Meta AI SEO: the technical guide to getting cited by Meta AI

How Meta AI answers with the web#

Meta AI answers from the model when it can and searches when a question needs something current or specific. Meta’s Search grounding documentation, written for developers using the same model family, says the model “skips the search when it can answer confidently from its training data” and that simple factual questions “typically do not trigger a search.” Since April 8, 2026 that model is Muse Spark, the first release from Meta Superintelligence Labs. It launched in the Meta AI app and on meta.ai, and on May 12, 2026 Meta said it was starting to bring it to Meta AI in WhatsApp, Instagram, Facebook, Messenger and Threads.

What it searches is less documented than on any other major assistant. Meta announced Bing in 2023 and Google in 2024, but its current terms name no engine. Alongside those partners sit Meta’s own crawler, Meta-WebIndexer, and a separate search over posts on Meta’s apps. The tool names below come from Meta’s API docs and from a tool list Meta AI returned to Simon Willison on launch day, which a Meta engineer confirmed belongs to the new harness.

1. 1

   Meta AI decides whether to search

   News, sports, prices, local and product questions search; stable knowledge comes from the model. Web search is a setting users can turn off; in incognito chats it “begins toggled on.”

   Your lever: Decide on Meta-ExternalAgent deliberately: it collects training data, and unsearched answers come from what the model learned.

2. 2

   It writes a main query plus alternatives

   browser.search takes a primary_query, a list of alternative_queries, an optional since date, and verticals for news, sports, weather, finance and local. For incognito chats, Meta’s help center says the query goes to “select search engines without connecting it to your account.”

   Your lever: Title pages the way people ask — product names, places, and the year where freshness matters.

3. 3

   Partners and Meta's own index return results

   Willison’s tool dump describes the engine as undisclosed. Meta-WebIndexer exists “to improve Meta AI search result quality,” and Meta ties it directly to citations and links.

   Your lever: Allow Meta-WebIndexer, and keep ranking in Google and Bing, the partners Meta has used before.

4. 4

   Meta's own posts are searched too

   meta_1p.content_search runs semantic search over Facebook, Instagram and Threads posts from 2025 onward, and shopping questions can query Meta’s product catalog and Marketplace.

   Your lever: Post on your own accounts and earn public posts from customers; both are retrievable.

5. 5

   It opens and reads the pages it needs

   browser.open loads a full result and browser.find pattern-matches inside it. Meta doesn’t document which user agent does the reading or whether it runs JavaScript.

   Your lever: Put the answer in the raw HTML, near the top, and don’t block Meta-ExternalFetcher at your WAF.

6. 6

   The answer cites a subset of what it read

   Links appear in the response, with a Sources list under it. In the API, the retrieved results are “every source the model considered” and the citations “the subset it actually cited.”

   Your lever: State each fact in one self-contained sentence a link can attach to.

Meta's five crawlers and robots.txt#

Meta documents five crawlers, each identified by a lowercase token such as meta-webindexer/1.1. Meta says it prefers “industry-standard practices like robots.txt rather than non-standard formats like NoAI tags,” and that its crawlers may cache robots.txt “for up to 24 hours.” There’s no FacebookBot on the current page, so an old rule targeting it no longer maps to a documented Meta crawler.

A robots.txt that keeps you in Meta AI search and in link previews while opting out of training looks like this. Named groups override a User-agent: * block, and the rules must be repeated on every subdomain:

# Meta AI search index: allow
User-agent: meta-webindexer
Allow: /
 
# User-requested fetches (may bypass robots.txt anyway)
User-agent: meta-externalfetcher
Allow: /
 
# Link previews in Meta's apps
User-agent: facebookexternalhit
Allow: /
 
# Model training: your call
User-agent: meta-externalagent
Disallow: /
 
# Ads and business products: your call
User-agent: meta-externalads
Disallow: /

Meta’s crawlers are also heavy. Fastly attributed 52% of the AI crawler traffic it saw from mid-April to mid-July 2025 to Meta — more than Google (23%) or OpenAI (20%). Rate-limit if you must, but a 429 to meta-webindexer is a page Meta AI can’t cite.

Meta’s page calls allowlisting by IP address “more secure” than by user agent, but the current version publishes no ranges. Meta’s own network is AS32934; Meta doesn’t document that every crawler request comes from it, but it’s the first check when a meta-* user agent looks spoofed:

# Routes announced by Meta's network (AS32934)
whois -h whois.radb.net -- '-i origin AS32934' | grep ^route | head

Where Meta AI's results come from#

Meta AI draws on four pools. You can influence all four, but each has a different door:

Search partners: named once, unnamed now

Microsoft announced in September 2023 that Bing would give Meta AI “more timely and up-to-date answers.” In April 2024 Meta added Google, though Search Engine Journal “could only get Meta AI to search using Bing” in its tests, and Ray-Ban Meta’s release notes described glasses search as “powered in part by Bing.” Since then Meta has stopped naming names. Its AI terms, effective May 13, 2026, say only that Meta “may share certain information with select partners, like search engines.” You can’t optimize for one partner, so being indexed and ranking in both Google and Bing remains the broadest route into the candidate pool.

