Prompts, keywords, and citations

Test an AI-search prompt and inspect brand mentions, model responses, related queries, and cited sources.

The AI visibility detail sections help you learn which questions surface your brand and which sources support the available answers.

Conditional: Prompt Explorer and its models can be plan- or provider-gated. Running a prompt requests new model output and may use paid capacity; review the current workspace notice before running it.

Entry path

Open a store project → SEO → AI visibility → Prompts, Keywords, or Citations.

Before you start

Complete AI-visibility setup.
Write one focused shopper question.
Select at least one available model.
Use web search only when you want the model to consult current public sources.

Controls and click results

Control	What happens after you select it
Prompt field	Accepts the question to test and warns when the input is too long.
Model selector	Chooses which available model responses to request.
Allow web search	Lets supported models consult public web sources for this run.
Brand highlight	Tells the result view which brand mention to call out. It does not force the model to mention the brand.
Run Prompt Explorer	Starts the selected model requests. It does not edit the store or save a visibility improvement.
Prompts	Shows tracked or explored questions and their available results.
Keywords	Shows terms associated with the AI-visibility dataset.
Citations	Shows sources cited in available model answers.
Mentions	Shows responses or records where the tracked brand was detected.

Explore one shopper question

Open Prompts.
Enter one natural-language question a shopper might ask.
Select the available model or models you need.
Decide whether to enable Allow web search.
Optionally choose the tracked brand for highlighting.
Select Run Prompt Explorer once.
Wait for each requested model to finish or fail.
Compare the response, Brand mentioned or Brand not mentioned state, citations, and related queries.
Open Keywords and Citations to find recurring topics and sources.

Interpret results safely

One answer is a sample, not a stable ranking.
“Brand not mentioned” describes that response, not every possible answer.
A citation means the model referenced a source; it does not mean Runner endorses the source.
Web-search-off and web-search-on results answer different questions and should not be treated as identical tests.

Troubleshooting

State	Meaning and next step
Run button disabled	Enter a valid prompt and select at least one model.
One model fails	Use the completed model results and retry the failed provider later.
No citations	The answer may not have used or returned public sources. Retry with web search only if that matches your goal.
Unrelated results	Make the prompt more specific about the product, audience, or buying need.
Provider access error	Choose an available model or ask a workspace owner about access.