To optimise for Meta AI, make your brand easy for it to identify and quote. Keep your Facebook and Instagram profiles complete, publish plain, crawlable pages that state what you do and who for, and earn consistent descriptions elsewhere on the web. Then test by hand, because no brand-facing Meta AI report turned up in the sources reviewed.
Key takeaways
- No brand-facing report for Meta AI answers turned up in the sources reviewed, so the only way to check visibility is a fixed prompt set, run by hand and logged.
- Meta AI was reported in 2024 to use Google and Bing results for some queries, so ordinary crawlability and indexing still matter. Confirm current behaviour by testing.
- Fix the profiles Meta already owns first: category, description, hours, contact details and replies, matched to your website.
- Meta's developer documentation names two crawlers with different jobs. Check your robots.txt, CDN rules and logs before assuming either is blocked or allowed.
- Noma by Nurdd does not track Meta AI. It tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, and most of the work below carries over to them.
What does visibility in Meta AI actually mean?
Meta AI is the assistant built into Meta's apps, such as Facebook, Instagram, WhatsApp and Messenger. A brand is visible when the assistant names it in an answer to a buyer question, describes it correctly, or links to its pages as a source. The general idea is covered in what AI visibility is.
The audience is large. Gradually, 2026 reports that Meta stated more than 1 billion monthly active Meta AI users on its Q3 2025 earnings call. That is a secondary report of a company statement, and usage counts for AI assistants are defined differently by each company, so treat it as a sign of scale rather than a count of people who see brand recommendations.
Visibility also depends on the surface: search boxes inside apps, chat threads and the standalone assistant may each behave differently. The test below logs the surface for every answer.
What is known and unknown about how Meta AI picks brands?
Little is confirmed in public. Separate what has a source from what is inference before you spend budget.
| Question | What is sourced | What is not confirmed |
|---|---|---|
| Does it use web search? | Search Engine Journal reported in 2024 that Meta's assistant chooses between Google and Bing results depending on the query. | How much of today's answers depend on those results, and whether the mix has changed. |
| Are sources shown? | Nothing sourced. | Whether each surface shows links, and when. Test each one. |
| Does Meta crawl the web itself? | Meta's developer documentation lists Meta-ExternalAgent, which crawls for training foundation AI models or indexing content directly. | Whether crawled pages decide which brands get named. |
| Do profiles and reviews count? | WebFX argues they do, in a scorecard it published. | No Meta documentation found states which profile fields influence answers. Treat it as a hypothesis. |
| Can brands see their data? | No brand-facing Meta AI report was found. | Whether Meta will release one. |
The Search Engine Journal report dates from the Llama 3 launch and is more than two years old. Use it as a reason to keep Google and Bing indexing healthy (Search Engine Journal, 2024), not as a description of Meta AI today. Botify's guide draws a similar line: Meta has relied on Bing for some topics while building its own crawler.
How do I get my brand recommended in Meta AI?
No one outside Meta can promise a recommendation. What you can do is remove the reasons an assistant would skip you: unclear identity, stale facts, pages it cannot read and no third-party corroboration. Work through the steps in this order.
Step 1: Fix the profiles Meta already owns
WebFX's scorecard lists the fields it expects to matter: complete profiles, accurate categories, current hours and contact details, and consistent branding. That is WebFX's reading, not a Meta statement, but the fixes cost almost nothing.
- Rewrite the Facebook page intro and Instagram bio to say what you are, who you serve and where, in one plain sentence.
- Match the business name, category, address, phone and hours to your website exactly.
- Turn on messaging, set a response time and write the automated reply as an accurate summary of your offer.
- Ask recent customers for reviews. Do not buy them.
Step 2: Make your website readable and indexable
Because Meta's assistant was reported to draw on Google and Bing, your pages need to be indexed there. Content that appears only after JavaScript runs may never be seen. The technical SEO for AI search checklist covers crawlability, rendering and indexing, and the Next.js indexed HTML guide shows how to test what a crawler fetches.