Meta's own index is growing fast

The Information reported in October 2024 that Meta was building a search engine that crawls the web, to reduce its reliance on Google and Bing (Social Media Today’s summary). Meta-WebIndexer is the visible result. In Promptwatch’s logs it sat near 2.2% of AI crawler requests in mid-July 2026, then became the heaviest AI crawler it tracks, at 37.8% on August 9. Days earlier, Pieter Levels posted that Meta staff told him Meta is “ALLEGEDLY” building its own index so its AI searches don’t end up at Google; Search Engine Roundtable’s report carried no comment from Meta. Treat it as unconfirmed — but the crawl volume is real.

Licensed news gets links

Meta’s first AI news deal, with Reuters in October 2024, came with each such answer expected to “include a link to the Reuters story on which it’s based” (SiliconANGLE). In December 2025 Meta added CNN, Fox News, Fox Sports, Le Monde Group, People Inc., The Daily Caller, The Washington Examiner and USA Today; TechCrunch reported those responses “will also include links to articles.” For a publisher, a deal is the documented route to linked news answers.

Meta's own platforms are a second index

This is what makes Meta AI different. Its tool list includes a “semantic search across Instagram, Threads, and Facebook posts,” built from captions, visual analysis and transcripts, with “Data coverage: posts since 2025-01-01.” At the Muse Spark launch Meta promised “Reels, photos, and posts woven directly into your answers, with credit back to the content creators,” and local answers showing “public posts from locals who know the area.” In June 2026 Facebook added AI Mode, a search tab with answers “grounded in what people are saying publicly across our apps like in Groups and Reels.” Shopping mode searches Facebook Marketplace “alongside options from across the internet.”

Answers inside WhatsApp, Instagram, Facebook and glasses#

Meta AI isn’t one interface. The same assistant answers in a standalone app, a website, three social apps, a messenger and a pair of glasses. Meta’s help center documents one constant: Meta AI “may use sources and links from the internet to inform its responses,” and you can review them by selecting Sources under the response. Beyond that, Meta publishes little about how each surface lays citations out.

Two details matter for visibility. Incognito chats, launched on WhatsApp and the Meta AI app in Q2 2026, still search — Meta’s help center says web search in them “begins toggled on” — so conversations even Meta can’t see are part of your audience. On glasses, the answer is heard, not read: since April 2025 the Meta AI app has been the glasses’ companion app, where a conversation started on glasses can be picked up from the history tab. Meta doesn’t document whether sources appear there for spoken answers.

Technical requirements#

Meta publishes no rendering, freshness or structured-data guidance for Meta AI. What’s documented is crawler access, link-preview requirements for facebookexternalhit, and how the search tool behaves in the API. The rest is inference, marked as such.

Meta-WebIndexer allowedRequired

In robots.txt and at the WAF. It’s the one agent Meta ties to citing and linking your content in Meta AI.

Content in the raw HTMLRequired

Undocumented for Meta-WebIndexer, but Vercel saw Meta-ExternalAgent fetch pages without running JavaScript. Treat client-rendered text as invisible — see server-side rendering.

Meta-ExternalFetcher not WAF-blockedHelps

It fetches at a user’s request and may bypass robots.txt anyway. A WAF block only means the answer is built from someone else’s page.

Fast, error-free responsesHelps

Meta’s crawl volume makes aggressive rate limits tempting. A 429, 5xx or timeout to a Meta agent removes that page from the answer.

The answer near the topHelps

The tool pattern-matches inside opened pages and passes extracted snippets to the model. A clear sentence early in the page is the easiest thing to find.

Open Graph in the first 1 MBHelps

Documented for facebookexternalhit: OG tags before the first 1 MB, gzip and deflate support, and a response “within a few seconds.” It governs your preview card whenever the URL is shared in Meta’s apps.

Structured dataUnconfirmed

No statement from Meta. The API hands the model extracted page text, so keep every fact in visible copy, not only in JSON-LD.

Linked social profilesUnconfirmed

Meta AI’s post search filters by author. sameAs links from your Organization schema to your Facebook, Instagram and Threads accounts can’t hurt; Meta doesn’t document entity matching.

