Can I audit my own AI search visibility? A checklist

Last updated October 14, 2026. Written by the Third Coast AI team.

Yes, and a useful first audit takes about a day: check that the AI crawlers can fetch your pages and that the important text is in the raw HTML, run a short set of fact-check and buyer prompts in each engine and write down what comes back, open the pages the engines cite for your category and note what they do that yours do not, confirm your business is described the same way on every profile and in your schema, and score your main pages on whether they answer the question early with sourced numbers, a quote, a list and a table. None of this needs a tool or a vendor, and the result is a short list of fixes you can price before deciding whether to buy more.

This checklist is deliberately the public version. It covers everything a competent reviewer can check from the outside and leaves out the parts of our method that need our own tooling, which is why it ends in a list of fixes rather than a content plan. If you want the full version run for you, the what-to-expect guide describes what a paid audit adds.

Can the engines reach your pages at all?

Check this first because nothing else matters if the answer is no. Cyrus Shepard's May 2026 synthesis of 54 studies scores crawler accessibility 9.5 out of 10, the highest of 23 factors, and it is the one we most often find broken on otherwise well-run sites, usually through a CDN's bot protection rather than robots.txt. OpenAI's crawler documentation names three agents: OAI-SearchBot, which surfaces sites in ChatGPT search and which a site must allow to appear there; GPTBot, which gathers training data; and ChatGPT-User, which fetches a page when a user asks about it. The same documentation says a robots.txt change takes about 24 hours to be picked up and that sites disallowing OAI-SearchBot will not be shown in ChatGPT search answers. The crawler guide lists every agent and the robots.txt lines.

1. Open yoursite.com/robots.txt and look for Disallow rules that apply to GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended or Bingbot. Note that blocking Google-Extended affects Gemini training but not Google Search.
2. Fetch your three most important pages with each crawler's user-agent string using curl or a similar tool and confirm you receive the page, not a challenge screen or a 403. If you use Cloudflare or a similar CDN, check its bot settings; the blocking is often there rather than in robots.txt.
3. View the raw page source (not the rendered page) and confirm the headings, the first paragraph and the tables are present. ChatGPT's retrieval does not execute JavaScript, so anything rendered client-side is invisible to it.
4. Verify the site in Bing Webmaster Tools and open the AI Performance report, which has been in public preview since Bing's February 2026 announcement and shows whether Copilot and Bing's AI answers cite you at all.

What do the engines say about you right now?

Write ten fact-check prompts (where you are based, what you sell, who you serve, what you cost if public, who runs the company, how long you have existed) and ten buyer prompts (the questions a customer would ask when choosing a company like yours), and run each in ChatGPT with search on, in Gemini, in Google AI Mode, in Perplexity and in Copilot. Run each prompt at least twice, because the answers vary; SparkToro and Gumshoe.ai found less than a 1 in 100 chance of the same brand list repeating on a single prompt. For each answer, record whether you were named, which companies were named instead, which pages were cited, and every factual error about you. The errors go into a log with the page the engine cited as its source for each one, because that page is where the correction has to happen, and the correction guide covers the rest.

Which pages do the engines cite instead of yours?

For each buyer prompt, open the pages that were cited, including the ones from competitors and the ones from directories or publications, and spend ten minutes with each. Note where the answer sits on the page, how many statistics appear in the first few hundred words, whether there is a table, how the headings are phrased, and what every cited page says that yours does not. This is the single most informative step in the checklist and the one most often skipped, because it is reading rather than checking. Ahrefs measured only 38% of AI Overview citations coming from Google's top ten in March 2026, so do not assume the cited pages are the ones that rank; they are often lower-ranked pages that answer a narrower question well.

Is your business described the same way everywhere?

• Your site's Organization schema has a name, URL, logo, address, founding date and a sameAs list pointing to every official profile.
• Every profile (Google Business Profile, LinkedIn, Bing Places, industry directories, the chamber, Crunchbase if relevant) uses the same legal name, the same description and the same address format.
• A Wikidata item exists for the company with the same facts, and the schema sameAs points to it; the entity guide walks through creating one.
• If a Google knowledge panel exists, it is claimed; Google's knowledge panel update page says verified users' suggested edits are reviewed within a few days, which is the fastest correction route Google offers.
• Old addresses, discontinued products and former names are either removed from third-party pages or clearly marked as former.

Do your main pages answer the way cited pages do?

Score your five most important pages against the elements the research associates with citation. Aggarwal and colleagues at Princeton measured quotations, statistics and cited sources lifting a page's visibility in generated answers by about 41%, 33% and 28%, and keyword stuffing lowering it by 9%. Zhang and Yao found top-quartile heading count cited at 12.5 times the bottom quartile and list density at 8.94 times, and found bolted-on Q&A formatting reduced absorption by about 5.7%, so the elements need to be part of the page rather than appended to it. For each page ask: is the question answered fully in the first 150 to 200 words, are there numbers with named sources, is there an attributed quote, a list and a table, are the headings phrased as a buyer would ask, and is all of it in the raw HTML? A page that scores poorly on three or more items is a rewrite candidate, and the decision-factors page explains what the content of the rewrite should prove.

What does the checklist not tell you?

It does not tell you which decision factors the engines weigh for your category or how often, it does not show you the sub-questions behind each buyer prompt, and it does not produce a content plan with briefs modeled on the cited pages. Those are the parts of a paid audit that need repeated retrieval-on runs and our own capture tooling, and they are what turns a list of fixes into a program. If the checklist leaves you with a short list and a clear site, you may not need more than a tool subscription and a quarterly re-run; the agency versus tool guide covers that call. If it leaves you with a long list and no idea which pages to write, that is the point at which an audit earns its fee.

Frequently asked questions