Run your key page through the free Noma AI SEO Checker. It shows whether ChatGPT, Claude, Perplexity and Google AI Mode can read the page and lists the top fixes across 30+ checks. It does not test Meta AI, but readable, well-structured pages help on any engine.
Step 3: Decide on Meta's crawlers deliberately
Meta documents two crawlers. Meta-ExternalAgent crawls the web for uses such as training foundation AI models or indexing content directly. Meta-ExternalFetcher fetches individual links at a user's request, and Meta's page says this crawler "may bypass robots.txt rules" (Meta for Developers, 2026). So a single Disallow line is not a complete decision: it addresses the first agent, and Meta says the second may not honour it.
- Open your robots.txt and search for meta-externalagent.
- Check your CDN or firewall bot rules, which often block unfamiliar agents even when robots.txt allows them.
- Search your server logs for meta-externalagent and meta-externalfetcher over the last 30 days to see what actually visits.
- Record your decision. The AI Crawler Policy Kit gives a structure for training bots versus search bots.
The free AI crawler checker tests the major AI bots against your site and generates a corrected robots.txt. Check its bot list for Meta agents before relying on it for them.
Step 4: Write pages that state facts in quotable form
Put a one-sentence definition of your brand at the top of your home and category pages, add dated pricing and comparison pages, and answer the follow-up questions buyers ask. Botify recommends mining question-style search queries for this. Use structured data where it describes the page honestly.
Step 5: Earn the same description off your site
If directories, review sites and press describe you in different ways, answers will be inconsistent. List where you appear, align the one-line description and fix outliers. For local brands the local AI visibility guide covers listing consistency.
Can I track my brand visibility in Meta AI?
Only manually, in the sources reviewed. Some vendors advertise Meta AI monitoring. Before you pay for one, ask it to show the raw answers, the Meta surface used and the run date, and check that a repeat run gives a comparable result. Noma by Nurdd does not track Meta AI, and a number you cannot trace to an answer is not one you can defend.
The manual method uses the same logic as the prompt-selection guide: a fixed set, repeated, with the denominator stated.
- Write 20 unbranded buyer prompts. Use questions a customer would type, such as "best [category] near [city]" or "which [product type] suits [use case]". Add 3 branded prompts to check accuracy.
- Pick the surfaces. Test the Meta surfaces your audience uses, such as in-app search, WhatsApp and the standalone assistant.
- Run each prompt three times on different days. Answers vary, so one run proves little. Use a fresh chat each time.
- Log every answer in the template below.
- Calculate mention rate. Mentions divided by total answers, with both numbers stated.
- Check accuracy. Compare every claim about you with your facts. The AI brand fact check walks through that.
A log template you can copy
One row per answer. The example rows are invented to show the format, not real data.
| Date | Surface | Prompt ID | Run | Named? | Position | Competitors named | Link shown | Exact words about you |
|---|---|---|---|---|---|---|---|---|
| 2026-11-12 | In-app search | P07 | 1 of 3 | Yes | 2nd of 4 | Brand B, Brand C, Brand D | None | "Brand A, a family-run bakery" |
| 2026-11-14 | In-app search | P07 | 2 of 3 | No | n/a | Brand B, Brand E | None | n/a |
Keep the prompt wording in a separate sheet keyed by ID so nobody edits it between rounds. Add a column for who ran it, since different accounts and locations can change answers.
A worked example of the arithmetic
This is a hypothetical to show the maths. Twenty prompts run three times gives 60 answers. If the brand is named in 9, the mention rate is 15%. Each answer is worth about 1.7 percentage points, so 9 versus 11 mentions (15% versus about 18%) is a two-answer difference. Treat changes this small with caution, and keep the prompt set unchanged between checks.