SitemapsUnconfirmed

Not documented for Meta-WebIndexer. Keep one with honest lastmod values, referenced from robots.txt.

llms.txtUnconfirmed

Meta publishes llms.txt files for its own developer docs, but hasn’t said its crawlers read other sites’ files.

IndexNowNo effect

Meta isn’t on IndexNow’s list of participating engines. It still reaches Bing, a past Meta AI partner.

NoAI meta tagsNo effect

Meta says it uses robots.txt “rather than non-standard formats like NoAI tags.”

Test the two agents that matter with Meta’s own documented user-agent strings. The second command is the crawler simulation from Meta’s page:

# Does Meta's search crawler get your real content?
curl -s -A "meta-webindexer/1.1 (+/documentation/sharing/webmasters/web-crawlers)" \
  https://yoursite.com/pricing | grep -c "Plans start at"
# 0 = text rendered by JavaScript, or a WAF block
 
# Meta's link-preview simulation
curl -v --compressed -H "Range: bytes=0-524288" -H "Connection: close" \
  -A "facebookexternalhit/1.1 (+http://www.facebook.com/externalhit_uatext.php)" "$URL"

What Meta AI cites#

Independent citation research on Meta AI barely exists. The large AI-referral panels — SE Ranking’s (101,574 sites, June 2026) and Similarweb’s (September 2026) — don’t break it out at all. The signals below lean on Meta’s own documentation, and say so where they’re our read.

• Crawl permission

  Official

  Meta says allowing Meta-WebIndexer “helps us cite and link to your content in Meta AI’s responses” — the only ranking-adjacent statement it has made.

• Public talk on Meta's apps

  Official

  Meta says local, trending and shopping answers draw on public posts and creators across its apps, with “credit back to the content creators.” Facebook’s AI Mode answers from what people say publicly.

• Licensed news

  Official

  Partner publishers’ articles are linked in news answers by agreement: Reuters, CNN, Fox News, USA Today, Le Monde and others.

• Pages that match the rewrite

  Our read

  Each search sends a main query plus alternatives, with an optional date filter. Pages whose titles and opening lines match those phrasings, with a real date, are easier to retrieve.

• Classic search rankings

  Our read

  With partners unnamed and a history of Bing and Google, strong rankings in both are the broadest way into the candidate pool.

• A trust gap to close

  Observed

  In the Reuters Institute’s six-country survey (Oct 2025), 12% trusted Meta AI against 29% for ChatGPT, and more people distrusted it than trusted it.

The practical upshot: Meta AI rewards the same answer-first, dated pages as other engines, plus something only Meta can search — brand mentions in public posts on Meta’s own apps. A brand nobody discusses on Facebook, Instagram or Threads is missing from the part of Meta AI’s retrieval no other engine has.

Measuring Meta AI#

Meta publishes no webmaster console for Meta AI and doesn’t document the referrer its answer links carry. GA4’s AI Assistant channel names ChatGPT, Gemini, DeepSeek, Copilot and Grok in Google’s channel documentation — not Meta AI. Our read: clicks from the meta.ai website should show meta.ai, while links opened inside Facebook can pass through Meta’s link shim, which reports as l.facebook.com on desktop and lm.facebook.com on mobile — mixed in with ordinary social clicks. Give Meta AI a custom channel, placed above Referral:

# Session source — Meta AI website
(^|\.)meta\.ai$
 
# Meta link shims — Meta AI mixed with ordinary Facebook clicks
^(l|lm|m)\.facebook\.com$

Your logs are more reliable. meta-webindexer hits show what Meta is indexing; meta-externalfetcher hits are pages fetched for a person’s request — the closest thing to a live-demand signal Meta gives you:

# Meta agent hits by type
grep -oiE "meta-webindexer|meta-externalfetcher|meta-externalagent|meta-externalads|facebookexternalhit" access.log \
  | tr 'A-Z' 'a-z' | sort | uniq -c
 
# Pages Meta fetched on a user's request
grep -i "meta-externalfetcher" access.log | awk '{print $7}' | sort | uniq -c | sort -rn | head -20

Finally, test retrieval directly. Meta’s Model API offers a web_search tool whose internals carry the same browser.search, browser.open and browser.find names as the consumer app’s tools, and it can return every result the model considered alongside the subset it cited. Run your target questions through it on a schedule — with the caveat that the app’s prompts, personalization and partners may differ:

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.meta.ai/v1", api_key=os.environ["MODEL_API_KEY"])
 
r = client.responses.create(
    model="muse-spark-1.3",
    input="What's the best project management tool for a five-person team?",
    tools=[{"type": "web_search"}],
    include=["web_search_call.results"],
)
 
# web_search_call.results = every URL retrieved
# url_citation annotations on output_text = the URLs cited

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Meta AI SEO: frequently asked questions#