Cancel Atlas

The Cancellation Friction Index

Insights

Cancellation friction, in aggregate

What 2075 cited company records say in the aggregate, where the cancel journey breaks, which industries trap you, and how far "easy in, hard out" really goes. Every chart is Documented-Policy tier (cited public pages); Verified-Flow behaviour audits are pending, so these describe what companies publish, not yet what they do. The same numbers are machine-readable at /api/v1/insights.json.

Context

The state of cancellation in 2026

Why a Cancellation Friction Index exists, and why this page measures what 2,075 companies publish, set against what the law, the research, and regulators say. All figures below are externally sourced; the charts that follow are Cancel Atlas’s own computed data.

  1. Regulation is in flux, and the headline US rule is not in force.

    The FTC's federal “Click-to-Cancel” rule was vacated by the 8th Circuit on 8 Jul 2025 (procedural grounds) and is not law; the FTC reopened rulemaking in 2026. Meanwhile California's Automatic Renewal Law (live 1 Jul 2025) and the UK's DMCC subscription regime (≈Spring 2027) do mandate easy, same-channel cancellation. Which rules bind you depends on jurisdiction, which is why every Cancel Atlas score is tagged per jurisdiction. Sources: 8th Cir. vacatur · California ARL · UK DMCC.

  2. The harm is large, recurring, and mostly invisible to the people paying it.

    US consumers underestimate their own subscription spend by about $133/month (guess $86 vs actual $219), 42% have forgotten an active charge, and only about 24% rate cancelling “very easy.” FTC negative-option complaints ran ~70/day in 2024, up from ~42/day in 2021. Sources: C+R Research · West Monroe · FTC.

  3. The friction is asymmetric by design, and now measured.

    Peer-reviewed research (CHI 2024) clocked subscription cancellation at 6.2 clicks vs 4.3 to subscribe in the EU (3.5→5.8 in the US; one outlet took 10 clicks to leave). Large-scale audits find 76-97% of sites use at least one dark pattern, with “obstruction” / hard-to-cancel among the most common. The principle regulators now codify: leaving should be as easy as joining. Sources: Staying at the Roach Motel, CHI 2024 · OECD · Dark Patterns at Scale.

  4. Enforcement is catching up, and validates naming names.

    Amazon settled the FTC's Prime-cancellation case for $2.5B in 2025 over its “Iliad” cancel maze; Adobe (DOJ), Uber One, Match/Tinder, Cerebral, ABCmouse and others face or have settled similar actions. When a regulator has already documented a company's friction, an adverse grade rests on public record. Sources: Amazon $2.5B · FTC dark-patterns report.

Full cited synthesis: see the methodology notes. Cancel Atlas turns this diffuse, well-documented harm into per-company, per-jurisdiction, dated, cited evidence, the gap no neutral, machine-readable index has filled.

Distribution

How the 2,075-strong index grades out

Grade distributionGrade distribution across 2075 companies: A 99, B 1224, C 572, D 151, F 29A grade99 (5%)B grade1224 (59%)C grade572 (28%)D grade151 (7%)F grade29 (1%)⊗ cancelatlas.com

Most companies land at B/C on documented transparency. A grades are rare (a clear, self-serve, well-disclosed exit); F grades mark phone-only or undisclosed terms. Documented-Policy tier, behaviour audit pending.

Easy in, hard out

The asymmetry: joining vs leaving

Cancellation channel asymmetry11% of companies offer no online cancel; 29% route you to a phone.Self-serve online cancel84%Phone is a cancel channel29%No online cancel at all11%⊗ cancelatlas.com

Signing up is almost always one online click. Leaving isn't: 227 companies (11%) offer no online cancellation at all, and 29% make a phone call one of the only ways out. That gap, easy in, hard out, is the core of the index.

