Meta AI Visibility Checklist: What to Check and What Meta Documents

Meta publishes a crawler page that names its agents and what each one does. One line on it matters more than anything else for this checklist: "Allowing Meta-WebIndexer in your robots.txt file helps us cite and link to your content in Meta AI's responses."

That is unusual. OpenAI, Google, Anthropic and Perplexity describe what their crawlers do, but none of them says allowing a bot helps you get cited. Meta does, in writing, as of its crawler documentation reviewed for this page.

Here is what Meta does not tell you:

• How Meta AI ranks or selects sources once it has them.
• How often Meta AI searches the web for a given question, rather than answering from what the model already knows.
• Whether pages that render content with JavaScript are read in full by Meta-WebIndexer.
• Whether Google's and Bing's snippet controls (nosnippet, max-snippet, data-nosnippet) apply to Meta AI. Those are documented by Google and Bing, not by Meta.

So the honest scope is narrow. You control whether Meta's search crawler can reach and read your pages, and whether your facts are consistent. You do not control what the model says. For the background on each Meta agent, see our guide to the Meta-WebIndexer crawler.

robots.txt rules apply per user agent. Meta lists several agents, and they do different jobs:

• Meta-WebIndexer: the crawler tied to Meta AI search and citation. This is the one to allow if you want Meta AI to cite you.
• Meta-ExternalAgent: a separate crawler. Our evidence base groups it with training crawlers such as GPTBot and ClaudeBot. Allowing or blocking it is a separate decision from Meta-WebIndexer.
• Meta-ExternalFetcher: documented as potentially bypassing robots.txt, so a Disallow line is not a reliable way to stop it.

The most common failure is a blanket AI block copied from a template, which removes the search crawler along with the training ones. Open /robots.txt and look for three things: a User-agent: * group with Disallow: /, a list of AI bots that includes Meta-WebIndexer, or no explicit rule at all.

If you want Meta AI search access, an explicit group is clearest:

User-agent: Meta-WebIndexer
Allow: /

Then decide Meta-ExternalAgent on its own merits. Keeping training out and search in is a valid setup.

robots.txt is a request, not a gate. Your firewall, CDN or bot-management rules are the gate. A site can allow Meta-WebIndexer in robots.txt and still return a 403 or a challenge page to it. The crawler gets nothing, and no dashboard tells you. Operators document this mechanism. How often it happens has not been measured.

This risk changed recently. From 15 September 2026, Cloudflare set Training and Agent crawler categories to blocked by default on ad-displaying pages for new domains, new sites on existing accounts and all existing free-tier customers, while Search stays allowed by default. You may have lost access without editing anything. Check which category your CDN puts each Meta agent in, as of October 2026.

How to test:

1. Request a real content page using each Meta crawler's user-agent string.
2. Record the status code. A 200 with your content is a pass. A 403, 429, CAPTCHA or JavaScript challenge is a fail.
3. Confirm /robots.txt itself returns 200, since a blocked robots file stops the crawler reading your rules.
4. Look at your own server or CDN logs for verified Meta-WebIndexer requests and their status codes.

The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away. Re-run this after any redesign or hosting change, as covered in our site migration checklist for AI crawler access.

Two separate problems make Meta AI repeat wrong facts about a brand.

Facts that only appear after JavaScript runs

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but do not execute them. Independent re-testing in June 2026 found the same for Meta-ExternalAgent. Meta does not document whether Meta-WebIndexer renders JavaScript. The safe assumption is that it may not. View your page source with JavaScript off and confirm your prices, plan names, product names and one-line description are present.

Facts that disagree with each other

If your pricing page, footer, About page and schema block state different prices or founding years, you have given the engine contradictory sentences. Pick one value per fact and use it everywhere.

The only positive experimental signal for on-page content in our evidence base is this: stating a specific fact (a price, a number, a one-line definition) on a fetchable page gives the engine a sentence it can quote. That is different from rewriting pages in an "AI-friendly" style, which the evidence does not support. Run this check again whenever you change pricing or launch a product, using our product launch checklist for AI accuracy.

You need your own record of what Meta AI says about you, and it has to be disciplined to mean anything. Most AI visibility tools focus on other engines, a gap we cover in whether any tools track Meta AI.

Pick 5 to 10 questions a buyer would actually ask, such as "best accounting software for a three-person firm." Ask the same questions each month. Record for every answer:

• Date and the exact question text.
• Whether the answer showed sources or links. If not, it may have come from the model's built-in knowledge, which nothing you do this quarter changes.
• Whether your brand was named, and which sources were cited.
• Every fact stated about you, marked right, wrong or unverifiable.
• Competitors named.

Do not record a "position" or "rank." Retrieval is discretionary and per question on every platform, and research on how models cite (arXiv 2606.28358) shows citation output can be separated from which sources the model actually used. A citation count is a noisy signal, not a measure of influence.

Wrong facts are the most useful finding. They point to a page you can fix. Fold the log into a quarterly AI visibility review.

Some common advice has no evidence behind it, for Meta AI or anyone else:

• Schema or FAQ schema to get cited. Ahrefs' matched test found AI Overviews citations fell 4.6% after adding schema. It is not a lever for AI answers, though existing markup does no harm.
• An llms.txt file. No engine, Meta included, documents reading it.
• Content rewrites or chunking for AI. A 2026 survey of 45 GEO studies (arXiv 2607.14035) found no technique with a stable, cross-platform effect on being discovered.
• Hidden text or instructions aimed at AI crawlers. This is manipulation. It is illegitimate and a reputational risk.

The checklist stays short because the evidence is strong about access and weak about everything else. If you want a broader first pass, our free GEO audit follows the same order: access first, then accuracy.