Because LLM answers are non-deterministic, the same prompt run three times can return your brand in first place, third place, and not at all. Monitoring brand mentions properly means sampling each prompt repeatedly and on a fixed schedule, not checking once and treating the result as ground truth - manual spreadsheets work at small scale, and this is exactly the workflow AskRank automates once your prompt set outgrows one.
Why it matters for AEO. Manually checking "does the AI mention us" a handful of times gives you a snapshot with high variance - LLM answers are non-deterministic, so a real read needs repeated sampling across the actual prompts your buyers ask, on a schedule.
How to do it
- Write down 10-50 real buyer prompts covering the questions and comparisons your buyers actually ask, not just branded searches.
- Run each prompt at least 3 times per AI assistant you care about (ChatGPT, Claude, Perplexity, Gemini) - a single sample per prompt is close to meaningless for a trend.
- Record whether your brand appeared, at what position, and with what sentiment, for every run - a spreadsheet works up to roughly 10-20 prompts.
- Repeat on a fixed cadence (weekly at minimum) rather than only when you remember to check - visibility drifts even when you have not changed anything.
How to verify. After 2-3 weeks of consistent sampling, you should have enough runs per prompt to compute a stable mention rate, not just a single yes/no.
Example
10-50 prompts x 3+ samples x 4 assistants, logged with:
prompt, date, mentioned (y/n), position, sentiment, cited URL.