Where it breaks

The weakest part of the cancel journey

Mean score per CES dimensionAverage of each CES-1.1 dimension across all companies; lowest = the system-wide weak point.Refund disclosed65.9 avgNotice disclosed68.5 avgSelf-serve cancel path68.9 avgFindability72.7 avgChannels disclosed77.6 avg⊗ cancelatlas.com

Averaged across the whole index, self-serve cancel paths are the weakest dimension (68.9/100), companies disclose that you can cancel more clearly than they give you a button to do it. In fact 185 companies (9%) publish no findable direct cancel path at all. Weighted: cancel-path 30, channels 20, notice 15, refund 15, findability 20.

By industry

Hardest and easiest industries to leave

Industry friction leaderboardMean documented-policy score per industry (>=8 companies), worst first.Business formation & legal (n=9)55.3SEO & social tools (n=10)57.3Telecom & ISP (n=216)64.3Dating apps (n=45)67.1Fitness & gym (n=146)67.3Energy & utilities (n=33)68.4Meal kits & food (n=102)69.7Travel & mobility (n=27)69.9Retail & membership (n=46)70.1Apparel & beauty (n=60)70.3Health & telehealth (n=54)70.6News & publishing (n=108)70.7AI tools (n=104)70.8Software & SaaS (n=265)71.9Developer tools (n=95)72.1Insurance (n=55)72.4Security & identity (n=75)72.4Mental health & wellness (n=38)73.1Personal finance & fintech (n=99)73.2E-learning (n=112)74.0Pet subscriptions (n=38)74.2Video streaming (n=208)74.8Food & grocery delivery (n=35)74.9Music & audio (n=48)75.8Gaming (n=24)76.0Audiobooks & ebooks (n=20)76.2⊗ cancelatlas.com

Mean documented score per industry (≥8 companies), hardest at top. The pattern is consistent with the thesis: contract/phone-gated categories (insurance, telecom, gyms) trail self-serve software/streaming. Category mix matters, see the jurisdiction caveat below.

Machine-readable signals

Structured signals, at a glance

Tier-A structured signalsShare of companies with each documented structured signal.Documented cancel method98%Cites a refund page39%Offers pause / freeze24%Phone / in-person to cancel7%⊗ cancelatlas.com

These are the first machine-readable structured signals (/api/v1/signals.json), derived per record from its cancellation channels, links, and pause data, for the agents and tools that increasingly cancel subscriptions on people's behalf. Pricing model across the index: 2,029 recurring, 43 one-time/lifetime, 3 hybrid. Deeper text-extracted signals (refund-claim conflicts, retention walls, notice windows) are in development.

Refunds

What happens to your money when you leave

Refund stance distributionShare of companies by documented refund stance.No refund (explicit)672 (32%)Conditional / windowed415 (20%)Not documented988 (48%)⊗ cancelatlas.com

Where a refund stance is documented, “no refund” is the most common (32%); 20% offer a conditional/windowed refund, and 48% publish no refund stance at all (an absence, not necessarily a denial). On notice: of 566 companies with a documented posture, 446 let you cancel anytime and 120 impose an advance-notice window. The source-page pass re-fetched 1,251 of 2,075 records' cited URLs (60% readable), adding source-verified governing law (299 named), arbitration clauses (323), and EU-withdrawal waivers. A cross-page heuristic flagged 127 possible marketing-vs-terms conflicts; on editor review 5 were genuine contradictions, the rest were consistent conditional policies (a money-back window with non-refundable exceptions), deliberately not published. Machine-readable at /api/v1/signals.json.

Documented by regulators

Enforcement-anchored: companies a regulator has named

These 12 companies have a government (FTC/DOJ/State-AG) or court-approved settlement specifically about cancellation difficulty or deceptive auto-renewal, their friction is established public record, not Cancel Atlas’s opinion. Note the tension: several grade B/C here on documented policy (e.g. Amazon Prime, yet fined $2.5B for its actual cancel flow). That is the disclosure-vs-behaviour gap our pending flow audit is built to close: what a company publishes and what it makes you do can diverge sharply. Each row links the official source. Machine-readable in /api/v1/signals.json (regulatory_action). Penalty figures are as reported; verify against the primary docket. Pending/ongoing matters are labelled.

