How to audit your brand's visibility in AI search

AI search visibility is easy to misunderstand because the result rarely looks like a traditional ranking page. A brand can be visible without getting a click, mentioned without being cited, or cited for the wrong thing. That means the audit has to answer a different question: when an AI system tries to help someone make a decision in our category, does it understand us well enough to include us accurately?

The good news is that the audit does not require a mysterious new discipline. Google’s current guidance says foundational SEO still matters for generative search, while Bing now exposes AI-specific citation data and OpenAI gives publishers direct control over whether OAI-SearchBot can crawl their pages. The work is new in places, but it is measurable.

What are you actually auditing?

An AI visibility audit looks at five layers at the same time: whether your brand appears, whether it is described correctly, which sources support the answer, whether your own content is eligible to be retrieved, and whether the experience changes when the question becomes more commercial or specific.

Do not turn those five layers into one vanity score. A brand mentioned often but described badly does not have strong AI visibility. Neither does a brand that only appears when someone types its exact name.

Step 1: Build a prompt set around real customer decisions

Start with questions, not keywords. AI assistants are strongest when the user gives context, constraints and a task, so your test set should look like real conversations rather than a list of short search terms.

1. Discovery prompts. Questions where the user knows the problem but may not know your category or brand yet. Example: “How can a retail brand improve conversion before spending more on ads?”
2. Category prompts. Questions that ask for options in your category. Example: “What should I look for in a marketing agency for GCC expansion?”
3. Comparison prompts. Questions that force trade-offs. Example: “Agency vs in-house team for a market launch: what are the pros and cons?”
4. Commercial prompts. Questions close to action: price, location, availability, integrations, service scope, timeline, proof or purchase criteria.
5. Brand prompts. Your brand name plus the questions a buyer would ask before trusting you: what you do, who you are for, what makes you different, where you operate and whether you are credible.

For a focused audit, 20 to 30 prompts is usually enough to reveal patterns. Keep the wording natural and save the exact prompt so you can rerun it later. AI answers can vary, so a single test should never be treated as permanent truth.

Step 2: Test more than one AI surface

Run the same intent across the AI surfaces that matter to your audience. At minimum, compare ChatGPT search, Google’s AI features in Search, and Microsoft Copilot/Bing. If your audience relies on another assistant or marketplace, include it too. The purpose is not to declare one platform “right”; it is to see where your brand is understood differently.

• Record the answer date. These systems change frequently, and freshness matters.
• Capture the exact brand language. Note the category, benefits, weaknesses, market, audience and alternatives associated with you.
• Open the citations. The source is often more useful than the answer because it tells you what information layer is shaping the model.
• Repeat important prompts. For high-value questions, test more than once and across logged-in/logged-out or location contexts when practical.

Step 3: Create a simple visibility sheet

You do not need an expensive platform to start. A spreadsheet with one row per prompt can tell you a lot if the columns force you to separate what happened.

After 20 or 30 rows, patterns usually become visible. Maybe you are strong on branded questions but absent from comparison queries. Maybe the same third-party article is teaching multiple systems about your category. Maybe your site is cited, but an outdated service page is the page being retrieved.

Check out what our CEO is writing about: SEO is dead. Learn AEO and GEO.

Step 4: Check what the platforms can measure for you

Manual testing tells you what an answer looks like. Platform data tells you whether the pattern is broader. In 2026, both Google and Bing added reporting that makes AI visibility less speculative.

Google Search Console

Google’s Generative AI performance report is available to websites worldwide and reports impressions from AI Overviews and AI Mode, including the pages shown, country, device and date. It does not give you the user’s full prompt, but it shows which URLs are actually earning visibility inside Google’s generative experiences.

Bing Webmaster Tools

Bing’s AI Performance report goes further into citation behavior. It shows total citations, cited pages and grounding queries, with additional views such as intent, topic and citation share in preview. That makes it useful for spotting which themes are already associated with your site and where a strategically important topic has little citation presence.

Use these reports as trend data, not as a perfect ledger. Bing explicitly notes that its AI data is aggregated and sampled, and Google reports impressions rather than a full explanation of why a page was selected.

Step 5: Check whether AI systems can actually access your content

If a page is blocked, difficult to crawl, hidden behind login, dependent on broken JavaScript or missing from the index, the content strategy can be excellent and still fail at retrieval.

• Google. Verify crawlability and indexability, canonical URLs, mobile/page experience, JavaScript rendering and duplicate-content issues. Google says its generative features are grounded in the Search index and core ranking systems.
• OpenAI. Allow OAI-SearchBot if you want eligible public content to be surfaced and cited in ChatGPT search. OpenAI separates OAI-SearchBot from GPTBot, so search visibility and model-training preferences can be managed independently.
• Bing. Confirm the site is eligible for Bing indexing and consider IndexNow for frequently changing content so updates can be discovered faster.

