The same prompt, different answers
Language models generate answers with some randomness, and AI search engines retrieve fresh sources for each query. Personalisation, location, model updates and changes to the index all add variation. Two people asking the same question at the same time can see different brands.
Why one screenshot isn't a measurement
A single answer tells you what happened once. If you check a prompt by hand on a good day you'll be over-confident; on a bad day you'll panic. Neither is a reliable basis for decisions.
Measure rates, not results
Treat each prompt as something to sample. Run it repeatedly and on every engine, then report the share of runs in which you appear.
- Run each prompt on a schedule, ideally daily.
- Report mention rate over a period such as 7 or 28 days, not a single run.
- Compare engines separately, because each uses different sources.
- Look at trends and step changes, and annotate them with your releases and content changes.
What you can control
You can't control the randomness, but you can control the evidence. Clear entity signals, accurate facts on your own site, strong third-party coverage and pages that answer the prompt directly raise the share of runs in which you're named.