Meta AI is the engine almost nobody works and the one your customers already use without deciding to. It's Meta's AI — built on its Llama model — and its oddity isn't size, though that too: it passed a billion monthly users. It's where it lives. You don't talk to it on a separate site: you talk to it inside WhatsApp, Instagram, Facebook and Messenger, in the search bar and chats you open every day. So «showing up in Meta AI» isn't one surface. And when it searches live, it doesn't crawl the web with its own index: it leans on Google and Bing at once. It answers two ways. Here's the map.
Why Meta AI isn't a destination: it's already in the apps your customers open
Every other engine you've worked takes a deliberate act: you open ChatGPT, you go to Perplexity, you head into Gemini. Meta AI doesn't. It's embedded by default in the apps people already have open — WhatsApp, Instagram, Facebook, Messenger — and it shows up in the search bar and the chats without anyone going looking. Per public usage data, its assistant splits like this: around 380 million monthly users on WhatsApp, 290 in Instagram search, 240 on Facebook and 110 on Messenger. The consequence for you is unlike any sibling: it's the assistant your least technical customers use, the ones who'd never open an AI app on their own.
Translated to your work: in Meta AI you're not competing to be an expert's pick, you're competing to be in the answer someone gets when they just wanted to ask something inside their usual chat. That audience is huge and almost nobody is working it.
The two ways Meta AI answers (and how to get into each)
Meta AI answers in two ways, and each is worked differently.
- By default, from what it knows. Like ChatGPT, its first deck is the knowledge Llama learned in training. There's no live link here: either you're in what the model learned, or you're not. You get in by being a recognizable entity corroborated off your own site — press, Wikipedia, reviews, forums — so the model learned you well.
- With live search, from the web. When the question needs fresh data, Meta AI retrieves real-time web results and cites them. Here you do compete to be retrievable: answer up front, updated content, sourced data. The difference from the rest is where it pulls those results from, and that's next.
Meta doesn't document when it draws on one deck or the other, so some of your answers will come from the trained model even though search exists. Work both.
The odd lever: Meta AI doesn't crawl with its own index, it borrows Google and Bing
Here's what makes Meta AI unique and the lever almost nobody uses. When it searches live, it doesn't retrieve from its own index like Perplexity or from Google's like Gemini: it leans on Google and Bing results at once. Meta struck deals with both to feed its assistant's answers with real search. That changes your game entirely: your best lever to show up in Meta AI isn't some new trick, it's your classic organic ranking on Google and Bing. If you rank well on both, you're already feeding Meta AI's search deck without touching anything else.
The levers that actually get you into Meta AI's answers
With the two decks and the borrowed indexes in mind, the game is five things.
- Rank on Google and Bing. It's lever number one here, because Meta AI's search deck feeds on both. Don't neglect Bing for being small: in Meta AI it weighs as much as Google.
- Be a clear, corroborated entity. It feeds Llama's trained layer, the one that answers without searching. Make who you are and what you do read the same on and off your site.
- Answer up front, with data and quotes. Front-load the answer in the first lines of each section and back it. It's what lifts visibility most and what clips cleanly when the engine cites.
- Mind your social surface. Meta AI lives inside Instagram and Facebook; a clear profile, substantive content and brand consistency are signal in the ecosystem where the assistant operates. It's not like X in Grok, but being well described on Meta's own platforms doesn't hurt.
- Let its retrieval crawler in. Check that your robots.txt doesn't block Meta-ExternalFetcher, the one that fetches pages when the assistant needs them. That's a separate decision from Meta-ExternalAgent, the training one.
Measure it and check your robots.txt: two named crawlers
Like every other engine, Meta AI gives you no citation dashboard. The method is the usual one: define 15 to 25 questions your customers would ask it, run them once a month — from WhatsApp or Instagram, where it lives — and note whether you show up in the text or among the sources. The trend over months is what counts.
And the technical detail worth controlling: Meta documents two named crawlers, controlled separately in your robots.txt. Meta-ExternalAgent collects content to train Llama; Meta-ExternalFetcher goes to fetch a page live when the assistant needs it to answer. The trap is the usual one: blocking the training one thinking you're protecting your content can shut out the retrieval one, the very one that gets you into the answer. Check which you allow. And remember: even if you close both, you can still show up via Google and Bing, because that deck doesn't depend on your crawlers.