: comprehension at the point of consent

Recruit people who have just completed a real signup - within minutes, not days. Recall decays fast, and the number you want is what they carried out of the flow, not what they can reconstruct later.

Ask unaided first, always: what did you just sign up for, and what happens next? Then move to specifics: will you be charged again, when, how much, and what would you do to stop it. The gap between the unaided answer and the true terms is your comprehension deficit, and it is the single most useful number in this whole area because it is the one a regulator would generate if they investigated you.

A useful KPI here, and one almost nobody tracks: the renewal-terms accuracy rate - the proportion of new subscribers who can correctly state the renewal date and the renewal amount without prompting. Score it against the actual terms. It is directly comparable across flow variants, it is trivially explainable to an executive, and unlike a conversion metric it moves in the right direction when you make the disclosure more honest.

Question 2: consent that was actually informed

The distinction the law draws is between agreeing to buy something and agreeing to a recurring charge. Those can feel like one action to a customer even when the interface technically separates them.

Test it directly. Did you agree to a subscription or make a one-time purchase? Was there a free trial, and what happens when it ends? Did anything on the page tell you how to cancel? Yes/no questions give you countable rates per flow variant.

Watch for the interference pattern specifically. Urgency banners, countdown timers, social proof, and pre-checked upsells all compete for the attention the disclosure needs. When comprehension drops between two variants that have identical disclosure text, the cause is almost always something else on the screen.

Question 3: the cancellation symmetry audit

This is the cheapest high-value thing in this guide, and you can do the first half without recruiting anyone.

Document both journeys side by side and count: number of screens, number of clicks, number of distinct decisions, whether a login is required, whether a human or chatbot interaction is required, available channels, and hours of availability.

Any row where cancelling is harder than subscribing is a finding. The vacated rule would have required the cancellation mechanism to be at least as easy to use as the mechanism used to consent, and to be available through the same medium; it also provided that a consumer must not be required to interact with a live or virtual representative - including a chatbot - to cancel if they did not do so to subscribe. Whatever the federal rule ends up saying after the new rulemaking, that symmetry principle is the intuition regulators and state legislatures keep converging on.

Then run the second half: give real subscribers the task of cancelling and observe. Time to completion, abandonment points, and the moments where people say they thought they had already cancelled are the outputs that matter. That last one is worth isolating - subscribers who believe they cancelled but did not are the population that generates complaints, chargebacks, and regulatory attention, and they never appear in your funnel analytics because from your system's perspective nothing happened.

Running this with Koji

The awkward part of this research has always been timing. Comprehension has to be measured immediately after the flow, at a sample size large enough to compare variants, and with follow-up probing that a static form cannot do. Scheduling moderated sessions destroys the timing; a survey link preserves the timing but gets you five words and no ability to ask why.

Koji resolves that by triggering an AI-moderated conversation right after signup or cancellation. The AI moderator asks the unaided question, then probes each answer - what on the page told you that, what did you expect to happen - in the same session, across every participant at once. You get verbatim comprehension data with the recall intact, plus countable rates from the same study.

All six structured question types earn their place in this design:

Because studies are versioned, running the same instrument against a new flow gives you a before-and-after you can put in front of counsel. Keep the results in a research repository with the asset versions attached - the acknowledgment record requirement runs three years, and your evidence of a good-faith comprehension process should outlive the flow it tested.

What good looks like

• A renewal-terms accuracy rate you track per flow version, not a one-off study
• A symmetry table with no row where cancelling is harder
• Observed cancel-task completion, including the believed-but-did-not-cancel rate
• Retained records of variants you rejected and why
• Re-testing whenever the page changes, since a promotional banner can undo a compliant disclosure without anyone touching the disclosure

Teams that measure this usually find the same thing: the disclosure text is fine, and something adjacent to it is doing the damage.

Frequently asked questions

Is the Click-to-Cancel rule still in effect?

No. The Eighth Circuit vacated the FTC's Negative Option Rule in its entirety on 8 July 2025 on procedural grounds, and the FTC opened a new rulemaking with an Advance Notice of Proposed Rulemaking announced 11 March 2026. However, ROSCA, Section 5 of the FTC Act, and state automatic renewal laws all continue to apply, and several state laws impose requirements comparable to or stricter than the vacated rule.

If the rule was vacated, why design research around it?

Because it remains the most detailed public description of what the agency enforcing ROSCA considers adequate disclosure, consent, and cancellation. It is not a compliance checklist any more, but it is an excellent hypothesis list. Its definition of clear and conspicuous - noticeable, understandable, unavoidable in interactive media, not contradicted by anything else on the screen - translates almost directly into testable study questions.

How is this different from a cancel-flow exit interview?

An exit interview asks why the customer is leaving, to inform retention. This study asks whether they understood the terms they agreed to and whether cancelling was as easy as subscribing, to inform compliance and design. They happen on the same screen and answer completely different questions, so keep the instruments separate or the retention framing will bias the comprehension answers.

When should we run the comprehension study relative to signup?

As close to immediately as you can manage - within minutes. Recall of a disclosure decays quickly, and delayed answers measure reconstruction rather than comprehension. Triggering an AI-moderated conversation at the confirmation step preserves the timing while still allowing follow-up probing.

What sample size do we need?

For surfacing the misunderstandings that exist and explaining why they form, 25 to 40 probed conversations per flow variant is usually sufficient. If you intend to report a comprehension rate as a number and compare variants statistically, size it as a proportion estimate and report a margin of error alongside it.

Does offering a discount during cancellation create a problem?

Not inherently. California's amended law expressly permits presenting a discount offer, retention benefit, or information about the effects of cancellation, provided the business prominently and proximately displays a button that immediately effectuates cancellation on the same page. The risk is a save offer that functions as a detour rather than an option, and observed cancel-task testing is how you find out which one yours is.

Related Resources

• Structured Questions Guide - the six question types and when to use each
• Cancel-Flow Exit Interviews - the retention-side conversation, kept separate from comprehension
• Reference Prices and Drip Pricing - testing a price display without making a deceptive claim
• AI Research for Subscription Businesses - the wider subscription research programme
• Content Testing - testing microcopy, labels, and UX writing with real users
• Churn Surveys - why customers leave, as distinct from what they understood
• Research Repository Guide - retaining evidence your legal team can find

---

Test your signup flow this week. New Koji accounts include 10 credits - enough to run a full post-signup comprehension study with AI follow-up probing and find out what your subscribers actually think they agreed to.