Home / Resources / AI answer spot-check
TemplateAI answer spot-check template
Once a month, ask the AI assistants your customers use the same fixed set of questions, and write down whether you're mentioned and which sources they cite. One run proves little; the trend over several months is what's useful.
Why a manual check, and its limits
- Published studies of which sites AI assistants cite use different prompts and tools, and one of them (Semrush) saw citation shares swing by tens of percentage points within weeks. So a single result is not a trend.
- Answers may differ by country, signed-in state, wording and day. Keep conditions as constant as you can and record them.
- This tells you what's being said. It can't tell you why, and it doesn't prove any change you made caused a change.
Step 1: write your fixed prompt set (8 to 12)
Use the same wording every month. Swap the words in brackets for yours.
- Local: "best vegan restaurants in [your city]"
- Local, specific: "vegan [dish or product type] near [your neighbourhood]"
- Product: "is [your product] vegan?"
- Category: "best vegan [your product type] in [your country]"
- Comparison: "[your business] vs [a competitor you know]"
- Reputation: "what do people say about [your business]?"
- Service businesses: "who can help with [your service] for vegans?"
Step 2: run them, the same way each time
- Pick the assistants you care about (for example ChatGPT, Perplexity, and Google's AI answers).
- Use a fresh chat each time, ideally in a private window.
- Note the date, the assistant, and whether web search or browsing was on.
Step 3: log what you see
| Date | Assistant | Exact prompt | Are you mentioned? (yes / no / wrong) | Sources or links shown | Is what it says about you accurate? | Notes |
|---|---|---|---|---|---|---|
Step 4: act on what you find
Sources (accessed 2026-10-21):
- Semrush: most-cited domains in AI search (2025-11-10)
- OpenAI: crawlers and bots
- Google Search Central: optimizing for generative AI features (2026-07-10)
Want help interpreting your results? See the free program.