# Two launch checks: trackers before consent and AI claims you can prove (free sample)

A free sample from Don't Get Sued: a pre-launch legal-risk gate for AI-built apps and agents, by PaperChaseWebb. Works in Grok Bot, Claude Code, Codex and Hermes, and pastes into Claude or ChatGPT. Version 1.0, 2026-10-27.

**This is not legal advice.** It is a method for finding two common risks before launch. It names where a lawyer would usually start reading, not what any law requires of your product, and it never tells you that you are compliant.

Two of the fourteen areas in the full gate, chosen because they are on almost every AI-built site: an ad or analytics script that loads before the visitor makes any cookie choice, and marketing lines about AI that nobody can back up. Both are quick to find and cheap to fix before launch.


## The rules block

Paste this first, or put it in the file your harness reads first.

```text
LEGAL-RISK GATE RULES
You are a read-only reviewer. You read code, templates and the pages I give you.
You never edit files, deploy, publish, send anything or contact anyone.
You are not a lawyer. Never write "compliant", "legal", "safe" or "no risk".
Write "found", "not found", "unclear" or "cannot see", and quote the file and line
or the page text. Never invent a law, a penalty amount or a deadline.
```

## Check 1. Trackers that fire before consent

| | |
|---|---|
| What goes wrong | An ad pixel or session recorder loads on page open, and the cookie banner is decoration |
| Usual place to read | GDPR and the ePrivacy rules for EU and UK visitors; CCPA and CPRA in California; other state privacy laws |
| What to check | Scripts in the page head and tag manager that load before any choice; whether "reject" stops them |
| Usual fix | Load trackers only after a choice; make reject as easy as accept |
| Evidence to keep | Two network captures, before and after rejecting |

```text
List every analytics, ad, pixel and session-recording script and where it loads.
For each, does it load before the visitor makes a cookie choice? Does "reject" stop
it? Tell me the exact steps to capture network requests before and after rejecting,
and what I should see if the gate works. Quote file and line.
```

Quick search from the project folder:

```sh
grep -rn -i -E "fbevents|gtag\(|googletagmanager|fbq\(|hotjar|clarity" --include=*.html .
```

A hit in the page head with no consent check around it is a finding. The banner code is not proof; the network capture is.

## Check 2. AI claims you cannot prove

| | |
|---|---|
| What goes wrong | "AI-powered", "never makes a mistake", "saves 10 hours a week" with nothing on file to back it |
| Usual place to read | FTC rules on substantiation of advertising claims, which the FTC has applied to AI claims; state consumer protection laws |
| What to check | Every claim on pages, listings, ads and decks, and the evidence for each |
| Usual fix | Cut or soften claims with no evidence; say where a person is in the loop |
| Evidence to keep | A claim table with a source for every line that stays |

```text
Read [pages, listings, ads, deck]. List every claim about what the product does or
what users get. Sort each: describes a feature, describes a result, describes the
future. For each, name the evidence on file or write "none". Draft a plainer version
of every claim with no evidence. Never add a number we cannot prove.
```

Example rewrite (illustrative): "The AI-powered planner that never makes a mistake" becomes "A planner that drafts your week with AI. You approve every change."

## Record the decision

Every finding ends in one of four outcomes: pass with evidence, fix before launch, an exception signed by a named person with a date to look again, or counsel. Launch when nothing is an open fix or an unanswered counsel line.

## Where it runs

Claude Code: save the rules block as a skill or in CLAUDE.md and keep permission prompts on. Codex: put it in AGENTS.md and run with approvals on, read-only. Hermes: add it as a skill and run it in a chat session. Grok Bot: make it the bot's instructions, with no send or publish connectors. Claude or ChatGPT: paste the page text in.

## In the full gate

All fourteen areas (marketing texts and email, trackers, accessibility, kids' data, biometrics, data security, AI disclosure, AI claims, takedowns, policies that do not match the product, subscriptions and refunds, licenses, and agents that act for users), a scope interview, a data map from your code, 12 prompts, the gate register, a seeded test app with nine expected findings, harness notes, a failure table, a worked example and a counsel packet for your lawyer. $29 at paperchasewebb.ai.

Credit: the list of risks this product grew from is a 2026-09-06 Instagram reel by @murphmaxxing, rewritten in our own words.

Made by PaperChaseWebb at paperchasewebb.ai.