Lettertrace

Search rankings used to be the whole game. Increasingly the answer a buyer sees is written by an assistant, and whether your product appears in it — and how it is characterised — is invisible to every traditional SEO tool. Lettertrace measures that surface directly.

What it tracks

You give it topics rather than exact prompts, and it expands each one into multiple natural phrasings automatically, because a single hand-written question tells you very little about a probabilistic system. It then runs those prompts across Claude, ChatGPT, Gemini and Google AI Overviews and records what comes back: whether you were mentioned, the sentiment of the mention, and your share of voice against named competitors. Scheduled daily or weekly runs turn those readings into a trend line rather than a one-off snapshot.

Bring your own key

Rather than reselling model access, Lettertrace runs on your own API credentials, encrypted at rest. That keeps the cost transparent — you pay the model providers directly at their rates — and means the tool has no incentive to limit how often you sample.

Open source

The whole project is MIT licensed and on GitHub. You can clone it, point it at your own Supabase database and run it locally, which matters if the prompt set or the competitor list is something you would rather not host with a vendor. The hosted version is free.

Who it is for

AEO and GEO teams, and anyone doing the work of understanding why an assistant recommends a competitor. Built by the Letter Company.