AI visibility
What to Save From an Answer So You Can Check It Again
A repeatable check needs more than the answer text. Save the exact question, the engine, the date, how you collected it, and the measurement status.
Say you asked ChatGPT a question about your industry and saw a competitor cited. Two weeks later you want to check whether that citation is still there. If all you saved was the answer text, you can reconstruct almost nothing. The exact question is gone, along with the engine, the date, how you collected the answer, and even whether that measurement existed at all. Rechecking an answer means keeping an observation record: a minimal set of fields that makes the observation repeatable and comparable. What follows is a working format from RankAI, not a universal standard.
What the observation record needs to hold
A useful record is not a folder of screenshots. It is a set of fields that let you say later, “this is the same observation” or “this is something else.” We propose six minimum fields.
1. The question, word for word
Not a paraphrase, not the topic. The exact string sent to the engine, with any variations noted separately. A reworded question produces a different answer, and the comparison falls apart.
2. The engine and how you collected it
RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. Claude and Grok are measured only after you connect them. Record the exact engine and, where relevant, whether you collected through an API. What comes back from an engine through an API is the API response, not the screen a person sees. Those are two different observations and should not be mixed in the same record.
3. The date and time
Without a timestamp, two identical answers prove nothing about stability, and two different answers prove nothing about change.
4. The status of the measurement
When a visibility figure is unavailable, RankAI shows it as N/A, never as zero. The record has to carry the same status. “Not available” is information; “zero” is a different claim, and often a false one.
5. The full answer text
Kept as it is, with no summary. Brand citations should be pulled from the text, not rebuilt from memory.
6. The verification note
Who collected the data, with which tool, and any anomaly worth flagging. This is the field that lets someone else recheck the record.
The point that often gets lost is the difference between the content of an observation and its conditions. The content is the answer: the text, the citations, the brand names mentioned. The conditions are everything that turns that answer into data rather than an anecdote: the exact question, the engine, the date, the collection method, the measurement status. Save only the content and, on recheck, you can say “this sentence was there before.” Save the conditions too and you can say “this sentence appeared on this engine, for this question, on this date, collected this way.” Only the second statement is verifiable. That is why the observation record is not bureaucratic paperwork: it is the minimum needed to avoid confusing a change in the engine with a change in how we are looking. One more thing: keep separate records per engine. Gemini and Google AI Overviews are not the same observation, and a record that merges them makes it impossible to tell which one changed. Finally, treat the status field as part of the data. A record marked N/A documents an absence; a record marked zero documents a result you probably did not measure.
A filled-in example
Question: “which tools measure brand citations on ChatGPT.” Engine: ChatGPT. Date: the day and time of collection. Method: API, with a note that this is the API response and not a person's screen. Status: available or N/A. Text: full. Verification note: operator and tool. Ask the same question of Perplexity and you need a different record: different engine, different observation.
The limit to keep in mind
The observation record preserves what was collected. It does not guarantee the engine will answer the same way next time. An API response is not a person's screen, and a missing measurement stays N/A. The record is for rechecking, not for predicting. For the full picture of how RankAI handles these observations, see the in-depth page.
Keep your prompt set fixed to compare AI visibility over time