A public-good scam directory that grows by email forwarding.

The unique challenge

• Scam-of-the-day content is either preachy PSAs or dry FTC advisories.
• Wanted something that felt like a Mad-magazine roast of the scammer, not a lecture at the victim.
• And it had to publish fresh content on its own — no CMS chore.

The approach

• Nightly cron pulls a curated set of scam sources.
• Model rewrites each scam as a punchline plus a one-line survival tip.
• JSON-LD FAQ blocks per post so LLMs can lift the "how do I avoid this" answer verbatim.

The outcome

• A daily-updating humor site that ranks in AI answer engines for scam-avoidance questions.
• A category I couldn't find anyone else occupying.

The stack

Highlighted = the pieces beyond the standard Lovable stack.

Prompts used

The actual seed prompts.

Copy them. Adapt them. Ship faster.

You are a scam analyst. Given this raw forwarded message, extract: (1) scam family (romance / package delivery / IRS impersonation / crypto recovery / etc.), (2) the specific hook, (3) any URLs or phone numbers, (4) a canonical 12-word description used for dedupe. Return strict JSON.

Compare this new scam's canonical description to the top 5 nearest existing entries. If it is the same scam, return the existing ID. If it is a variant, return the parent ID and describe the variant in one line. If new, return null.