How we log AI visibility: 10 questions × 4 engines, weekly
Starting September 2026, we log 10 buyer questions across 4 AI engines every week, 3 runs per question per engine — 120 sampled answers per week — and publish the method here.
Why we log
A single AI answer is one sample from a distribution. Ask ChatGPT the same question twice and you can get different brands cited. Any vendor showing you one screenshot as “proof” of visibility is showing you one coin flip.
So we do the boring version: the same fixed questions, asked fresh every week, on every engine, with the citations written down. Over weeks, that becomes a share-of-voice measurement you can actually act on.
The method
- Fix the question set. 10 questions a real buyer would ask an AI assistant. They are written once and never reworded mid-series, because rewording invalidates the trend.
- Fix the engines. ChatGPT (with web search), Perplexity, Gemini and Claude. Each engine retrieves differently — ChatGPT leans on Bing’s index and OpenAI’s own search crawler, Perplexity on its own crawl, Gemini on Google — so results are logged per engine, never blended into one “AI visibility” number.
- Ask fresh, log everything. Each question is asked in a clean session, 3 times per engine. We record: date, question, engine, whether the target brand was cited (yes/no), which domains were cited instead, and notes. How each engine’s answers are collected (API or app) is in the table below.
- Weekly cadence. Every week, same day. 10 questions × 4 engines × 3 runs = 120 logged answers per week.
What this does and does not tell you
- It does show which domains each engine repeatedly pulls from for your buyers’ questions — those domains are your target list.
- It does show movement after content or schema changes, engine by engine.
- It does not prove causation. If citations improve after a change, that is evidence, not proof. We say this to everyone we work with.
How to cite this
metabygaurav (2026). How we log AI visibility: 10 questions × 4 engines, weekly. First published 20 September 2026; updated 22 October 2026.
How we measure, per engine
| Engine | How answers are collected | Country and login | Runs | Cadence |
|---|---|---|---|---|
| ChatGPT | OpenAI API with web search, plus manual checks in the ChatGPT app | App checks: India, logged out | 3 per question | Weekly |
| Perplexity | Sonar API | API call, no login | 3 per question | Weekly |
| Gemini | Gemini app, checked by hand | — | 3 per question | Weekly |
| Claude | Claude app, checked by hand | — | 3 per question | Weekly |
| Google AI Overviews (paid audits) | Google Search in a browser | India, logged out | 3 checks per question, on 2 different days | Once per audit |
The free check covers ChatGPT, Perplexity, Gemini and Claude. Paid audits cover ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. Answers fetched through an API can differ from what people see in the apps, so we record the access method for every run.