AI visibility
Whether assistants mention and cite a brand, share of voice against competitors, and what the same prompt gets from four models.
Both tools cost credits and require a plan with AI visibility (Starter and above). Calling them on the Free plan fails with a plan-upgrade error.
get_brand_visibility
How often a brand is mentioned by ChatGPT and in Google's AI Overviews, its share of voice against named competitors, and which sources those answers cite.
| Argument | Type | Required | Default | Meaning |
|---|---|---|---|---|
projectId | string | yes | ||
query | string, 2 to 200 chars | yes | Brand, product, or domain to measure. Usually the project's own. | |
competitors | string[], up to 9 items | no | Competitors to compare against. Supplying these is what turns raw mention counts into a share-of-voice number. | |
locationCode | integer | no | project market | |
languageCode | string | no | project market |
Returns: total mentions, a share-of-voice table (brand, whether it is the target, mentions, share), and the 25 most-cited sources. Without competitors the share-of-voice table is omitted and the result says so.
ChatGPT visibility data is available for the United States and English only.
run_ai_prompt
Ask the same prompt of ChatGPT, Claude, Gemini, and Perplexity at once, and get each answer with its citations. Shows what an assistant actually tells someone asking about your category. Costs credits per model selected.
| Argument | Type | Required | Default | Meaning |
|---|---|---|---|---|
projectId | string | yes | ||
prompt | string, 3 to 4,000 chars | yes | The prompt, exactly as a user would type it. | |
models | array of chat_gpt, claude, gemini, perplexity; 1 to 4 items | yes | Which assistants to ask. Running several side by side is the point: they disagree, and the disagreement is the finding. | |
highlightBrand | string, up to 200 chars | no | Brand to flag wherever it appears in the answers. | |
webSearch | boolean | no | true | Let the models search the web before answering, which is closer to how real users hit them. |
webSearchCountryCode | two-letter country code | no | Localise the models' web search. |
Returns: one section per model with its answer, its citations (marked when they mention the highlighted brand), and its own status. A model that fails is reported as failed without hiding the others. Responses are cached for a week, so re-running the same prompt within that window is free.