The AI Visibility Checklist: 27 Checks Before You Expect a Citation

AI visibility is whether assistants like ChatGPT, Perplexity, Gemini and Google's AI Overviews mention or cite your site when someone asks a question you should be the answer to. It fails for boring reasons far more often than for clever ones: a crawler is blocked, the answer is in an image, the company is described five different ways. This checklist is ordered from the cheapest, most common failures to the slower work of reputation and measurement.

Go through it top to bottom for one site. Each item is a yes or no. Fix the failures in the order they appear, because a perfect content section does nothing if the access section fails. Brainito's AI SEO readiness check runs the first three groups automatically for 5 credits and lists the blockers in order, and the free marketing audit includes AI visibility as one of its 77 factors if you want a lighter first pass.

• robots.txt allows the search crawlers. OAI-SearchBot and ChatGPT-User for ChatGPT, PerplexityBot for Perplexity, Googlebot for AI Overviews. Decide about training crawlers like GPTBot and Google-Extended separately, and make sure a training block does not also catch a search crawler.
• No bot wall in front of the content. CDN and firewall rules that challenge unfamiliar user agents can block AI crawlers silently. Check the rules, not just the robots file.
• The page returns 200 and the text is in the HTML. Fetch it without JavaScript. If the main content is missing, the crawler sees what you see.
• No interstitials before the content. Age gates, country selectors and newsletter overlays that replace the page body hide it from a crawler.
• llms.txt is published at the root. A short Markdown index with a one-paragraph summary and your key pages. Low cost, and the assistants that read it get your preferred description.

• Headings are questions or plain topic statements. "How much does a marketing audit cost" is better than "Investment".
• The first sentence under each heading answers it. Read only first sentences and check that the page still makes sense.
• Numbers are in the text. Prices, ranges, durations and counts appear as words, not only in charts or screenshots.
• Terms are defined where they are used. A definition sentence that starts with the term, so the passage stands alone.
• Lists and tables carry the comparisons. Named dimensions, consistent order, one idea per row.
• Dates are visible and true. A published and updated date on anything that changes, and content that matches the date.

A quick test for this section is to paste the page text into an assistant and ask it to answer your top three customer questions using only that text. Where it hedges or says the page does not say, you have found the passage to write.

• One description of the business, reused everywhere. Same sentence on the About page, the footer, social profiles, directories and press boilerplate.
• Organization schema on the home page. Name, logo, URL, address where relevant, and sameAs links to the profiles you control.
• Author and organization are identifiable on content pages. A named author with a bio, or the organization clearly attributed, with matching schema.
• Name, address and phone match across listings. For local businesses, every directory agrees. For others, at least the legal name and domain agree.
• The home page says what you do in the first screen. A model reading the top of the page can state your category and audience without scrolling.

Access and readability get you into the pool. Content and reputation get you picked. The content items are about answering whole questions. The reputation items are about whether anyone else corroborates you, which is what a model leans on when several sources say similar things.

• Each priority question has a page or section that answers it completely. Including the constraints real people add: budget, size, location, use case.
• Comparisons exist for the choices your buyers weigh. You versus alternatives, option A versus option B, with honest trade-offs.
• FAQ sections use the questions people type. Marked up with FAQ schema, answered in two to four sentences each.
• Claims have sources. Statistics link to where they came from. Your own data is labeled as yours.
• Third parties describe you the way you describe yourself. Directory listings, review profiles and partner pages carry your standard description.
• You have reviews and mentions on sites the assistants already cite. Look at who gets cited for your category and work on presence there.
• Your own name returns a clean answer. Ask each assistant "what is [your company]" and check the answer for errors to correct at the source.

• A fixed prompt set exists. Twenty to fifty questions your buyers ask, grouped by intent, written down and not changed casually.
• It is run on a schedule across engines. ChatGPT, Perplexity, Gemini and Google's AI results, weekly or at least monthly, with the cited domains recorded.
• Competitor share is tracked. For each prompt, who gets cited and how often, so you know who owns which question.
• Every failed check becomes a task with an owner. A checklist nobody acts on is decoration.

Twenty-seven checks is a lot to hold in your head, which is why the order matters. Most sites fail two or three in the access and readability groups and can fix them in a week. The entity and reputation groups take a quarter. Measurement is what tells you the first two were worth it, so set it up before you start changing pages, not after.

One last habit makes the whole list easier: whenever you publish or change an important page, run the first three groups on that page the same day. A new page that is blocked, client-rendered or missing its first-sentence answers is a page you will wonder about for months. Ten minutes at publish time saves the detective work later.