AI visibility audit: how to check whether AI mentions your brand

What is an AI visibility audit

An AI visibility audit checks how your brand, product, category or experts are represented in answers from AI systems: ChatGPT Search, Perplexity, Google AI Overviews / AI Mode, Bing Copilot, Gemini, Claude and other answer engines.

The audit answers five questions:

1. 1Does AI mention your brand at all?
2. 2In which queries does it mention you?
3. 3Does AI cite your site or third-party sources about you?
4. 4How exactly does it describe the brand — is the description accurate, neutral, positive, outdated or superficial?
5. 5Who does AI mention instead of you — and why?

SEO, GEO, AI Search visibility: what’s the difference

The term GEO became popular after the “Generative Engine Optimization” paper, where the authors described a new task: optimising content not only for SERP rankings, but for visibility inside the answers of generative systems.

But GEO shouldn’t be treated as a set of “hacks”. Google is explicit that basic SEO practices still apply to AI features in Search: the page must be crawlable, indexed, snippet-eligible, useful for people, and structured data must match the visible content.

Mention, citation, grounding and discoverability

In an AI visibility audit it’s important not to conflate several different signals.

What matters for a business isn’t just any mention, but a mention in the right context: when the user is already comparing solutions or asking who to trust.

When a business needs an AI visibility audit

An AI visibility audit is worth running if:

• customers in your niche often ask AI “what should we choose”, “which vendors should we compare”, “what are the best tools”;
• the brand has SEO traffic but doesn’t appear in ChatGPT, Perplexity or AI Overviews;
• AI mentions competitors but not you;
• AI describes the brand in an outdated or inaccurate way;
• you invest in PR, content, reviews and ratings but don’t know whether any of it affects AI answers;
• the CMO wants brand visibility not only in Google, but across the new answer surfaces.

This is not a hypothetical shift. Gartner predicted that traditional search volume could drop by 25% by 2026 because of AI chatbots and virtual agents. Pew Research also found in 2025 that users clicked traditional results less often in sessions where Google showed an AI summary.

That doesn’t mean SEO is “dead”. It means part of the decision-making now happens in the answer layer, not only in the click layer.

What to prepare before the audit

Before opening ChatGPT or Perplexity, assemble a basic map of the brand. Otherwise the audit quickly turns into a pile of random screenshots.

1. Brand entity map

Put these in a table: the official name, spelling variants, domain, product names, categories, key experts, geography, languages, competitors, main use cases and target segments. For agentMention.ai it could look like this:

2. The AI platforms to check

For a B2B/SaaS/tech brand, the minimum set is: ChatGPT Search; Perplexity; Google AI Overviews / AI Mode (where available); Bing / Copilot; Gemini; Claude (if your audience actually uses it).

ChatGPT Search matters because OpenAI describes it as a way to get current answers with links to sources. Bing is moving towards measuring AI presence too: Microsoft shipped AI Performance in Bing Webmaster Tools, which shows citations, cited pages, grounding queries and visibility trends.

3. The context of the check

Record: the query language; the country or region; logged-in or logged-out mode; date and time; the model or platform; whether web search was on; the exact prompt; the answer; citations / sources; the competitors mentioned.

How to build a prompt set for the audit

The most common mistake is checking a single branded prompt: “What is [brand]?”. That’s useful, but it doesn’t show whether the brand makes it into real buyer journeys. Instead build a prompt set that covers 5 query types.

1. Branded prompts

They show whether AI knows the brand and describes it correctly. For example: “What is agentMention.ai?”, “What do people use agentMention.ai for?”, “Explain agentMention.ai to a CMO”. Watch for: whether AI mixes your brand up with another one, whether it gets the category right, whether it mentions current features, whether it links to the official site, whether it invents facts.

2. Category / unbranded prompts

They show whether your brand appears when the user doesn’t know its name yet. For example: “Which tools help track brand mentions in ChatGPT?”, “Best AI visibility monitoring tools for B2B SaaS”. Watch for: whether your brand makes the list, in which position, next to which competitors, and whether there’s a citation.

3. Comparative prompts

They show how AI compares you with competitors. For example: “Compare [brand] and [competitor] for AI visibility monitoring”, “Alternatives to [competitor] for GEO analytics”. Watch for: whether your brand is in the comparison, whether the criteria are correct, whether AI overstates the competitor’s advantages, whether sources are given.

4. Problem-based prompts

They show whether AI recommends your brand as a solution to a specific pain point. For example: “AI recommends competitors but not our brand. What should we do?”, “How do we measure brand visibility in AI answers?”. Watch for: whether AI names your product category, whether it recommends tools, whether it mentions your brand.

5. Local / market prompts

For a local market this is critical. An English prompt can paint one picture and a local-language prompt a completely different one. For example: “Which AI visibility tools are available to Ukrainian companies?”, “How can a Ukrainian B2B SaaS check brand mentions in ChatGPT?”. Watch for: whether AI understands the local context, whether it cites local sources, whether it silently replaces the local market with global examples.

The step-by-step audit method

Step 1. Build the audit sheet

The minimum columns:

Step 2. Run the prompt set across several platforms

Don’t treat ChatGPT and Perplexity as the same system — their search, citation and source-selection logic differs.

