Visibility measurement

How to Write Prompts for Visibility Tests

Use relevant requirements, consistent wording, repeated runs, and recorded settings.

A useful visibility test starts with a clear question and a repeatable method. Prompt wording, buyer requirements, model settings, and search use can all change the answer.

Include the requirements that matter

Write prompts from real customer questions. Add context when it affects which products or services qualify:

• The task: what the customer needs to do.
• The buyer: relevant role, company size, or experience.
• Requirements: price range, timeline, integrations, or contract terms.
• Location: where the service or product must be available.
• Exclusions: conditions that would rule out an option.

A short category prompt can still be useful. It measures a broad question. Keep broad prompts separate from prompts that describe a specific buyer so the report makes that difference clear.

Keep a fixed set for comparisons

Save the exact prompt text. When wording changes, record a new version. A change from “cheap” to “affordable” introduces another variable into the comparison.

Use a fixed set to measure changes over time. Maintain a separate set for new questions and prompt experiments.

Repeat each prompt

One answer records one outcome. Repeat the prompt to measure how often each brand appears. Report both the count and the number of completed runs, such as “mentioned in 7 of 10 answers.”

Ten runs can be a practical starting point. Increase the sample when differences are small or answers vary widely. Do not describe a small sample as a precise measure of all customer experiences.

Record settings and search use

Keep model version, available tools, location, conversation history, and supported generation settings consistent where possible. Record what you cannot control.

Test search-enabled and search-disabled conditions separately. Where tool logs are available, record whether search actually ran. Making a search tool available does not prove it was used.

Choose a useful schedule

Set the schedule according to the decisions the report supports. Weekly or monthly comparisons may be sufficient for ongoing content work. A launch or a rapidly changing category may justify more frequent checks.

Compare repeated batches before attributing a change to marketing work. A model update, a changed source, or ordinary response variation can also affect the result.

Report the observed outcomes

• How often each brand is mentioned or recommended.
• Which customer requirements change those results.
• Which competitors appear in the same answers.
• Whether the descriptions of your business are accurate.
• Which sources are cited or recorded in available tool logs.

Keep citations, opened pages, and returned search results as separate fields. Each records a different operation.