# What an AI Visibility Audit Should Show You > An AI visibility audit should show which brands ChatGPT, Gemini, Claude, Perplexity and AI Overviews name for the questions your buyers ask, which sources those answers read, whether AI crawlers can reach your pages, and how far visibility is from pipeline. If the prompts contain your brand name, or the output is one score, it's a dashboard, not an audit. By Othmane Khadri, founder of Earleads. Published 2026-10-08. Canonical: https://www.earleads.com/blog/what-an-ai-visibility-audit-should-show-you/ ## What is an AI visibility audit? An AI visibility audit is a dated record of how AI assistants answer the questions your buyers ask before they buy. For each question it shows which brands ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews name, which pages they cite, whether AI crawlers can read your site, and what the gaps cost you in pipeline. Some vendors call it an LLM visibility audit or a GEO audit. The contents should be the same. A quick word on who's writing this. [Earleads](https://www.earleads.com/) runs AI visibility audits, so we sell a version of the thing this guide teaches you to judge. We don't rank tools or agencies here. Read it as a checklist, and hold our audit to it as hard as anyone's. A checklist matters because two very different things get sold under the same name. An audit tells you where you stand on the questions that lead to revenue and what to fix first. A pretty dashboard gives you a big number, a trend line and a feeling of progress. Only one helps you decide anything. ## Why so many AI visibility reports are pretty dashboards Most weak reports fail in one of three ways, and you can spot all three in the first five minutes of a demo. **The prompts flatter you.** Prompts get copied from your own blog titles, or they contain your brand name. "What is [Brand]?" "Is [Brand] good for small teams?" An engine asked about you by name will talk about you, so the chart starts high and stays high. It measures recognition by people who already know you, which isn't the problem you're paying to solve. **One run is treated as a fact.** AI answers change from one run to the next. SparkToro had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI Overviews a combined 2,961 times. There was less than a 1 in 100 chance of getting the same list of brands twice, and roughly 1 in 1,000 of getting the same list in the same order ([SparkToro, January 2026](https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)). The same research concludes that a visibility percentage across many prompts, each run several times, is a reasonable thing to track, and that tools reporting a "ranking position" in AI are "full of baloney." A report built on one run per prompt is mostly measuring noise. **Everything collapses into one score.** An AI visibility score out of 100 is easy to put in a board deck and hard to act on. It hides which questions you lose, on which engine, to whom, and because of which source. When the number moves, nobody can say why. ![Two columns comparing a pretty dashboard, built on branded prompts and one run each and ending in a single score out of 100, with a real audit built on buyer questions approved by you, a citation map per question and engine, and ending in a short list of first moves](/blog/what-an-ai-visibility-audit-should-show-you/diagram-1.svg) *Same name, different product. Only one of them tells you what to do on Monday.* None of this means tracking tools are useless. It means the prompt list, the run counts and the sources have to be visible, or the output can't be checked. A recent evidence standard published by Pixelmojo puts it well. It asks every report to disclose "the exact prompts; which ones contain the brand name," and every rate "with its count, such as '3 of 8 responses'" ([Pixelmojo, AI visibility evidence standard](https://www.pixelmojo.io/blogs/ai-visibility-evidence-standard)). The six steps below are what a real AI visibility audit should contain, in the order it should be built. ## Step 1: The buyer questions, approved by you Every number in the audit depends on this list, so it comes first and it's yours to approve before anything is measured. Good questions come from the places buyers already speak. Sales call notes, support tickets, review sites, the threads where people compare options, and the search terms that already bring you visits. Most guides on page one recommend somewhere between 10 and 50 prompts. Fewer questions that buyers really ask beat a long list padded with variations. What a solid list looks like: - Discovery questions, such as "best payroll software for a 50 person team" or "fragrance free moisturizer for sensitive skin." - Comparison questions, such as "alternative to [incumbent] for ecommerce" or "[competitor] vs [competitor] for a mid-size sales team." - Problem questions, such as "how do I stop losing leads between the form and the CRM?" - Constraint questions with real limits, such as budget, location, team size or delivery time. Then run the two-minute test. Count how many prompts contain your brand name. In a visibility audit the answer should be zero. Branded prompts still have a job, which is checking whether AI describes you correctly, but they belong in a separate reputation section and shouldn't count toward any visibility number. We go deeper on this in our guide on [how to choose a GEO agency](/blog/how-to-choose-a-geo-agency/), where the same test sorts serious agencies from the rest. ## Step 2: A citation map per question and per engine The core of an AI visibility audit is a grid. Questions down the side, engines across the top, and in each cell who won and how. "Won" needs a definition, because AI visibility isn't one thing. For every question and engine, a useful map records: 1. **Mentioned.