Transparency

Which official policy pages are cited

Official-policy-link coverageShare of records with a cited cancel / refund / terms / privacy page.Cancellation page97%Privacy / data page49%Terms page56%Refund / returns page39%⊗ cancelatlas.com

Every record cites a cancellation page; far fewer surface a dedicated refund, terms, or privacy page. Low coverage means “not published or not yet recorded,” not a definitive negative, these links never affect the score.

Pause instead of cancel

Where you can freeze rather than quit

Pause availability by industryShare of companies per industry offering a documented pause/freeze.Apparel & beauty (n=60)87%Meal kits & food (n=102)87%Pet subscriptions (n=38)82%Fitness & gym (n=146)57%Health & telehealth (n=54)44%Telecom & ISP (n=216)33%E-learning (n=112)21%News & publishing (n=108)17%Audiobooks & ebooks (n=20)15%Food & grocery delivery (n=35)14%Retail & membership (n=46)13%Video streaming (n=208)13%⊗ cancelatlas.com

A pause option can be genuine flexibility, or a retention nudge that keeps you subscribed. Shown per industry (≥8 companies); informational, not scored.

By jurisdiction

Average score by market (read with care)

Mean score by jurisdictionMean documented-policy score per jurisdiction (>=8 companies).MX (n=11)62.3SG (n=18)64.1PH (n=9)65.2PT (n=8)66.4ZA (n=16)66.5AE (n=15)66.7UK (n=49)67.5ES (n=28)67.8CH (n=13)68.5ID (n=9)69.0IN (n=43)69.5CA (n=47)70.0DK (n=10)70.2BR (n=25)70.4PL (n=11)70.6US (n=1228)70.7IL (n=9)71.7IT (n=19)71.9FR (n=41)72.1SE (n=17)72.2NZ (n=15)72.5CZ (n=9)73.6NO (n=9)74.2JP (n=16)74.4KR (n=19)74.6AU (n=63)75.3DE (n=60)75.8NL (n=13)76.3GB (n=85)76.6⊗ cancelatlas.com

Directional only, heavily category-confounded. A market's average reflects which kinds of companies we’ve indexed there as much as its law. The rigorous law-effect comparison needs same-category pairs (e.g. German vs US gyms, where Germany’s mandated cancel button moves the score within the same category). Treat this as a coverage map, not a league table.

Freshness

How current the index is

Freshness of the index1% of records were verified within 30 days of the most recent check.Verified ≤ 30 days ago2731-90 days ago2048Over 90 days ago0⊗ cancelatlas.com

1% of the index was checked within the last 30 days. Freshness is measured against the most recent verification date; this view surfaces decay as records age between re-checks (the weekly snapshot + planned auto-freshness engine keep it current).

Evolution

How the index is changing over time

Index evolution over timeMean score and key friction metrics across weekly snapshots.02550751002026-06-102026-08-17Mean score% A-grade% F-grade% no online cancel⊗ cancelatlas.com

Tracked across 3 snapshots (2026-06-10 → 2026-08-17). Mean score +2.3 since the first snapshot; net grade shift A +82, F +8.

The transparency floor

Companies with no published cancellation page

A clear, self-serve cancellation page should be table stakes. These 48 companies we checked publish no findable cancellation page at all, so there is nothing to grade. This is the absence of a policy, not an F (an F is a poor documented policy). We name them to push the standard forward, and we will score each one the day it publishes a real cancellation page. Listed a company in error? Show us the page and we'll grade it. Machine-readable at /api/v1/no-policy.json.

Generated 2026-06-04 from the live index (2075 companies). Charts are statements of opinion grounded in dated, cited public facts, computed from a published methodology; they never alter any company's score. Found an error? Submit a correction.