Home · Blog · Cricket markets Cricket

We let our AI CMO market itself. Here's the playbook it ran.

July 27, 2026 · Cricket AI

There's an old test for tool-makers: do you use the thing you sell? Cricket AI's marketing department is Cricket AI — the same agents, the same daily loop, the same honesty rules every customer gets. This is a walk through what the product prescribed for its own site, what we shipped because of it, and the uncomfortable lessons dogfooding forced on us.

Day zero: running the audit on ourselves

We pointed Cricket at cricketai.io like any user would. The audit did not flatter us. The findings read like a checklist of shoemaker's-children problems: our own site was thin on the structured data we score other sites on, our buyer questions had no dedicated pages, and the GEO panel called out gaps in exactly the machine-readability we preach. The first lesson arrived immediately: an honest auditor is uncomfortable, and that's the product working. A tool that flattered its own homepage would be lying to yours too.

What it prescribed — and what we shipped

• The site files. robots.txt welcoming crawlers, sitemap.xml that's actually current, and an llms.txt telling AI engines plainly what Cricket is, who it's for and where the key pages live. The same three files the GEO agent checks on every site it audits.
• Pages for buyer questions. The audit's coldest observation: people ask "what is GEO?" and "how do I know if ChatGPT recommends my product?" — and we had no retrievable page answering either. So we wrote them (the GEO guide among others), structured the way the product says citable pages should be: direct answers up top, structured data underneath, claims a model can quote.
• Structured data everywhere it's true. FAQ schema on the pages that answer questions, article schema on the blog, organization schema on about — machine-checkable facts instead of prose-only claims.
• The daily drafts. The X pair, the LinkedIn post, the community answers — drafted by the same agents customers use, approved by a human every time. The approve-and-ship loop we sell is the loop we actually run; when the approving felt heavier than it should, that was product feedback, and it changed the product.
• The citation scoreboard. Cricket asks the engines the questions our own buyers ask — "best AI marketing tool", "how to show up in ChatGPT recommendations" — and tracks whether we're cited and who's named instead. Every gap on that list is literally our content roadmap.

What dogfooding changed in the product

Running your own tool daily is a brutality no beta program matches. A sample of what it forced: drafts got anti-repetition memory because we caught our own LinkedIn posts opening with the same hook three days running. Reddit relevance got a full two-stage re-rank because we felt the sting of being offered an off-topic thread. Empty states got honest because "0 threads today — here's why" builds more trust than a padded list. And the daily fix became one prioritized fix with a paste-ready snippet because, as our own user, two vague ones was one too many. The feedback loop is the moat: every rough edge cuts us first.

The honest scoreboard

We won't dress this up with invented numbers — our own rule is that a metric you can't verify is worse than none. What we can say plainly: the work above is real and recent, the citation checks run on us like they run on customers, and the gaps they show are the pages we write next. When those numbers are mature enough to publish with a straight face, they'll appear here with their methodology. Until then, the process is the proof — and you can run the exact same audit on your own site in about a minute and see what it tells you.