Meta AI crossed one billion monthly users in 2025, and it still holds just 0.1% of the dedicated chatbot market, according to First Page Sage's September 2026 tracking. That gap explains why most brands leave Meta AI out of their AI Share of Voice checklist entirely. It's also why the timing here matters: Meta-WebIndexer, the crawler behind Meta AI's citations, jumped from roughly 2.2% of tracked AI crawler requests in mid-July 2026 to 37.8% by August 9, a seventeen-fold increase in under a month, according to AI-visibility platform Promptwatch. Here's what changed, and how to check whether your own site can be cited.
What Meta AI actually is, and why 0.1% market share is misleading
Meta AI isn't a single app. It's an assistant folded into Facebook, Instagram, WhatsApp, and Messenger, plus a standalone app and website, and First Page Sage's September 2026 market-share report puts its share of dedicated chatbot sessions at 0.1%, ninth among the platforms it tracks, behind ChatGPT at 51.5%, Gemini at 27.6%, and Claude at 10.2%. That number counts active chatbot queries the same way it counts every other assistant on the list, and on that narrow basis Meta AI barely registers next to the market leaders.
The same report calls Meta's own usage figure “unverifiable as a standalone chatbot,” since it blends passive social-feed interactions with active queries, and classifies Meta AI as a social-platform-embedded assistant rather than a destination app. That's a fair caveat, but it doesn't make the underlying reach fictional. Mark Zuckerberg said in May 2025 that Meta AI had surpassed one billion monthly active users across the company's family of apps, and Microsoft Copilot's 1.3% chatbot share carries a similar mismatch with its roughly 218 million active users inside Windows and Microsoft 365. A low measured share and a large embedded audience can both be true of the same assistant, and tracking tools built around dedicated chatbot apps will keep missing that second number.
Meta AI now runs on Muse Spark, not Llama
The model behind Meta AI changed in 2026. Meta Superintelligence Labs, the unit Meta formed to pursue what it calls personal superintelligence, introduced Muse Spark in April 2026 as the first model in a new series, distinct from the Llama family that had powered the assistant since its 2023 launch. Meta says Muse Spark now runs Meta AI across WhatsApp, Instagram, Facebook, Messenger, AI glasses, and the standalone app and website, replacing Llama as the model users actually talk to.
A July 2026 update, Muse Spark 1.1, added a Thinking mode inside the Meta AI app built for agentic tasks, tool use, and coding, and Meta describes it as including built-in search with citations. The company also opened a public preview of a Meta Model API so outside developers can call Muse Spark through an OpenAI-compatible package, a sign Meta wants the model treated as infrastructure, not just a consumer feature. What Meta hasn't published is a retrieval architecture the way OpenAI documents how its own retrieval tool chunks content by default. That gap is the practical problem for brands: Meta AI's citation behavior has to be inferred from crawler logs and robots.txt guidance, not read from a spec sheet.
Three different Meta crawlers, three different jobs
Meta operates three separate crawlers, and mixing them up is the most common mistake brands make when trying to manage AI visibility. Meta-ExternalAgent indexes content directly to train and improve Meta's models, similar to how GPTBot handles training crawls for OpenAI. Meta-WebIndexer builds the search index that feeds Meta AI's live answers. Meta-ExternalFetcher retrieves a specific page mid-conversation, when a user asks the assistant to look at a link, and Meta's own documentation notes that this fetcher, like the older FacebookExternalHit preview bot, may bypass robots.txt entirely for certain user-initiated requests and security checks.
Allowing Meta-WebIndexer in your robots.txt file helps us cite and link to your content in Meta AI's responses.
That's a direct line from Meta's own developer documentation, and it draws a sharp distinction most robots.txt files don't make. Disallowing Meta-ExternalAgent keeps content out of model training without touching whether Meta AI can cite it. Disallowing Meta-WebIndexer removes the one crawler Meta explicitly ties to citing and linking a site in its answers, and a general “block all AI bots” rule often does exactly that by accident, the kind of gap this blog's bot-by-bot breakdown of what robots.txt actually stops was built to catch across every major engine, not just Meta's.
None of that matters if the traffic claiming to be Meta-WebIndexer isn't really Meta's. Research into AI-crawler impersonation has already turned up hundreds of IP addresses spoofing other major AI bots to hunt for exposed credentials, and a well-known crawler name is exactly what a scraper would fake to slip past a server that only checks the user agent string. Verify a hit against Meta's published autonomous system, AS32934, and its listed IPv4 and IPv6 ranges before trusting the request at all, the same discipline that separates a genuine citation crawler from a scraper wearing its name.
Why Meta-WebIndexer's traffic jumped seventeen-fold in a month
Meta-WebIndexer's share of tracked AI crawler requests moved from about 2.2% in mid-July 2026 to 37.8% by August 9, according to Promptwatch's published daily data, a swing that dwarfs anything ChatGPT's or Perplexity's crawlers did over the same stretch. Promptwatch, which says it monitors more than 26 billion data points drawn from prompt responses and citations across major AI platforms, measures the share against every AI crawler user agent it tracks, not against ordinary search bots like Googlebot or link-preview fetchers such as FacebookExternalHit. That framing matters, because it means the jump reflects Meta specifically pulling more weight inside the AI-crawler mix, not just a general rise in bot traffic across the web.
