How to Appear in Meta AI and Get Your Brand Cited

Meta AI is the assistant Meta builds into its apps. Since 8 April 2026 it runs on Muse Spark, the first model from Meta Superintelligence Labs, which Meta announced as powering the Meta AI app and meta.ai on that date. In its update of 12 May 2026, Meta wrote that it was starting to bring the same model to Meta AI across WhatsApp, Instagram, Facebook, Messenger and Threads.

The audience behind those apps is the reason a company should care. On its earnings call of 29 July 2026, Meta reported 3.6 billion people using at least one of its apps each day, and said that the number of people interacting with Meta AI each day had risen 60% since it rebuilt the assistant on Muse Spark.

There is no switch to flip in that definition. Meta AI decides which brands to name when it writes the answer, and a company influences that decision only through what is published about it, on its website, on its own accounts and on other people’s pages.

WhatsApp is where people engage with Meta AI the most, according to what Mark Zuckerberg said on Meta’s earnings call of July 2026. For a company selling to the public, that means a buyer may ask about it inside the same chat app they use to message friends.

Meta AI is not the same thing as a company’s own chatbot on WhatsApp. Meta’s Business Agents, which more than 1 million businesses were using every week as of the July 2026 call, answer a company’s own customers in its own chats. Setting one up does not make Meta AI recommend the company to someone who has never contacted it.

The training route is fed by Meta-ExternalAgent. Meta describes it as crawling the web for uses such as training foundation AI models or improving products by indexing content directly. What the model knows about a company without searching comes from pages collected this way, mostly written by people other than the company.

The search route is fed by Meta-WebIndexer. In the same documentation, Meta says that “Allowing Meta-WebIndexer in your robots.txt file helps us cite and link to your content in Meta AI’s responses.” That is the robot that decides whether Meta AI can link a company’s pages at all.

Meta AI also reads what is posted on Meta’s own apps. Meta’s Muse Spark announcement says the assistant can show public posts from locals about a place and answer what people are talking about with content and community posts, so public posts on Instagram, Facebook and Threads are a source in their own right.

Read across the rows, the table shows why a fix on the website moves only part of Meta AI. Rewriting pages helps the web index, but the model’s memory changes only with the next training run, and the posts route depends on the company’s accounts. How each assistant decides which sources to cite is covered surface by surface in how AI models choose their sources.

Meta’s crawler documentation, updated on 21 May 2026, describes four robots that matter here:

• Meta-WebIndexer improves the quality of Meta AI’s search results; allowing it helps Meta cite and link a site in Meta AI’s answers.
• Meta-ExternalAgent crawls for uses such as training foundation AI models or improving products by indexing content directly; it has its own line in robots.txt.
• Meta-ExternalFetcher fetches individual links at a person’s request, and Meta warns that it may bypass robots.txt rules.
• FacebookExternalHit reads a page when its link is shared on Facebook, Instagram or Messenger, to build the preview with title, description and image.

The most common mistake is not in robots.txt at all. A bot-protection rule at the firewall or at the CDN, the layer that serves a site before the server does, can stop Meta’s robots even when robots.txt allows them, and nobody notices because people browsing the site see nothing wrong. Checking whether AI robots can reach the site, Meta-ExternalAgent among them, shows whether pages answer at both layers.

Meta’s original announcement of 8 April 2026 says shopping mode draws on the styling inspiration and brand storytelling already happening across Meta’s apps. A company that keeps its Instagram and Facebook profiles empty, or describes itself differently on each, gives the assistant little to work with in exactly the place where it looks first.

Consistency is the practical rule. The same company name, the same one-line description of what it does and for whom, and the same website address on the Instagram profile, the Facebook page and the website make it easier for Meta AI to recognise that the three are one brand.

Each robot answers a different question. Meta-WebIndexer’s visits show which pages are entering the index Meta AI searches. Meta-ExternalFetcher’s visits, made at a person’s request, are the closest signal to “Meta AI read this page during a conversation”. Meta-ExternalAgent’s visits speak only about training.

The robots are identified by the name they give when they request a page, their user agent, and, where the provider publishes its addresses, verified by IP so that an impostor using the same name is not counted. This is what DoGeo sets up for each client, with a monthly count per robot and per page.

What the logs do not show is what Meta AI writes. Whether it names the company, how it describes it and which page it links can only be followed by asking Meta AI the same questions every month, in WhatsApp and on meta.ai, and recording the answers.