How do you build a prompt basket?
Category: 10 questions
"Best [category] providers for [industry]"
Comparison: 8 questions
"[Vendor A] vs [Vendor B] for [use case]"
How-to: 8 questions
"How to choose a [service] supplier"
Proof: 8 questions
"Is [vendor] reliable for [use case]?"
Brand: 6 questions
"What does [company] do?"
Source the questions from sales calls, enquiry forms and the "people also ask" boxes on Google, not from your own marketing language.
How do you run it?
- Use identical wording in every tool and every run.
- Run each prompt three times per tool, across ChatGPT, Gemini, Perplexity and Claude. That is 480 answers per cycle.
- Log the tool, model version, date and whether web search was on.
- Repeat monthly on a fixed date.
How do you score the answers?
For each answer record four things: mentioned or not, cited or not, described accurately or not, and position. Combine them into the Growth Torque AI Visibility Index: AVI = 0.40 × mention rate + 0.25 × citation rate + 0.20 × accuracy + 0.15 × position score. Position score is 100 if you are named first, 70 if second or third and 40 if later.
What does a simple tracking sheet look like?
| Prompt | Tool | Run | Mentioned | Cited | Accurate | Position |
|---|---|---|---|---|---|---|
| [Prompt 1] | ChatGPT | 1 | Y / N | Y / N | Y / N | 1, 2-3, later |
| [Prompt 1] | ChatGPT | 2 | Y / N | Y / N | Y / N | 1, 2-3, later |
| [Prompt 1] | Gemini | 1 | Y / N | Y / N | Y / N | 1, 2-3, later |
Which mistakes spoil a prompt basket?
- Changing the wording between months, which breaks the trend.
- Testing only brand questions, which always flatter you.
- Judging from one run.
- Not recording whether web search was on.
Frequently asked questions
How many prompts do I need?
Thirty to fifty is practical. Fewer than thirty gives noisy results, and more than fifty is hard to maintain by hand.
Can I automate the runs?
Partly. Tools and APIs can run prompts, but you should still review a sample by hand because scoring accuracy needs judgment.
How often should I run it?
Monthly is enough for most B2B companies. AI answers change often, but a monthly trend smooths the noise.
Sources and further reading
Next step
Skip the spreadsheet: our free audit runs a 40-question basket for you and compares you with three competitors.
