See what will stop your ad copy, email or post before it goes out.

Each stage files its output in outputs/claim-check/ in your project folder. Intake through filing run in /claim-check. The record after the piece goes out is /claim-recheck.

Intake.

You give the copy (pasted, as a file, or as the URL of a piece already live), where it goes and in what format, who it reaches, and one line on what you want to know. For email you also say how the list was built. You say whether the piece is part of a campaign and whether it uses AI to make decisions about people or to generate content shown to customers. You can add what unlocks checks: the sources behind each claim, any affiliate, sponsor or gifted relationship, whether AI made or altered an image, voice or video, consent records for quotes, and the day’s post count for the account. For email you send a test message from your real sending system to an inbox you control and paste it with its full headers. Anything missing makes its checks read NOT CHECKED, and the report says what would run them. For a piece already sent, you give the copy as delivered, and each fix applies to the next send.

Check.

Claim Check runs the sections in order, each against a written rubric: claims ledger, slop pass, platform fit, disclosures, send preflight for email, quotes and testimonials, a launch preflight for a campaign, and AI readiness when it applies. Every check names its evidence and returns pass, fix, stop or NOT CHECKED. A check that could not run is never a pass. Claim Check also reads the page your copy links to and compares the price, offer, terms and deadline with the copy. It reads a checkout or a form and never opens or submits it.

What will stop this.

Under the summary table, the report counts the blockers it found by type, then gives each blocker an entry. The types are in the next list.

Fix.

The fix list is numbered, addressed to whoever wrote the copy, and ordered by what each fix prevents. A fix can say to remove the sentence, supply the source, or limit the claim to what the source says, quoting the source’s words. It never writes replacement copy. A fix that changes something already public or already configured, such as a live post, a DNS record, a sender setting, a list or an ad in flight, is marked owner decision, with a rollback copy first. A step only you can take, such as sending the test to your own inbox or setting a platform flag, is marked owner step.

Gate and file.

Before you see it, the report goes to a verifier running in a fresh context. Confirmed findings get fixed. Judgment calls are listed for you. Claim Check then saves the report as a preflight file, with the copy verbatim and a stable id on each blocker.

After it goes out.

Run /claim-recheckwith the preflight file and tell it what happened: when the piece went out, and any of these you have: delivered, bounced, complaint or spam report, a platform rejection or label with its message, replies, edits made, and numbers you choose to share. Claim Check compares the copy as sent with the copy it checked and lists every difference. It marks each fix made, not made or NOT CHECKED, then sets your reported outcome beside each blocker and labels it “moved with”, never “caused”. Your numbers go in a column headed “Reported by the owner, not verified”. Claim Check computes no rate, compares nothing to a benchmark, and reads an open as an attention figure, never as proof that a message landed. The audience, the list, the timing, the subject line and the platform may all have changed too, and any of them can account for a difference. Claim Check has no server and watches nothing, so you start it.

The blocker types.

Unbacked claim.

A result, number, comparison or guarantee with no source, or a source that contradicts it. The evidence is the claim quoted, the source and the date it was read. A claim that matches only your own files reads as verified for consistency, and the row says so.

Platform rule.

A limit, format or policy that the rule on file for the platform states, which the copy or the plan breaks: text length, media, links, hashtags, daily caps and required flags. The evidence is the rule quoted with its date.

Will not land.

Email only. A failed send preflight check: sender authentication, one-click unsubscribe, where the list came from, suppression or how the message is classified. The evidence is the DNS record or header quoted. It reports a failed check and makes no prediction about any inbox.

Missing disclosure.

An affiliate, sponsorship, testimonial, AI-generated or regulated-claim disclosure that a platform or claim type calls for, or a quote with no recorded consent.

Each count is what the checks found on the day they ran. A zero means a check ran and found nothing. A type whose check did not run reads “not checked”, never 0. Each blocker’s re-check proves the change was made, such as “search the revised copy for the quoted words; expect no match”. It does not measure delivery, approval, reach or sales.

/claim-check

The full check on ad copy, an email or a post. It asks what you want to know, where the piece is going, who it reaches, whether it is part of a campaign, and whether it uses AI to make decisions about people or to generate content shown to customers. The preflight is filed in outputs/claim-check/.

/copy-check

The claims ledger and the slop pass alone, when you need to know which claims hold and whether the writing reads as machine-made.

/send-preflight

The email section alone: authentication, unsubscribe, list source, suppression and classification.

/launch-preflight

The campaign section alone, when the piece is part of a larger launch. It grants no permission to launch.

/claim-recheck

Run it after the piece has gone out, with the preflight file and what happened.

Lost? Type /bb, or /next for where the project stands.

Markdown files in your project folder, filed by date. This is the layout of a report. It contains no findings; those come from your copy.