** Your brand appears somewhere in the answer. 2. **Recommended.** The answer suggests you for the need in the question. 3. **Cited.** One of your pages is listed as a source. 4. **Who won instead.** The brands named when you weren't, and the ones named first. 5. **How often.** The rate across several runs, with the count, like "named in 3 of 8 runs." Engines differ enough that each needs its own column. ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews don't read the same sources or answer the same way, and a brand can lead on one while missing from another. A map with one engine, or one blended number, hides the very gap you'd want to close. Watch for position claims. Given SparkToro's finding that the order of a list almost never repeats, "you're number 2 in ChatGPT" isn't a stable fact. "Named in 6 of 10 runs, first in 2 of them" is. ## Step 3: The sources each answer reads This is the part most AI visibility tools skip, and it's where the plan comes from. For each question, an audit should list the pages, threads and videos the engines draw on, with links you can open. Google describes why this matters. Its AI Overviews and AI Mode "may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources" ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). An answer about your category is built from many pages, and most of them aren't yours. In practice the sources fall into five groups: - Articles written by practitioners and review sites. - Community threads where buyers compare options. - Your own pages, if crawlers can read them. - YouTube videos and their transcripts. - Other sites that mention and link to you or your competitors. When the sources are listed, the work becomes obvious. If every answer to "best [category] for startups" cites the same three roundup articles and one thread, those four pages are the job. If your own comparison page is cited but says something out of date, that's a one-hour fix. An audit that names sources hands you a to-do list. One that doesn't hands you a mood. ## Step 4: A crawler and technical check None of the work above helps if AI crawlers can't fetch your pages. An audit should check, page by page for the pages that matter, that the right bots get through. The bots aren't interchangeable, and blocking the wrong one is common: - OpenAI says "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features," and sites opted out of it "will not be shown in ChatGPT search answers, though can still appear as navigational links." GPTBot is a separate crawler for training ([OpenAI, bots](https://developers.openai.com/api/docs/bots)). Blocking GPTBot and allowing OAI-SearchBot is a valid choice. Blocking both by accident isn't. - Perplexity says PerplexityBot is "designed to surface and link websites in search results on Perplexity" and isn't used to train foundation models ([Perplexity, bots](https://docs.perplexity.ai/guides/bots)). - For Google's AI features, a supporting link must be "indexed and eligible to be shown in Google Search with a snippet," and Google adds there are "no additional requirements" beyond that ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). Robots.txt isn't the only gate. Firewalls and CDNs make their own calls. Cloudflare announced that from September 15, 2026, new domains get Training and Agent bots blocked by default on pages that display ads, while Search crawlers stay allowed, and that multi-purpose crawlers such as Googlebot, Applebot and BingBot get blocked for customers who choose to block Training ([Cloudflare blog](https://blog.cloudflare.com/content-independence-day-ai-options/)). An audit should test what a bot actually receives, not just read the robots file. The rest of the technical check is ordinary. Pages render without heavy JavaScript, key facts sit in plain text, prices and plans are public where they can be, and the same one-line description of what you sell appears across your site and profiles. ## Step 5: The gap to pipeline A citation is a line in a chat window. It isn't revenue until someone clicks, gets identified and buys. An AI visibility audit should show how far today's visibility is from that, even if the honest answer is "we can't see it yet." Pixelmojo's standard separates five outcomes, a mention, a recommendation, a cited source, a visit and a sale, and asks that they're kept apart. That's the right frame for this step. ![A ladder of five outcomes from mention to recommendation to cited source to visit to sale, with most dashboards stopping after the cited source and a real audit continuing to visits, CRM records and revenue by engine](/blog/what-an-ai-visibility-audit-should-show-you/diagram-2.svg) *Most dashboards stop at the citation. The money is two rungs further down.