This is also a good moment to review your marketing and digital visibility foundations. AI search is another discovery layer, not a reason to abandon the technical and content work that already makes a site understandable.

Step 6: Audit whether each important page is easy to answer from

AI systems do not need every page to read like an FAQ, but they do benefit from content that is clear enough to retrieve and reuse without guessing. The page should make the subject, entity, evidence and answer obvious to a human reader first.

• Answer the main question early. Do not make the reader cross five paragraphs of setup before learning what the page is about.
• Use descriptive headings. A heading should tell the reader what the next section answers, not simply say “More information.”
• Make important statements self-contained. A useful sentence should still make sense if it is quoted outside the paragraph around it.
• Show evidence. Use original examples, first-hand experience, data, named sources, methodology, comparisons and dates where relevant.
• Keep entity facts consistent. Names, locations, service scope, product details and other core facts should not contradict each other across pages and profiles.
• Update the content. Outdated pricing, products, people, market data or regulations weaken both user trust and retrieval quality.

Google’s 2026 guidance is unusually clear on one point: unique, non-commodity content matters more than trying to manufacture hundreds of pages for every possible AI query. If the page could have been produced from the first page of generic search results, it is probably not adding much information to the ecosystem.

Step 7: Find the sources teaching AI systems about your brand

The answer may not come from your website. Reviews, industry publications, retailer listings, directories, social profiles, partner sites and community discussions can all become part of the source layer around a brand. This is why AI visibility is partly a content problem and partly an authority and consistency problem.

• List recurring third-party sources. Which domains appear again and again beside your category or competitors?
• Check whether they represent you accurately. A strong directory listing with the wrong service description is not a win.
• Look for missing proof. If competitors are cited through awards, expert articles, reviews, case studies or retail data and you have none of that evidence, the gap is not solved by rewriting your homepage.
• Do not manufacture mentions. Artificial references and low-quality placements create noise. Build genuine sources that have a reason to talk about the brand.

Step 8: Add the commerce and local layers when they matter

For ecommerce, structured product information is becoming part of discoverability. OpenAI says merchant rankings in shopping can use factors such as availability, price, quality and whether the merchant is the maker or primary seller, and it supports direct product feeds for merchants that want fresher product data. For local businesses, accurate location, hours and contact information are equally important.

Treat product feeds, merchant data and local listings as part of the same brand truth. An AI assistant cannot recommend an item confidently if the website says it is available, a feed says it is out of stock and a marketplace uses an old price.

A practical AI visibility scorecard

Use the scorecard to prioritize work, not to create a vanity benchmark. Score each area from 0 to 2: 0 = weak or absent, 1 = inconsistent, 2 = strong and repeatable.

What should you fix first?

Prioritize the gap that affects the most valuable customer decision, not the easiest edit. A sensible order is:

1. Wrong information. Correct facts that could mislead customers or AI systems.
2. Access problems. Fix crawl, index, robots, canonical, JavaScript or protection issues that prevent retrieval.
3. Missing decision content. Create or improve pages that answer the questions buyers actually ask before choosing.
4. Weak evidence. Add first-hand expertise, sources, proof, comparisons, examples and current data.
5. Authority gaps. Strengthen legitimate third-party references, reviews, partnerships and industry coverage.
6. Measurement. Rerun the prompt set and compare Google/Bing visibility after enough time has passed for systems to recrawl and update.

What not to do

• Do not create one thin page for every prompt variation. Google explicitly warns against scaled content created mainly to manipulate rankings or generative responses.
• Do not treat llms.txt as a magic switch. Google says unnecessary AI text files such as llms.txt are not required for visibility in its generative Search features.
• Do not confuse citation with conversion. A citation tells you your information was used; it does not prove the user clicked, trusted or bought.
• Do not optimize only for your brand name. The important visibility often happens before the user knows what company to ask for.
• Do not let AI visibility become detached from the business. Prioritize the questions connected to consideration, trust, qualified demand and revenue, not simply the prompts where you can get a mention.

A 30-day audit rhythm

AI visibility changes too quickly for a once-a-year report. For most brands, a light monthly check is enough to catch meaningful movement without turning the exercise into daily rank watching.

1. Week 1, baseline. Run the prompt set, export Google/Bing AI visibility data and identify the highest-value gaps.
2. Week 2, fixes. Correct technical access, inaccurate facts, weak pages and missing evidence.
3. Week 3, authority. Improve legitimate third-party profiles, citations, reviews and expert content where the audit shows a source gap.
4. Week 4, recheck and document. Rerun priority prompts, note what changed and keep the next action list short.

The goal is not to make every AI assistant say exactly the same thing. The goal is to make the web around your brand clear, current and useful enough that different systems can reach a trustworthy answer.