CASE FILE // Exhibit A
The pattern

What actually works.

We coded hundreds of real-world reports, Reddit, X, Hacker News, GitHub, developer forums, provider status pages, real support transcripts, to find what moves the needle. The pattern is brutally consistent.

You don't have to argue your case. You have to show it.
Gets results✓
  • Their documents, quoted backInvoice numbers, status incidents, in-app banners.
  • Numbered facts, one precise ask"Invoice 0042, paid July 5, $200," with an exact remedy.
  • TimestampsLet them check your account against their own incident log.
  • Owning your own errorsA precise correction makes every other claim more credible.
  • Calm persistenceResolved cases take two to four exchanges. Round one is a bot.
Gets closed
  • The angry first messagePattern-matched into cancellation or feedback flows. Sometimes in minutes.
  • "Cancel" or "refund" to a botExecuted literally, even when you meant a credit.
  • "Your service is terrible"Filed as feedback. Closed.
  • Threats and caps lockFiled under "angry." Disengaged.
  • One giant blended grievanceA blur gets a blur back. Separate claims must each be answered.
The quiet part

What we're expected to accept.

The embarrassing part, and why nobody wants to make a fuss: we're being led to believe this is normal. That the fault is ours.

What they imply
  • "The model's just like that sometimes."Degraded output inside a fully billed month is a service failure, not a mood.
  • "You must be using it wrong."You didn't change. The model and the meter changed under you, mid-subscription.
  • "That's just how limits work."Advertised capacity that never materializes is not your misunderstanding.
  • "AI is new; faults are expected."Faults, maybe. Charging full price through them, silently: no.
The actual positionyours
  • You are not imagining itHundreds of public reports describe the same failures, the same months, the same walls.
  • It is not your faultNot finding workarounds for a degraded product is not a skills gap.
  • It is not normalIn no other industry is "we quietly made it worse" an accepted answer to an invoice.
  • Enterprise customers don't accept itTheir disputes get escalated to humans. The double standard is the tell.
You are not making a fuss. You are asking a company to honor its own receipts.
The realities

True at every provider.

01

Round one is a bot

First contact is almost always automated. Expected, not failure.

02

Emotion gets misfiled

Story-led first messages get routed into cancellation or feedback flows.

03

Their documents win

Numbered facts, one precise ask, and their own records get further than any argument.

04

Words are triggers

"Cancel" or "refund" to a bot can execute literally, even when you meant a credit.

05

Degradation is provable

Dated examples plus the provider's own status history beat adjectives.

06

App stores complicate

Apple and Google add a second party. Providers can sometimes still refund directly.

The through-line: most documented cases end unresolved, unless the user escalates with evidence. That is the gap this tool closes.

The anatomy of a winning message

Exhibit B, case anatomytemplate
  • Opens with identifiers: ticket ID, account email, plan. Trivially easy to look up.
  • Numbered facts referencing evidence IDs (E01, E02…), each verifiable against a document.
  • One separate numbered claim per issue, never blended. Each must be answered individually.
  • One numbered ask per claim with exact amounts and mechanism.
  • A 7-day response deadline and a request that any declined item be explained individually.
  • A request for a human with authority.
  • Under 800 words. No threats, no adjectives, no history lesson.

The wizard generates a prompt that makes your own AI write this for you, from your actual documents.

About the data

The patterns on this site come from a coded dataset of public reports (currently in the hundreds and growing), gathered across Reddit, X, Trustpilot, Hacker News, GitHub, developer forums, provider status pages, and support transcripts. It is a sample of public reports, not market-wide failure rates, and it currently reads deepest in Claude communities, so provider-to-provider comparisons should be read cautiously. The useful signal is consistent everywhere: documented cases mostly end unresolved unless the user escalates with evidence.

Put it to work, start my case → Read the provider dossiers
Copied