Key takeaways
- Track a fixed set of buyer questions across the same assistants, on the same day each month.
- Four metrics: mention rate, average position, share of voice, and cited sources.
- Record the full answer text — it tells you why you were or weren't named.
Step 1 — Build the question set
Start with 30 questions a real buyer would ask an assistant. Draw them from three places: Search Console queries with commercial intent, your sales team (“what do prospects ask on the first call?”), and the assistants themselves (ask ChatGPT what questions people ask about your category). Group them into discovery, comparison and decision.
Write them the way a person would type them, with context: “Which agency should I hire for technical SEO for a Shopify store in the UK?” not “technical SEO agency”. Keep the set fixed for at least three months so you can compare.
Step 2 — Run the questions
On the same day each month, run every question through each assistant you care about — typically ChatGPT (with search enabled), Perplexity, Gemini and Google (noting whether an AI Overview appears). Use a clean session or a browser profile with no history so personalisation doesn't skew results.
For each run, record: the full answer text, every business named in order, and every cited source URL. This takes two to three hours for 30 questions across four assistants; it's worth it.
Step 3 — Calculate four metrics
| Metric | How to calculate | What it tells you |
|---|---|---|
| Mention rate | Questions where you're named ÷ total questions, per assistant | Your baseline visibility |
| Average position | Mean of your position in the list when named | Whether you're a first choice or an afterthought |
| Share of voice | Your mentions ÷ total mentions of you and named competitors | Competitive standing |
| Cited sources | Count of citations by domain, yours and others' | Which sources to target |
Step 4 — The spreadsheet layout
One sheet per month, one row per question × assistant. Columns:
Date | Stage | Question | Assistant | AI Overview shown (Y/N)
| Named? (Y/N) | Position | Businesses named (in order)
| Cited sources (URLs) | Answer textA summary tab pulls mention rate, average position and share of voice per assistant and per stage, with a month-on-month column. Keep the raw answer text — when a mention disappears, the text tells you whether the assistant changed its sources or your competitor did something.
Step 5 — Read it and act
- Low mention rate, you're not retrieved: check whether you rank for the comparison queries behind the questions. Fix ranking first.
- Retrieved but not named: your page isn't quotable or your entity is unclear. Restructure for specificity; fix consistency.
- Named but low position: competitors have stronger third-party presence. Target the cited sources.
- Named in Perplexity but not ChatGPT: your live presence is good but your training-layer footprint is thin. Build consistent mentions over time.