* What the pipeline part should check: - **ChatGPT referrals.** ChatGPT adds utm_source=chatgpt.com to links it sends from search, according to [OpenAI's publishers and developers FAQ](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq), so those visits can be split out in analytics today. - **Google AI features.** Google says sites in AI features are "included in the overall search traffic in Search Console," in the normal Web report ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). There's no separate AI Overviews line, so any report claiming "AI Overview traffic from Search Console" should explain how it was separated. - **Landing pages.** Which pages AI-referred visitors land on, and whether those pages have a next step worth taking. - **The CRM.** Whether an AI-referred visit can become a named company or contact, with an owner, so that pipeline and revenue can be reported by engine later. Many audits find this plumbing missing. That's a finding, not a failure, and it's worth more than a chart that pretends otherwise. ## Step 6: First moves, ranked An audit ends with a short list of what to do first, not a 90 slide appendix. Each move should name the question it's for, the source or page it touches, and why it's ranked where it is. A good list is usually short and specific: - Unblock OAI-SearchBot on the pricing and comparison pages. - Correct the out of date plan details on the comparison article cited for two questions. - Answer the community thread cited by three engines for the top discovery question. - Add a visible starting price and an integrations list, so answers can confirm the facts they repeat. - Tag AI referrals and route identified visits into the CRM with an owner. If you can't hand the list to someone on Monday and have them start, the audit isn't finished. ## Audit, tool or report: which one do you need? The three get sold under the same names, so it helps to separate them. An audit is a one-time diagnosis. An AI visibility tool is software that reruns prompts and tracks the results over time. An AI visibility report is the recurring summary you or a vendor send to the team. Most teams need the audit first, because it produces the question list the tool and the report should track. | | AI visibility audit | AI visibility tool | AI visibility report | |---|---|---|---| | What it is | A dated diagnosis with a plan | Software that reruns prompts on a schedule | A recurring summary for the team | | Best for | Deciding what to fix first | Tracking movement across many runs and markets | Keeping everyone looking at the same numbers | | Who picks the prompts | You approve them | Often the tool suggests them, so check | Inherited from the audit or the tool | | Shows the sources answers read | It should, per question | Varies widely, ask for exports | Only if the tool collects them | | Connects to pipeline | It should find the gap | Rarely on its own | Only if analytics and CRM are wired | | Main risk | A slide deck with no ranked moves | Suggested branded prompts and one blended score | A trend line nobody can explain | When you evaluate an AI visibility tool or an AI visibility checker, three questions settle most of it. Can I load my own prompts and lock them? Can I export the raw answers and every cited URL? Does it show run counts per prompt instead of one result? A tool that says yes to all three can carry a real audit forward. One that says no will give you a number you can't check. ## The AI visibility audit checklist Hold any audit, tool or report you're offered against this table. Fill the last column during the demo or when the deliverable lands. ![A flow of the six parts of a real AI visibility audit, buyer questions approved by you, a citation map per question and engine, the sources each answer reads, a crawler and technical check, the gap to pipeline, then a ranked list of first moves](/blog/what-an-ai-visibility-audit-should-show-you/diagram-3.svg) *Six parts, in this order. Skip the first and every number after it is suspect.* | What to check | Real audit | Pretty dashboard | Yours | |---|---|---|---| | Prompt source | Buyer questions you approved | Copied from your articles, or suggested by the tool | | | Brand name in prompts | Zero in the visibility set, branded kept separate | Many, counted toward the score | | | Engines | ChatGPT, Gemini, Claude, Perplexity, AI Overviews, each shown apart | One engine, or one blended number | | | Runs per prompt | Several, with counts like "3 of 8" | One run, shown as fact | | | Outcomes | Mention, recommendation, citation, visit and sale kept apart | One AI visibility score | | | Competitors | Who wins each question, by engine | A share of voice chart with no questions behind it | | | Sources | Pages, threads and videos per answer, with links | Not shown | | | Crawler check | What each bot actually receives, CDN included | Robots.txt glance, or nothing | | | Pipeline gap | ChatGPT referrals, Search Console caveat, CRM routing | Stops at citations | | | Output | A short list of ranked first moves | A score and a trend line | | | Ownership | You keep the prompts, data and exports | Lives in the vendor's login | | The first two rows matter most. If the prompts are flattering, every other row inherits the bias, however polished it looks. ## B2B vs B2C: what the audit should look at The structure of the audit doesn't change between B2B and B2C. The questions, the sources and the outcome you trace do. ### B2B: shortlists and committees B2B buyers ask AI for shortlists, alternatives and integration checks. "Alternatives to [incumbent] for a 200 person sales team." "Does [category] tool integrate with HubSpot?" The sources tend to be practitioner articles, review directories and community threads. The audit should check whether your pricing, integrations and security pages are public enough for an answer to confirm them, and it should trace AI-referred visits to named accounts, meetings and pipeline. ### B2C: products and reviews B2C buyers ask for products with constraints. "Best running shoes for flat feet under $120." "Is [brand] worth it?" Reviews, community threads and videos carry more weight, and product data matters as much as content. The audit should look at whether price, size and stock are readable by AI shopping surfaces, a topic we cover in [how to get recommended by AI agents](/blog/how-to-get-recommended-by-ai-agents/), and it should trace AI referrals to accounts created and orders. ## How to check if AI agents recommend your brand yourself You can run a rough version of this audit in an afternoon, with no tool. It won't have the run counts or source extraction of a full audit, but it will tell you whether you have a problem. 1. Write 10 buyer questions without your brand name in any of them. 2. Run each one in ChatGPT, Gemini, Claude, Perplexity and a Google search that shows an AI Overview. Use fresh sessions, and run each question at least three times per engine. 3. For each run, note whether you were mentioned, recommended or cited, who was named instead, and the links in the answer. 4. Open the cited links and list the pages and threads that show up again and again. 5. Check your robots.txt for OAI-SearchBot and PerplexityBot, and filter your analytics for utm_source=chatgpt.com. If you'd rather have it done properly, our [free AI visibility audit](/#audit) maps the 5 conversion queries competitors are winning in 7 days, shows who wins each one on every engine, lists the exact pages and threads the engines read, and gives you the first moves to win them back. Hold it to the checklist above, the same as any other. ## Frequently asked questions ### What is an AI visibility audit? An AI visibility audit is a dated snapshot of how AI assistants answer the questions your buyers ask before they buy. It records which brands ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews name and cite for each question, which pages and threads those answers draw on, whether AI crawlers can read your site, and what the gaps mean for pipeline. ### Is an AI visibility score reliable? A single score is the least reliable part of most reports. SparkToro found AI tools return the same list of brands less than 1 time in 100 for the same prompt, so one run per prompt is noise. A visibility rate across many runs of a fixed, buyer-written prompt set is defensible. A score with no prompts, run counts or sources behind it isn't. ### Should branded prompts be part of an AI visibility audit? Yes, but kept apart. Prompts that contain your brand name test whether AI describes you correctly, which is a reputation check worth running. They shouldn't count toward your visibility number, because an engine asked about you by name will nearly always mention you. Visibility should be measured on questions buyers ask before they know you exist. ### How is an AI visibility audit different from an SEO audit? An SEO audit checks crawling, indexing, rankings and page quality for search engines. An AI visibility audit keeps the crawl check but adds what search audits skip, meaning which brands AI answers name per question and engine, the third party pages and threads those answers read, and AI referral traffic. Google says AI features run on the same foundations as search, so the two overlap. ### How often should you rerun an AI visibility audit? Run the full audit once to set a baseline, then rerun the same prompt set on a fixed schedule, weekly or monthly, with several runs per prompt each time. Keep the question list fixed between reruns so changes mean something. Redo the full audit when you enter a new market, launch a product, or a competitor starts appearing in answers you used to win. ### Do I need an AI visibility tool to run an audit? Not for a first pass. You can run 10 to 20 buyer questions by hand across five engines in an afternoon and log the results in a spreadsheet. A tool earns its fee when you need many runs per prompt, several markets or weekly tracking. Pick one that lets you load your own prompts and export the raw answers and cited URLs.