How to Track Changes in ChatGPT Citations Over Time

Correction — 31 October 2026: the earlier version included unsupported research or outcome claims without supporting records. Those findings and dataset promises are withdrawn. This is practical guidance, not a verified original study or client result.

Keep your prompt panel stable

Select relevant questions before testing and keep their wording and language unchanged for comparisons. Separate brand-led questions from generic discovery prompts. Record additions as a new panel version instead of silently changing the denominator.

Record the conditions of every run

Log date, assistant, model label when available, search mode, location context and fresh versus existing conversation. If a setting cannot be controlled, disclose it. Do not treat runs across different models or modes as identical experiments.

Save sources rather than a single score

Preserve answer text and exact source URLs for every completed run. Distinguish your own site, other domains and mentions without links. Mark errors and no-source answers explicitly. Do not overwrite last week’s evidence with the latest result.

Compare sets with a declared rule

For each matched prompt, compare distinct cited URLs or domains using the same unit. One optional descriptive measure is intersection divided by union of the source sets; declare how empty sets are handled. Show counts as well as percentages and keep failed runs out of successful-run totals.

Investigate before changing a page

A changed source does not by itself prove that a page lost authority or needs rewriting. Check page availability, material changes and observed answer accuracy. Update content when it has a real gap, not simply to reset a modification date.

Report uncertainty

Show the observation window, panel size, settings and variation. This workflow provides no verified universal overlap rate, decay rate or expected time to regain a citation. Referral visits and enquiries require separate measurement.

Sources and further reading

Google guidance on AI features