The AI visibility checklist: 30 checks for your brand

This is the checklist we'd work through for any brand that wants to be named more often when buyers ask ChatGPT, Gemini or Claude for a recommendation. It has 30 checks in three groups: your own site, the third-party sources AI engines cite, and how you measure. Each one is specific enough to mark as done or not done, and most take minutes rather than days.

Copy it into a doc, go through it once, and you'll end up with a short, ranked to-do list instead of a vague feeling that you should "do something about AI".

How to use this checklist

Mark each check with one of four statuses. Be strict: if you're not sure, it's not a pass.

Work through Part 1 first. It's mostly quick, it's fully in your control, and the other parts depend on it. Part 2 is where most of the gains are, but it's slower. Part 3 is what tells you whether any of it worked, so set it up before you change anything big.

Part 1: your own site (12 checks)

Can AI crawlers read it?

• 1. robots.txt loads. Open yoursite.com/robots.txt in a browser. It should return a plain text file, not an error page or a redirect to your homepage.
• 2. OAI-SearchBot is allowed. This is OpenAI's crawler for ChatGPT search results. If it's blocked, your pages can't be fetched when ChatGPT searches the web for an answer.
• 3. GPTBot, ClaudeBot and Google-Extended are a deliberate choice. GPTBot is OpenAI's training crawler. ClaudeBot is Anthropic's crawler for model training and Claude web results. Google-Extended covers Gemini training and grounding, not Google Search ranking. Allow or block each on purpose, not by accident. Our robots.txt guide has copyable snippets.
• 4. Your firewall or CDN isn't blocking them anyway. Some hosting and CDN dashboards have a bot-blocking switch that overrides robots.txt. Check that it matches what your robots.txt says.
• 5. Key facts are in the page text. Your product description, pricing and features should be in the HTML as text, not only inside images, videos, PDFs or a widget that loads after the page.

Is it easy to quote?

• 6. Your homepage says what you do and who it's for in the first lines. Use the words your buyers use. "Invoicing for freelance designers" beats "The future of getting paid".
• 7. Your price is written in plain text. A number on a pricing page, with what's included. If prices vary, give a starting point.
• 8. One consistent product description. The same one or two sentences on your homepage, your About page, your social bios and your directory listings. Engines notice when sources agree.
• 9. Headings that match real questions, answered directly underneath. "Does it work with Stripe?" followed by "Yes. Connect your Stripe account in Settings and..." is easy to quote.
• 10. A fair comparison page for each main rival. One page per competitor buyers compare you with, with an honest table. How to write one has a template.
• 11. Use-case pages for the buyers you serve best. One page per audience, like "for agencies" or "for freelancers", describing their specific needs and how you meet them.
• 12. Facts are current. Features, integrations, limits and plan names on your site match the product today. Old help articles that mention a retired plan are a common source of wrong answers.

Some sites also publish an llms.txt file, a proposed plain-text summary of the site for AI tools. It's cheap to add with the llms.txt generator, but treat it as optional; the checks above matter more.

Part 2: third-party sources (10 checks)

AI answers lean heavily on pages you don't own: "best of" articles, review sites, community threads and other vendors' pages. Which websites AI engines cite goes through each type in depth.

• 13. You know which pages are cited for your questions. You have a list of the URLs engines cite when answering your buyer questions, grouped by site and counted. Without this list, the rest of Part 2 is guesswork.
• 14. You're on, or have pitched, the three most-cited "best of" articles. For each one: are you listed? If not, have you sent the author a short, honest note on who you're for and a fact they can verify?
• 15. Your profile is complete on the review site cited most for your questions. Category, pricing, screenshots, integrations, a current description. The review sites playbook has a full profile checklist.
• 16. You have recent reviews. At least a few reviews from the last 90 days, from real customers, asked for at a natural moment.
• 17. You respond to reviews. Including the critical ones, briefly and helpfully. Buyers read the replies, and so can engines.
• 18. You take part in the community threads where buyers ask for recommendations. As yourself, with your affiliation stated, answering the question before mentioning your product. The Reddit playbook covers how.
• 19. Your listings agree with each other. Directory entries, app marketplace listings, social profiles and partner pages use the same name, category, description and price as your site.
• 20. Integration partners list you. If you integrate with other products, you're in their app directory or integrations page, with a working link and a current description.
• 21. You have something worth citing. Original data, a useful free tool, a clear guide or a benchmark. Writers link to things like this, and linked pages get found.
• 22. Wrong facts about you have been corrected. If a cited page lists an old price, a missing feature or the wrong category, you've asked the owner to fix it, with the correct facts and a source.

Part 3: measurement (8 checks)

• 23. You have a written list of buyer questions. Usually 20 to 50, in the words buyers actually use. How to choose them has 45 examples.
• 24. The list covers several types of question. Category ("best X for Y"), alternatives ("X alternatives"), comparisons ("X vs Y"), problems ("how do I...") and use cases.
• 25. The wording of each question is fixed. You don't rephrase questions between checks, so a change in the answer is a change in the engine, not in you.
• 26. You check each engine separately. ChatGPT, Gemini and Claude often disagree. A blended number can hide a gap on one engine.
• 27. You've recorded a baseline. Your score, the questions where you're missing, the rivals named instead and the pages cited, written down before you start changing things.
• 28. You track rivals, not only yourself. You know which competitors are named for your questions and how their share of answers is moving.
• 29. Your numbers handle noise properly. Failed answers are left out rather than counted as misses, your rank in each answer is recorded, and you look at trends over weeks rather than single days. Why answers change explains why this matters.
• 30. Someone reviews the numbers on a schedule. A weekly look at what moved, and a judgement on each change after four to eight weeks. One big change at a time, so you can tell what worked.

The order to work through it

If you're one person with a few hours a week, this order works for most small teams:

1. Week 1: checks 1 to 5 (crawlers) and 23 to 27 (questions and baseline). Fast, and everything else depends on them.
2. Weeks 2 to 3: checks 6 to 12. Fix your homepage copy and pricing page, and publish your first comparison page.
3. Weeks 3 to 6: checks 13 to 17. Find the cited pages, pitch the top articles, tidy the main review profile and ask for reviews.
4. Ongoing: checks 18 to 22 and 28 to 30. Community participation, consistent listings and a weekly review.

Write down the date you finish each step. When a number moves, you'll want to know what changed in the weeks before.

Common reasons a check fails

A few failures come up again and again:

• A staging robots.txt copied to production. "Disallow: /" for all user agents, left over from before launch, blocks every crawler.
• Pricing only behind a "contact us" form. If there's no number anywhere, answers about price will name rivals that publish one.
• Five different descriptions. The homepage says one thing, the review profile another and the social bio a third. Pick one.
• No baseline. The team made several changes in the same month and can't tell which one helped.

How Citemark covers the measurement part

Part 3 is what Citemark is built for. It puts your buyer questions to ChatGPT and Gemini (and Claude on Pro) on a schedule, and shows whether you're named, your rank, the rivals named instead and the sources cited, grouped by type. The score is a position-weighted share of answers naming you, with failed answers left out. You get alerts when you drop out of an answer or a rival enters one, a Fixes tab with up to three suggestions per question drawn from the cited pages, and a weekly email on Mondays.

You can see where you stand today with the free AI visibility check: 5 buyer questions on ChatGPT and Claude, about a minute, no signup. The AI SEO audit covers checks 1 to 3 in seconds.