• Google AI Overviews / AI Mode depend on indexing, snippet eligibility, preview controls and Google’s own search systems.
• ChatGPT Search can use search partners and rewrite the prompt into more targeted queries.
• Perplexity has its own crawler rules: PerplexityBot for indexing and surfacing, Perplexity-User for user-initiated actions.

That’s why visibility has to be counted separately for each platform.

Step 3. Separate mentions from citations

The brand can be: not mentioned at all; mentioned without a link; mentioned with a link to your site; mentioned with a link to a third-party source; recommended as one of the best options. For a CMO these are different levels of value: a mention without a citation builds awareness, a citation to the site can drive traffic, a recommendation in a BOFU query affects pipeline.

Step 4. Score the quality of the mention

Not every mention is useful. An example scoring model:

Step 5. Check the sources

For every citation ask: is it your site or a third-party source? Is the source current? Does the page really talk about your brand? Did AI use the correct information? Does the source give competitors more prominence than you?

Vendor research shows that different systems prefer different sources. Yext’s analysis of 6.8 million citations found that Gemini cited brand-owned websites more often, ChatGPT leaned more on third-party sites and directories, and Perplexity relied more on niche/industry sources. It’s vendor research, but it makes the point well: your own blog alone isn’t enough.

Step 6. Compare competitors

For every prompt count not only your own presence, but also: who is mentioned instead of you; which competitors have citations; which sources back them up; which topics they cover better; where they have reviews, ratings and guest articles.

How to read the result: if you weren’t mentioned but neither were competitors, the prompt may simply be weak. If competitors are mentioned regularly and you aren’t — that’s a real gap.

Step 7. Check technical access

Not all AI bots are the same. OpenAI separates OAI-SearchBot (for appearing in ChatGPT search), GPTBot (for training crawls) and ChatGPT-User (for user-initiated actions). Google separates Googlebot from Google-Extended, which controls whether your content is used to train Gemini but doesn’t affect ranking in Search.

So don’t just put “open the site to all AI bots” in the recommendations. A better sequence:

1. 1check robots.txt and WAF/CDN blocking;
2. 2separate search/grounding crawlers from training crawlers;
3. 3check server logs;
4. 4allow only the user agents you need, and only where that matches company policy;
5. 5don’t block Googlebot if you need visibility in Google Search;
6. 6document the decision together with the SEO, legal and security teams.

Which metrics to track

For Bing, part of this logic is already formalised in AI Performance: total citations, average cited pages, grounding queries, page-level citation activity and visibility trends. In Google Search Console, a dedicated AI Overviews / AI Mode breakdown isn’t always available, so a full audit still needs your own prompt sets.

How to work out why AI doesn’t mention your brand

If your brand doesn’t show up in AI answers, the reason isn’t always “bad SEO”. A diagnostic table:

How to improve AI visibility after the audit

1. Make pages clear for people and machines

An AI-friendly page is not a page “for a robot”. It’s a page that clearly states: who you are, which category you represent, who the product is for, which use cases it covers, how it differs, which facts support that and when the information was updated. Google emphasises unique useful content, page experience and crawlability — not “AI hacks”.

2. Add citation-friendly blocks

AI finds it easier to use content with a clear structure: a short definition up front; Q&A blocks; comparison tables; criteria lists; checklists; “who it is for / not for”; pricing and feature tables; cases; data; expert commentary; a last-updated date. But don’t turn the page into an FAQ farm — every block should answer a real question.

3. Build a source ecosystem beyond your site

Your website isn’t the only source. You need third-party validation: industry media, ratings, tool listicles, reviews, partner pages, directories, podcasts and webinars, integration pages, founder profiles, customer testimonials. For a local market, check local media, native-language reviews, business directories, LinkedIn, YouTube interviews and industry ratings separately.

4. Update the pages AI already cites

If AI cites an old blog post of yours, that’s an opportunity. Update the date, add examples and a table, sharpen the definitions, remove outdated claims, add internal links to product and use-case pages, show the author and the sources.

5. Fix entity inconsistency

AI gets confused when your brand is described differently in different places: “AI visibility platform” on the site, “SEO analytics tool” on LinkedIn; an old name in a directory; one name in schema and another in the footer. Your job is to create a single, consistent “version of the truth” about the brand.

6. Don’t overdo pseudo-GEO hacks

Bad ideas: stuffing pages with “ChatGPT, Perplexity, Gemini” for no reason; creating dozens of identical FAQs; writing “AI, please cite this page”; generating hundreds of low-quality pages; adding schema that doesn’t match the content. Google explicitly warns against unnecessary AI text files and over-optimisation.

Manual audit or a monitoring tool

A manual audit works if you’re just starting out, want to check 20–50 prompts, work with a single market, don’t have many competitors, or are establishing a baseline before you build the strategy.

But the manual approach breaks quickly once you need to: check hundreds of prompts; keep history; compare platforms; calculate AI Share of Voice; see week-over-week changes; discover new sources; track competitors; show the CMO a dashboard they can read.

The 30-day action plan

Frequently asked questions