Meta AI Visibility Factors, Graded by Evidence Strength

Meta documents its crawlers. It does not document how Meta AI picks sources. That gap shapes every grade below.

The pattern matches other engines. The levers that hold up are access levers, and they work by removing blocks. If you want the full explanation of Meta's crawler set first, start with the Meta-WebIndexer guide.

Meta's web crawlers page (dated 21 May 2026) says: "Allowing Meta-WebIndexer in your robots.txt file helps us cite and link to your content in Meta AI's responses."

That sentence is unusual. OpenAI, Google, Anthropic and Perplexity document what their search crawlers do, but none of them says that allowing a crawler helps citation. Meta does.

Read it carefully, though. "Helps" is not a number. Meta publishes no effect size, and no independent study in the evidence base has tested a before and after. It is an assertion from the operator, which is the best evidence available for Meta AI and still not a measurement.

The check is simple. Your robots.txt should not disallow Meta-WebIndexer, either by name or through a blanket rule. A minimal explicit allow looks like this:

User-agent: Meta-WebIndexer
Allow: /

The common failure is a blanket AI block copied from a template that catches retrieval bots along with training bots. For more on what the statement does and does not promise, see whether allowing Meta-WebIndexer helps Meta AI cite you.

Meta runs more than one crawler. Each user agent needs its own decision.

• Meta-WebIndexer. Meta AI search indexing. This is the token tied to citation.
• Meta-ExternalAgent. A training crawler, grouped in the evidence base with GPTBot, ClaudeBot and Google-Extended. It is the only channel with a theoretical path into answers Meta AI gives without live retrieval, and that path takes a full model generation.
• Meta-ExternalFetcher. Meta documents it as potentially bypassing robots.txt. A Disallow line for it is not a reliable block.

What this means in practice: blocking Meta-ExternalAgent to keep content out of training is a legitimate choice, but nobody, including the vendors, has measured whether it changes how Meta AI describes your brand. The honest label is unmeasured, not harmless and not harmful.

The same split exists at OpenAI, where GPTBot governs training and OAI-SearchBot governs search. Our page on what OpenAI discloses about ChatGPT covers that parallel.

Two gates sit outside robots.txt. Meta's page in the evidence base does not describe them, but the mechanisms are documented by other operators and apply to any crawler.

Network reachability

A firewall, CDN or bot-management rule can return a 403 or a challenge page to a crawler that robots.txt allows. OpenAI's help centre states eligibility needs both robots.txt permission and a host or CDN that lets its searchbot traffic through. The same logic applies to Meta. How often this happens is not measured anywhere.

As of October 2026, one dated change matters. From 15 September 2026, Cloudflare blocks Training and Agent crawler categories by default on ad-displaying pages for new domains, new sites on existing accounts and existing free-tier customers, while Search stays allowed. Which category each Meta crawler falls into is not stated in the evidence base, so check your own Cloudflare settings rather than assuming.

Rendering

Vercel's crawler study found the ChatGPT and Claude crawlers "don't execute" JavaScript. An independent re-test in June 2026 found the same for Meta-ExternalAgent. Meta-WebIndexer was not in that test set. Until someone tests it, the safe default is that prices, product names and key claims appear in the HTML before JavaScript runs.

The AI Crawler Checker on 99Visibility requests a page as each AI crawler and shows which ones the firewall turns away: run the AI Crawler Checker.

The list of unknowns is longer than the list of knowns.

• Ranking or selection factors. Meta publishes none in the evidence reviewed.
• Snippet controls. Google and Bing document nosnippet, max-snippet and data-nosnippet as ways to keep content out of AI answers. The evidence base has no equivalent documentation from Meta.
• Knowledge cutoff. The evidence base records cutoffs for OpenAI models, not for Meta AI.
• Retrieval behaviour. When Meta AI searches the web, and what queries it runs, is not documented.
• Measurement. As of the August 2026 research pass, no vendor tracked Meta AI at all. See whether any AI visibility tools track Meta AI.

Two cross-platform findings still apply as general context. Lab studies, including the 2026 critical survey of 45 GEO studies, found topical relevance and position in the context list are the most reproducible drivers of citation among documents already retrieved. Brand prominence shows a strong association with citation on every platform studied. Neither is a Meta-specific finding, and neither is a lever you can pull this quarter. Treat any study claiming otherwise with the caution in correlation vs causation in AI search studies.

Schema, FAQ schema, llms.txt, E-E-A-T scores and domain authority do not appear in anything Meta documents. No evidence supports them as Meta AI levers.

1. Read your robots.txt. Confirm nothing disallows Meta-WebIndexer. Decide on Meta-ExternalAgent separately, as a training choice.
2. Test the network gate. Request a key page with each Meta user agent and look for 200, 403 or a challenge page. If you use Cloudflare, check the post-September 2026 defaults.
3. Diff raw and rendered HTML. List the facts that only appear after JavaScript runs, then move them server-side.
4. State specific facts plainly. A price, a number or a one-line definition on a fetchable page gives any engine a sentence it can quote. It is the only positive experimental signal for on-page content, and it is not a promise of citation.
5. Don't buy Meta AI rank tracking. No position data exists to check such a number against.

What stays outside your control: whether Meta AI retrieves your page for a given question, and what it says when it does. The same access-first approach applies on Perplexity.

Does blocking Meta-ExternalAgent remove me from Meta AI answers? The evidence does not say. Meta-ExternalAgent is a training crawler, and Meta ties citation to Meta-WebIndexer instead. Blocking training use is a legitimate choice. Whether it changes how Meta AI describes your brand in answers given without live retrieval has not been measured by anyone, including the vendors, so treat the effect as unknown rather than zero.

Will schema markup or an llms.txt file help Meta AI cite me? Nothing Meta documents mentions either. No engine documents reading llms.txt, and Google says its Search ignores it. Ahrefs' matched-control test found adding schema produced no major uplift in AI citations on any platform it tracked. Existing markup does no harm, but it is not a lever for Meta AI answers.

Is robots.txt enough to keep Meta's crawlers out? Not for every crawler. Meta documents Meta-ExternalFetcher as potentially bypassing robots.txt, so a Disallow line for it is not a reliable block. If you need hard exclusion, enforce it at the server, firewall or CDN layer. Expect that a firewall rule applied broadly can also block Meta-WebIndexer, the crawler tied to citation.

Does Meta AI favour big brands? There is no Meta-specific data in the evidence base. Across the platforms that have been studied, a small set of large sources takes a large share of citations, and brand size is the main uncontrolled confounder in correlational studies. That is an association, not a lever. It does not tell you how Meta AI treats a smaller brand that is fully reachable.

How do I check whether Meta's crawler can reach my site? Check two things. First, parse robots.txt for any rule that disallows Meta-WebIndexer by name or with a wildcard. Second, request a real page using Meta's crawler user agent and confirm a 200 response rather than a 403 or a challenge page. The [free Meta AI crawler check](/free/free-meta-ai-crawler-check) walks through both.