---
name: ai-visibility-audit
description: Measure whether a brand or website appears in a defined sample of AI search answers and citations. Use for AI visibility, GEO visibility, brand mentions in ChatGPT or Google AI Mode, and answer-source comparisons using Toolmu DataForSEO.
---

# AI search visibility audit

## Before you start

Read [the shared research contract](../connect-toolmu/references/research-data.md) and [marketplace usage](../connect-toolmu/references/marketplace-usage.md). Use the existing Toolmu connection. The agent performs analysis; Toolmu supplies the evidence.

## Inputs

Brand name, website, market, competitors, and the buying questions customers ask. Specify engines and a bounded question set.

Ask only for information that changes the task. Accept ordinary links and names; resolve technical identifiers from verified results. Confirm missing market or scope before dependent paid calls.

## Tool selection

Inspect the current contract with Toolmu `get_tool`, then call `execute_tool` only for the steps that need data:

- `google-ai-mode-live-advanced-dataforseo`
- `chatgpt-llm-scraper-live-dataforseo`
- `gemini-llm-scraper-live-advanced`

The catalog's current input and credit ceiling override examples in this document. Do not fabricate parameters, silently use a different supplier, or treat a search snippet as full source content.

## Workflow

1. Build three neutral questions representing discovery, comparison, and purchase evaluation. Avoid putting the target brand in every question; record the exact wording.

2. Choose one engine for the first pass and a supported location/language. Keep question, engine, date, and market fixed when comparing brands.

3. Fetch each answer once. Preserve the complete returned answer, cited URLs, run ID, supplier task ID, and retrieval time.

4. Record separate states for brand mentioned, brand recommended, own domain cited, competitor cited, no citation returned, and failed collection. Check wording in context, not only occurrence of the brand name.

5. Calculate counts only over successful answers and show both numerator and denominator. Failed calls are unknown, not absent mentions.

6. Inspect cited pages to identify useful evidence or coverage the user's site lacks. Suggest specific content improvements with linked examples.

7. Repeat engines or dates only when the user requests a broader audit or has authorized that scope; do not repeatedly query until a favorable answer appears.

## Deliver

A question-by-engine evidence matrix with answer excerpts and source links, explicit sample counts, competitor observations, and prioritized content recommendations.

Include source links, actual market/date, returned sample size, missing coverage, run IDs, and confirmed credits. Let the user's agent choose the presentation; do not invoke an MCP App or a render tool.

## Boundaries

Results describe supplier-collected API samples, not every consumer session or all AI users. Answers vary by date, personalization, and engine. No visibility guarantee.

## Example request

> Check whether [brand] is mentioned for [three customer questions] in [market] and show the cited sources.