The timing lines up with two other 2026 moves. Meta signed commercial data agreements with CNN, Fox News, Fox Sports, Le Monde Group, several People Inc. titles, The Daily Caller, The Washington Examiner, and USA Today in December 2025, letting Meta AI surface real-time news with working links back to the publisher rather than a summary with nowhere to click through. Meta has also been building its own search index since at least late 2024, when reports first surfaced that the company wanted to cut its reliance on Google and Bing for the assistant's answers. Meta-WebIndexer's growth looks like that plan reaching production scale: more of Meta AI's live answers pulled from a corpus Meta crawls, indexes, and controls directly, rather than licensed from a rival's search engine.
Meta AI still borrows from Google and Bing, and that's changing
Meta AI has drawn on both Google and Bing for real-time results since the company first added web-grounded answers on Llama 3 in April 2024, choosing between the two engines depending on the query rather than relying on one exclusively. That dependency hasn't been publicly retired even as Meta-WebIndexer scales up, so ordinary search visibility with Google and Bing still feeds part of what Meta AI can surface for a given question, alongside whatever Meta-WebIndexer has crawled and indexed on its own.
A brand that's invisible to Bing and has never explicitly allowed Meta-WebIndexer in robots.txt is working with neither on-ramp into Meta AI, which is worth checking directly rather than assuming a single AI-crawler policy covers both. Cloudflare's Content-Signal work already separates search indexing, real-time AI input, and model training into three distinct robots.txt permissions, and Meta's three crawlers map onto almost the same split: one for training, one for the live index, one for on-demand fetches. Treating all three as a single “AI bots” category is how brands end up blocking the crawler that actually earns them a citation.
How to check whether Meta AI can actually cite you
Start with robots.txt itself. Confirm Meta-WebIndexer specifically has access, since allowing FacebookExternalHit for link previews doesn't cover it, and neither does a rule aimed only at Meta-ExternalAgent. Then pull server logs for the meta-webindexer user agent and Meta's AS32934 IP ranges to see whether it's actually visiting, since Google Analytics won't show any of this traffic at all, the same blind spot that hides every AI crawler that never fires the JavaScript tag analytics platforms depend on.
Once you know Meta-WebIndexer is actually crawling a page, write that page to survive being lifted whole. Open each section with a direct, self-contained answer to the question it covers, the same framing that already raises citation odds in ChatGPT and Perplexity, rather than a paragraph that only gets to the point on the third sentence. Given that Meta hasn't disclosed how Muse Spark ranks or selects passages, an answer an assistant can lift whole is the safest bet available until that changes.
Where Meta AI belongs in your AI Share of Voice tracking
Most AI Share of Voice tools default to ChatGPT, Gemini, and Perplexity, and treat Meta AI as an afterthought if they track it at all, the same blind spot already flagged for Mistral and DeepSeek. A billion-user surface, even one First Page Sage calls unverifiable as a standalone chatbot, isn't one a brand tracking AI visibility can skip by default, especially once its own citation crawler is growing this fast.
Meta hasn't said whether Meta-WebIndexer's growth will keep pace with August's spike, but the crawler is still listed in the community-maintained ai.robots.txt registry as of September 2026, alongside the handful of other AI bots most sites already track.
Frequently Asked Questions
What is Meta-WebIndexer?
Meta-WebIndexer is the crawler Meta uses to build the search index behind Meta AI's live answers, separate from Meta-ExternalAgent, which trains Meta's models, and Meta-ExternalFetcher, which fetches individual pages during a conversation. Meta's own developer documentation says allowing it in robots.txt helps the company cite and link to a site's content in Meta AI's responses.
How many people use Meta AI?
Mark Zuckerberg said in May 2025 that Meta AI had surpassed one billion monthly active users across Facebook, Instagram, WhatsApp, and Messenger. First Page Sage's September 2026 tracking puts Meta AI at just 0.1% of dedicated chatbot sessions, and calls the billion-user figure unverifiable as a standalone chatbot metric since it blends passive feed activity with active queries.
Does Meta AI still rely on Google and Bing?
Partly. Meta first added real-time Google and Bing results to its assistant on Llama 3 in April 2024, and that arrangement hasn't been publicly retired. Since then, Meta has grown its own Meta-WebIndexer-powered index and signed real-time news agreements with publishers including CNN and Le Monde Group, shifting more retrieval toward sources Meta controls directly.
Should I block or allow Meta's AI crawlers?
Treat them separately. Blocking Meta-ExternalAgent keeps content out of Meta's model training without affecting citations. Blocking Meta-WebIndexer removes the one crawler Meta explicitly ties to citing and linking content in Meta AI's answers. Most brands trying to appear in Meta AI's responses should allow Meta-WebIndexer while making their own separate call on Meta-ExternalAgent.
Is Meta AI worth tracking in AI Share of Voice tools?
Yes. Meta-WebIndexer's share of tracked AI crawler requests grew roughly seventeen-fold between mid-July and early August 2026, according to Promptwatch, which signals Meta is investing heavily in the index behind Meta AI's answers. A billion-user surface that most tracking tools ignore by default is worth adding before a competitor closes that gap first.