By Milod · 5 August 2026 · 1 min read · Measurement

Measuring citations across four assistants

Rank tracking doesn't tell you whether ChatGPT quotes you. Here's a workable way to measure it.

There is no Search Console for assistants. No impressions, no average position, no export. If you want to know whether ChatGPT, Claude, Perplexity or Gemini quote you, you have to ask them, repeatedly, and record what comes back.

Track prompts, not keywords

The unit is the question a buyer would type: "best running shoes for beginners", not the keyword "running shoes". Write 20 to 50 of them, in the words customers use, covering the whole journey from problem to comparison.

Sample on a schedule

Answers vary between runs, so a single check tells you very little. Ask the same prompts weekly and record whether your domain appeared, in which position among the sources, and which page was cited. The trend is the signal; any one answer is noise.

What to record

• Cited or not, per prompt, per assistant.
• Which URL was cited, so you learn which pages are quotable.
• Which competitors appear beside you, and how often.
• The date, because model updates move these numbers without warning.

What to do with it

Pages that get cited tell you the format that works for your subject; write more like them. Prompts where a competitor is cited and you aren't are the clearest brief you will ever get: their page answers something yours doesn't.

Expect volatility. A prompt you own one week can vanish the next because the model changed, not because your page did. Judge over months.