Define the evidence

For every important citation observation, save the exact prompt, service, date, source URL and relevant response context. A screenshot should show enough information to establish what was asked and which source was referenced. Avoid cropping away a condition that changes the meaning. Store the record in a structured folder or table so a reviewer can find the evidence behind a dashboard number without searching through informal messages.

Make it operational

Define a naming convention and access policy for the archive. Remove unnecessary personal data from test conversations before sharing evidence widely. Record whether the source link still opens and whether the response accurately represents the cited article. Keep corrections and invalidated observations rather than silently deleting them from the history. When presenting results externally, explain the test scope and avoid implying that the archive measures every user’s exposure. It is a documented set of observations, not a census of AI behaviour.

Imagine this

Hypothetical example

A team initially stores only attractive screenshots of source mentions. It changes the process to include prompts, dates, links and full relevant context. During review, one screenshot turns out to cite a different publication with a similar name and is marked invalid rather than counted.

Make it practical.

Use this checklist as you read. Checks are not stored.

References & further reading

Further reading: Google AI features and websites ↗

Published by Tap Tap Go. Product availability and eligibility vary by programme and market. About our stories.