AI SEARCH (GEO)

Build a prompt set for brand monitoring

Use when you need a repeatable list of questions to track how assistants describe your category.

prompt-set-brand-monitoring.md

Download .md

You are designing a monitoring prompt set for tracking how AI assistants answer questions in a category.

Brand: {{BRAND_NAME}}
Category: {{CATEGORY}}
Core use cases: {{USE_CASES}}
Known competitors: {{COMPETITORS}}

Produce 40 prompts a real buyer might type, grouped into these five buckets, 8 per bucket:
- Unbranded problem prompts (no product named)
- Category and shortlist prompts ("best X for Y")
- Direct brand prompts
- Comparison prompts (us vs a named competitor)
- Objection and risk prompts (pricing, migration, reliability, alternatives to us)

Format as a table:
| # | Bucket | Prompt text | What a good answer for us looks like | What would count as a bad outcome |

Then list the 8 prompts you would track weekly if you could only track 8, with one line each on why.

Constraints:
- Prompts must be phrased as a buyer would type them, including sloppy phrasing. Do not write keyword strings.
- Do not include prompts that only our brand could plausibly answer.
- Do not assume our product has features not listed in {{USE_CASES}}.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{BRAND_NAME}}
{{CATEGORY}}
{{USE_CASES}}
{{COMPETITORS}}

Getting a better result

Keep the wording frozen once you start tracking; changing a prompt resets your baseline.
The unbranded bucket matters most early, since that is where you are absent rather than misdescribed.
Log the date and the assistant version alongside every answer you capture.

Questions about this prompt

When do I need a prompt set rather than the one-off brand test?

When you have decided to track this over months rather than look once. The brand description test gives a snapshot of one question. This builds the fixed list you will rerun, which is what turns anecdotes about what an assistant said last Tuesday into something comparable across periods.

What do I need before running it?

Your real use cases and the competitors buyers actually name, not the ones on your battlecard. The prompt is told not to assume features beyond {{USE_CASES}}, so an inflated list produces prompts nobody would type. Examples of how buyers phrase things, including the untidy phrasing, improve the output noticeably.

What comes back, and which column earns its place?

Forty prompts across five buckets with, for each, what a good answer for you looks like and what would count as a bad outcome, plus eight to track weekly. The bad outcome column is the useful bit. It settles what you are measuring before any results arrive to argue about.

What is the mistake to avoid?

Improving the wording of a prompt after tracking has started. A reworded prompt is a new prompt and the baseline it had is gone. Freeze all forty, including the awkward ones, and log the date and assistant version with every capture. If one is genuinely wrong, add a replacement alongside it.