How to audit AI search visibility

Imagine a spend-management company tests this question:

Which tools suit a US software company that needs NetSuite integration and multi-entity controls?

The answer names the company but does not recommend it. It links to a publisher comparison and two competitor integration pages, but not to the company’s site.

The description is broadly correct, although it leaves out multi-entity support.

That gives the team a specific finding: the brand was named, its own domain was absent from the sources, and the description missed a relevant capability.

The team could check whether its integration and multi-entity pages explain that capability clearly. Missing information or weak supporting evidence are possible explanations to investigate.

The next test is to review the relevant product and integration pages against the buyer’s question.

If an approved change is made, wait until the page can be crawled and repeat the same prompt under recorded conditions.

This is an illustrative example, not a Maxtor Labs or client result. The answer alone cannot establish its cause.

The prompt set matters more than the size of the monitoring dashboard.

Start with questions buyers ask while they define the problem, compare approaches, build a shortlist or validate a purchase.

Good inputs include sales questions, customer emails, support tickets, site search and recurring objections, provided your team can use that material legitimately.

Keep the buyer’s constraints in the prompt. Company size, market, existing software, security needs and implementation limits often change which answer is useful.

Cover the decisions that matter commercially:

• How does the buyer describe the problem or category?
• Which approaches or vendors are being compared?
• Which products fit a particular use case?
• What integrations, proof, risks or implementation questions affect the final choice?

There is no universal prompt count. Use enough questions to cover the important decisions, remove near-duplicates and stop when another prompt would not change what your team does.

Run the same set across the AI platforms that matter to your buyers.

Save the wording, date, platform or product surface, known market and language, and any relevant account state.

AI answers can change, so each result is a dated sample rather than a permanent rank.

Keep the complete answer and every visible source URL.

Record named, recommended or discouraged brands, owned-domain citations and the exact wording of an inaccurate description.

Use unknown for conditions you cannot confirm.

A colleague should be able to inspect the record without reconstructing the test from a summary.

Do not compress every appearance into one “visibility” label. A brand can appear in several distinct ways:

• Named: the answer mentions the company or product.
• Recommended: the answer presents the brand as a suitable choice for this use case.
• Cited: the answer includes a visible source link to a page.
• Owned citation: the answer cites the brand’s canonical domain.
• Named plus owned citation: the answer names the brand and cites its canonical domain somewhere in the same response.
• Discouraged: the answer advises against the brand or says it is unsuitable for the stated use case.

Counting the brand name as a recommendation would hide the problem in the example.

A source link records a different outcome from a recommendation. It does not prove which page caused the answer.

Next, compare the answer with the company’s current offer and approved evidence. Mark the description as accurate, partial, inaccurate or not applicable.

If it is partial or wrong, save the exact sentence. That gives the product-marketing or content team something concrete to investigate.

Record competitors with the same care. “Competitor present” says little.

“Competitor recommended for companies that need international payments, supported by an integration page” gives you a useful comparison.

Visible sources show some of the material used to support an answer. They do not reveal the full reason the model produced it.

Classify a source only as far as the page allows.

Useful categories may include the company site, an independent publication, a review or directory, a partner, a community or another clearly relevant source type.

Then ask four questions:

1. Can the relevant search engine or AI search crawler reach and process the priority page?
2. Does the company’s page answer the buyer’s question and support its main claims?
3. Do current sources describe the company, product, audience and category consistently?
4. Is there credible outside evidence for the claims a buyer needs to trust?

A competitor that appears beside several independent sources may point to an outside-proof gap. It does not prove that buying one listing or earning one article will produce a citation.

Microsoft’s Bing Webmaster Tools AI Performance reporting can add page-level evidence for supported Microsoft experiences.

Microsoft documents citation totals, cited-page activity, grounding queries and trends, while warning that those totals do not show placement, authority or the role a page played in an individual answer.

Its Topics Portfolio beta was described in October 2026 as directional rather than a precise ledger.

See the Bing AI Performance announcement and the Topics Portfolio beta announcement.

Before rewriting a group of pages, check whether the priority pages can be discovered and used.

Review indexability, snippet eligibility, robots and CDN rules, crawlable internal links, visible text, canonicals, page experience and structured data that matches the page.

Google says a page must be indexed and eligible to show a snippet before it can appear as a supporting link in AI Overviews or AI Mode.

Google also says those features do not require special AI text files, special markup or extra Schema.org data beyond normal Search requirements.

See Google Search Central’s AI features guidance and AI optimisation guide.

OpenAI lists OAI-SearchBot for surfacing websites in ChatGPT search results. GPTBot serves a separate purpose: crawling content that may be used to improve foundation models.

Allowing OAI-SearchBot can make a page eligible for ChatGPT search. Access does not guarantee a citation. See the crawler documentation for the platform’s controls.

Do not present llms.txt as a Google Search requirement. A team may run a clearly labelled platform-specific experiment, but it should not be sold as a universal visibility fix.

AI visibility matters when it helps the right buyer move forward. Keep four kinds of evidence separate:

• Answer evidence shows whether the brand appeared, how it was described and which sources were visible.
• Page evidence shows which pages were cited or received identifiable referral visits.
• Conversion evidence shows whether a visitor continued to a relevant service, tool or contact path.
• Revenue evidence shows whether a qualified buyer reported that AI-assisted research influenced the decision.

Direct referral traffic is useful but incomplete. A buyer may discover a brand in an AI answer, search for it later, visit directly or mention the discovery during a sales conversation.

Google includes traffic from AI Overviews and AI Mode within Search Console’s Web performance reporting rather than providing a clean standalone attribution view.

Bing’s AI Performance report provides a separate citation view for supported Microsoft experiences. Check the current platform documentation when reporting is configured because these products change.

Finish the audit with a short list of findings that affect a real buying decision. For each finding, state:

• what happened;
• the page or source to inspect;
• the possible explanation;
• the next test;
• the condition for repeating the prompt.

Prioritise a wrong or missing description in a vendor-shortlist answer before a broad informational prompt with no clear path to an enquiry.

Prefer a small check that can change a decision over a speculative site-wide project.

Do not promise a citation increase or a timeline without evidence. Name the result you can check, such as a corrected description, an indexable page, a stronger source, an identifiable referral or a changed answer in the next dated run.

An audit can show how selected AI tools answered selected buyer questions on recorded dates.

It can surface missing or wrong brand information, visible sources, access problems, weak page evidence and measurement gaps.

It cannot produce a permanent rank, guarantee a future mention or prove causation from one answer.

Google notes that AI Overviews and AI Mode may use different models and techniques, so answers and supporting links can vary.

Treat the audit as a decision-making baseline, not a universal score.

If you can already see that important buyer questions produce missing, partial or unsupported answers, the next step is to document the gap and inspect the pages and sources behind it.

The Maxtor Labs GEO Visibility Audit is a concise 6-page diagnostic of your AI visibility, key gaps and priorities.

• Google Search Central: AI features and your website
• Google Search Central: Optimizing your website for generative AI features
• OpenAI Developers: Overview of OpenAI Crawlers
• Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools Public Preview, 10 February 2026
• Microsoft Bing: Introducing Topics Portfolio (Beta) in AI Performance, 13 October 2026