Why it matters.
Research by SparkToro with Gumshoe.ai, reported by Search Engine Journal on 30 January 2026, ran 2,961 prompts across ChatGPT, Claude and Google’s AI in November and December 2025. Repeated runs of the same prompt returned the same brand list less than 1% of the time, and the same list in the same order less than 0.1% of the time.
The same research found that visibility percentage across many runs was more consistent than position. So a single screenshot, good or bad, tells you little. A count across runs tells you a lot.
The 10-minute self-check.
Take your main commercial question and do this now:
- Run it three times on ChatGPT in a temporary chat, so memory does not shape the answer.
- Run it three times on Perplexity, logged out.
- Write down every company named in each run.
- Count how many runs named you, and how many names appear in every run.
If the lists differ a lot, you have just seen why one run is not a verdict.
Steps.
- Decide a fixed number of runs per question per engine. Five is a practical minimum for a small team.
- Remove personalisation: temporary or logged-out sessions, no custom instructions, the same browser location each time.
- Record the date, engine, mode (search on or off), every name and every source for each run.
- Calculate the named rate: runs naming you divided by total runs, per question and engine.
- Compare month to month on the named rate. Ignore single-run swings.
This is the counting method behind our AI search visibility reports.
What we saw on prolinkage.com.
How to measure it.
- Named rate: per question, per engine. Monthly.
- Stable competitors: names that appear in more than half of runs. These are the firms the engines are confident about.
- Run log completeness: every run dated and stored, so anyone can recount.
Sources.
- Search Engine Journal, AI recommendations change with nearly every query (SparkToro and Gumshoe.ai research), 30 January 2026.
- Prolinkage AI self-test (ChatGPT, Perplexity, Google AI Overview, one run each), 6 November 2026.