Building a Repeatable Way to Measure AI Citation Rate

The question

I kept seeing citation rate numbers quoted in single screenshots, from a single day, for a single query. That's not a rate, it's an anecdote. I wanted a methodology that could actually produce a trustworthy number over time.

What "citation rate" needs to mean

A citation rate is only meaningful relative to a fixed query set, checked on a fixed cadence, over a defined window. Change any of those three and the number stops being comparable to last month's number.

The tracking approach

I defined a query set per topic cluster, large enough to smooth out day-to-day noise but small enough to check consistently. Each query gets checked against Google AI Overviews, ChatGPT, and Perplexity on the same cadence, and every citation (or non-citation) gets logged with a timestamp, not just tallied into a single percentage.

This is deliberately more work than a single screenshot. It's also the only way I've found to tell a real shift in citation behavior apart from ordinary day-to-day noise.

Where this is still incomplete

Gemini tracking isn't part of this methodology yet, mostly because its citation format has changed enough times that I don't yet trust a stable parsing approach for it. That's next on the list.

What this became

This is the tracking logic that Citeable is built on. The manual version described here is what I ran by hand before deciding it was worth turning into software.

References

1. Ongoing daily tracking, weekly snapshots. Methodology documented for replication, internal data.