Direct Answer
To audit what AI says about your business, do not start by counting mentions. Start with accuracy. Write down the facts a buyer must understand, ask the same six to ten questions in search-enabled ChatGPT, Google, and Bing experiences, and record the answer, cited sources, omissions, and errors. Then fix the source layer that produced the problem.
A useful audit separates five conditions: accurate, incomplete, unsupported, wrong, and absent. That distinction matters. An absent brand may need stronger relevance and third-party proof. A wrong brand description needs factual reconciliation. An accurate mention with no click may be doing its job at the discovery stage.
Source Note
OpenAI's ChatGPT Search documentation, updated on 13 July 2026 when checked for this article, says search may rewrite a prompt into multiple targeted queries and may show inline citations or a Sources panel. It also says there is no way to guarantee top placement. Google explains that AI Mode and AI Overviews can use query fan-out, may use different models and techniques, and do not show the same links every time.
Bing's February 2026 AI Performance preview documents citations, cited pages, grounding queries, and visibility trends. A 30 June TechRadar guide recommends fixed prompt logs and trend tracking, but some product-reporting details in that article conflict with Google's current documentation. This audit therefore uses official platform documentation for platform facts and treats manual prompt tracking as a practical method, not an official ranking report.
Link Map
| Resource | Date checked | What it supports |
|---|---|---|
| OpenAI: ChatGPT Search | 13 Jul 2026 | Query rewriting, citations, source panels, crawl eligibility, and no guaranteed placement. |
| Google: AI features and your website | 13 Jul 2026 | Query fan-out, variable links, SEO fundamentals, and eligibility requirements. |
| Bing: AI Performance preview | 10 Feb 2026 | First-party citation, cited-page, grounding-query, and trend metrics. |
| TechRadar: Track brand visibility | 30 Jun 2026 | Manual prompt logs, scheduled checks, and separating mentions from clicks and crawl access. |
1. Prepare a Brand Truth Sheet
Keep the sheet short enough to verify in minutes. Include the official brand name and aliases, founder or owner, location and service area, primary services, ideal clients, proof links, price or availability boundaries, and the preferred next action. Add a source URL beside every claim. If the business itself cannot agree on these facts, an answer engine has no clean truth to retrieve.
- Identity: exact name, spelling, owner, location, and official domain.
- Offer: what is sold, who it is for, and what is explicitly not offered.
- Proof: case studies, credentials, products, reviews, and independent references.
- Decision facts: geography, pricing model, timelines, availability, and contact route.
2. Use a Fixed Buyer-Question Set
Run the same questions with search enabled where the product offers it. Do not use only the brand name. Test the questions that lead to discovery, comparison, verification, and action.
- What is [brand], and who runs it?
- What services does [brand] provide?
- Is [brand] suitable for [specific buyer and problem]?
- What proof shows that [brand] can do this work?
- How is [brand] different from common alternatives?
- How can I contact or hire [brand]?
Add two unbranded questions that a real prospect might ask, such as "Who builds practical AI research systems for a small consultancy?" This tests category discovery without assuming the buyer already knows the name.
3. Score the Answer, Not the Vibe
| Dimension | 0 | 1 | 2 |
|---|---|---|---|
| Identity | Wrong or confused | Partly correct | Correct and specific |
| Offer | Missing | Generic | Accurate and buyer-relevant |
| Proof | Unsupported | Self-claimed only | Traceable evidence |
| Source quality | No usable source | Weak or stale source | Canonical and credible sources |
| Next step | Wrong route | Vague | Clear, current action |
A score is useful only when the evidence stays attached. Save the prompt, date, product, search state, answer, citations, and a short note explaining the defect. A screenshot without those conditions is not a repeatable audit record.
4. Trace the Source Path
Open every cited source. Check whether it actually supports the answer. Then search the exact incorrect phrase and compare your homepage, service pages, profiles, directories, review sites, partner pages, and older articles. The goal is to locate the conflict that a retrieval system could reasonably encounter.
If there is no citation, treat the answer as a clue rather than evidence. OpenAI explicitly warns that models can still be wrong. Important identity, legal, financial, medical, pricing, and availability claims should always be verified against the source itself.
5. Fix the Source, Not the Generated Answer
- Correct the canonical page that owns the fact.
- Align high-value profiles, directories, and partner references.
- Add proof close to the claim instead of hiding it in a distant PDF.
- Use descriptive internal links so supporting pages are easy to discover.
- Make the important fact visible in text, not only inside an image or video.
- Request normal recrawling through the platform tools you already use, then wait before judging the result.
6. Repeat Without Chasing Noise
Run the audit monthly at first and after material changes to the offer, team, pricing, location, or proof. Keep the prompt set stable. Add a new prompt only when buyer behavior changes, not because one answer was disappointing. Compare four-week direction: fewer factual conflicts, stronger sources, better category relevance, and more accurate next steps.
Sources
- OpenAI Help Center: ChatGPT Search (updated 13 Jul 2026 when checked)
- Google Search Central: AI features and your website (checked 13 Jul 2026)
- Bing Webmaster Blog: Introducing AI Performance (10 Feb 2026)
- TechRadar Pro: How to track your brand's visibility in AI search results (30 Jun 2026)