What can and cannot be verified?
| Claim you may hear | Can you verify it? | How |
|---|---|---|
| "Meta AI named us on 3 of 20 prompts" | Yes, for that surface and date | Screenshot and log each answer. |
| "Our reviews raised our Meta AI rank" | No | No ranking or position data is published for brands. |
| "Meta AI cites our pages" | Sometimes | Only where a link is shown. Record the URL. |
| "Blocking Meta's crawler removes us" | Not confirmed | Retest the prompt set after the change and compare. |
| "Meta AI reaches N users" | Only as a reported company statement | Gradually reports more than 1 billion monthly actives from Meta's Q3 2025 call; definitions vary. |
You can verify what you observed, not why it happened. That limit is the same on every engine, and it is why a fixed prompt set beats a one-off check. The AI Mode checker explainer makes the same point for Google.
How do I know the work changed anything?
Set the baseline before you edit. Change one group of things at a time, such as profiles first and pages second, and rerun the full prompt set two to four weeks later. Hold prompt wording constant. Record each edit with its date so any movement has a candidate cause.
Add two indirect checks. First, add a "how did you hear about us" option that includes AI assistants. Second, review messaging enquiries for customers who say an assistant sent them.
Look at your logs too. If a Meta user agent starts fetching pages after you open access, that is evidence of crawling, not of being recommended. Keep the two findings separate in your report. For the wider method, see how to measure AI share of voice across engines.
Where Noma fits
Noma by Nurdd is an AEO and AI visibility platform that asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions your buyers ask every day. It shows where your brand is named, which competitors are named instead and which pages the answers draw on. Meta AI is not one of the engines it tracks, so use Noma for the engines it covers and the manual log above for Meta.
See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged.
Why the same groundwork helps on other engines
Complete profiles, crawlable pages, answer-first copy and consistent third-party descriptions help brands on ChatGPT, Gemini, Perplexity and Claude as well. If hours are limited, do the shared work first, then add Meta-specific profile fixes. Compare the disciplines in AEO vs SEO vs GEO.
How we researched this
We reviewed Meta's developer documentation on its web crawlers, trade press on Meta AI's search integration, and vendor guides on Meta AI optimisation (WebFX, Botify). Competitor vendor claims were removed unless they could be confirmed. We ran no tests inside Meta AI. Facts were checked on 2026-11-11; claims we could not confirm were removed or marked as unconfirmed.
What to do next
This week, write your 20 prompts, run the baseline on each Meta surface and check your robots.txt and logs. Then fix profiles and pages and rerun. To track the 13 AI engines Noma covers every day, book a demo or start the free trial.
See what AI engines say about your brand
Free · no account · or start a 3-day trial
Questions people ask
How do I get my brand recommended in Meta AI?
Meta publishes no ranking factors, so no one can guarantee a recommendation. Complete your Facebook and Instagram profiles, keep facts identical across your website and listings, publish crawlable answer-first pages and earn consistent third-party descriptions. Then test with a fixed prompt set to see whether anything changed.
Can I track my brand visibility in Meta AI?
Only by hand in the sources reviewed. Run a fixed set of buyer prompts on each Meta surface, repeat them on different days and log whether you are named. Noma by Nurdd does not track Meta AI, so use it for the 13 AI engines it covers.
Does Meta AI use Google and Bing results?
Search Engine Journal reported in 2024 that Meta's assistant chooses between Google and Bing results depending on the query. That report is old, so confirm current behaviour by testing. Keeping your pages indexed in both search engines is low cost either way.
Should I block Meta-ExternalAgent in robots.txt?
It depends on whether you want your content used for Meta's AI training and indexing. Meta's documentation says Meta-ExternalAgent crawls for those uses and that Meta-ExternalFetcher, which fetches links at a user's request, may bypass robots.txt. Record your decision and retest your prompts afterwards.
How many people use Meta AI?
Gradually reports that Meta stated more than 1 billion monthly active Meta AI users on its Q3 2025 earnings call. That is a secondary report and the definition of an active user is Meta's. Describe the audience as large rather than building plans on one number.
How often should I retest Meta AI visibility?
Take a baseline, then rerun the same prompt set two to four weeks after each batch of changes. Run each prompt several times on different days because answers vary. Report mentions over total answers, with both numbers shown.