Meta AI Is 32% of Classified Traffic. GPTBot Is About 2%.

Robauto Team · Published Sep 24, 2026 · Last updated Sep 24, 2026

Robauto’s Sep 24 engine mix: Meta AI 32.1%, ChatGPT 2.3%. Meta External Agent leads visits as other_bot — use Meta’s separate robots.txt tokens, not one “AI bot” switch.

Most AEO playbooks still treat GPTBot as the center of gravity. Robauto’s live engine mix on September 24, 2026 says otherwise: Meta AI accounts for 32.1% of classified bot visits across the network, while ChatGPT sits at 2.3%. That is not a rounding error. It is a different map of who is actually reading your pages.

The headline AI share of page views is 20.89% over 90 days (264,884 of 1,267,777 views). Inside the classified bot pool, only the llm_crawler class — 143,039 visits, 54% of bot traffic — counts as AI search alongside ai_referral (currently 0%). That is about 11.3% of total page views. The rest of the headline is classic SEO crawlers, headless automation, and other non-search agents — including Meta’s own External Agent on the visit leaderboard.

Engine mix: Meta AI dwarfs ChatGPT and Claude

Robauto’s engine breakdown ranks every classified visitor by share of the 264,884-visit pool. Meta AI is first. ChatGPT is eleventh. Claude is tenth. Microsoft Copilot is ninth. If your robots.txt and content budget only name OpenAI tokens, you are optimizing for a minority of the observed mix.

Meta also shows up as smaller sibling engines in the same mix: meta_externalads at 1.0% and meta_webindexer at 0.9%. Those are not duplicates of “Meta AI” in the ranking above — they are separate keys in Robauto’s engine table, which matches how Meta documents distinct crawler tokens.

Meta’s robots.txt tokens are not one bot

Meta’s developer docs (updated May 21, 2026) list separate crawlers with different jobs. Treating “Meta” as a single Allow/Disallow is how publishers accidentally block link previews, ads quality crawls, or Meta AI citations while still letting training traffic through — or the reverse.

Cloudflare’s September 15 Search / Training / Agent split is the policy-layer version of the same idea: one “AI bot” switch is too blunt. Meta’s documented tokens already give you per-purpose levers if you use them.

The visit leaderboard is mostly not llm_crawler

Robauto’s public Agent Leaderboard ranks by visits, not by “how AI” a visitor is. On September 24, the top of that board is dominated by Meta External Agent (other_bot) and classic SEO crawlers. The first clear llm_crawler in the top five is Bingbot (Copilot) at rank 5. GPTBot (OpenAI) is rank 10 with 724 visits in 7 days.

That does not make Ahrefs or Semrush “AI search.” It means raw visit rankings mix SEO ops traffic with model and agent traffic. Robauto’s taxonomy note is explicit: only llm_crawler + ai_referral are AI search; search_bot and headless are classic crawlers and automation.

What to do this week

1. Name Meta tokens separately in robots.txt — at minimum meta-webindexer (Meta AI citations) vs meta-externalagent (training / direct indexing). Do not assume one Disallow covers the family.
2. Stop treating GPTBot volume as your AEO north star. On this network snapshot, ChatGPT is 2.3% of classified engines and GPTBot is 10th on 7-day visits.
3. Report two KPIs: headline non-human share if you must, and a strict AI-search line that keeps only llm_crawler (+ ai_referral). Today that compresses 20.89% down to about 11.3% of page views.
4. On Cloudflare, map Training vs Agent preferences to the Meta tokens that match each job — Disallow AI Training is not the same decision as allowing Meta AI search citations.
5. Read the leaderboard with bot_class on. Rank 1 visits can be other_bot; that is still useful operations data, not proof you are “winning AEO.”

{
  "as_of": "2026-09-24T15:00:08.742+00:00",
  "network": {
    "total_page_views_90d": 1267777,
    "ai_agent_views_90d": 264884,
    "headline_ai_traffic_pct": 20.89,
    "llm_crawler_visits": 143039,
    "llm_crawler_share_of_bot_traffic_pct": 54,
    "approx_llm_crawler_share_of_page_views_pct": 11.3,
    "ai_referral_visits": 0
  },
  "engine_highlights": {
    "Meta AI": {"visits": 85051, "share_pct": 32.1},
    "ChatGPT": {"visits": 6090, "share_pct": 2.3},
    "Claude": {"visits": 8543, "share_pct": 3.2},
    "Microsoft Copilot": {"visits": 9646, "share_pct": 3.6}
  },
  "leaderboard_highlights_7d": {
    "1_Meta_External_Agent": {"bot_class": "other_bot", "visits_7d": 10585},
    "10_GPTBot_OpenAI": {"bot_class": "llm_crawler", "visits_7d": 724}
  },
  "sources": [
    "https://robauto.ai/ai-search-data",
    "https://robauto.ai/agent-leaderboard",
    "https://developers.facebook.com/docs/sharing/webmasters/web-crawlers/",
    "https://blog.cloudflare.com/accountable-mixed-use-ai-crawlers/"
  ]
}

What this means for builders

If your content and crawl policy were written for a GPTBot-first web, update the map. On Robauto’s September 24 snapshot, Meta AI is 32.1% of the classified engine mix and ChatGPT is 2.3%. Meta External Agent leads the visit leaderboard as other_bot, while GPTBot is tenth. Split Meta’s tokens the way Meta documents them, split your KPIs the way Robauto classifies them, and optimize for citation, training refusal, or live agent action — not a single bot name that no longer describes the traffic.

Sources

• Robauto AI Search Data
• Robauto Agent Leaderboard
• Meta Web Crawlers documentation
• Cloudflare: Disallow AI Training / Search-Training-Agent controls

Where can an agent get this data live?

Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure. A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp, callable with no key. Setup and the full tool list are in the Robauto MCP directory.

All blog posts