How We Add Meta AI to a Client's AI Visibility Tracking

Until this month, Meta AI sat outside our standard scope for a simple reason: Promptwatch, the platform we run client programs on, didn't track it. As of its October 28, 2026 update it does, through Meta AI's real interface rather than a model API. Meanwhile Meta's own documentation now describes a crawler whose job is to help Meta AI cite websites.

So Meta AI is now part of our standard scope for the right clients. This is how we add it.

Step 1: decide if the client needs it

We don't add every engine to every client. Each one costs response budget and reporting attention, and a line in a report that never moves teaches the client to skip the report.

Meta's help center says the Meta AI assistant runs in the Meta AI app and on meta.ai, and inside Facebook, Instagram, Messenger and WhatsApp, with availability varying by country (Meta Help Center). That tells us who it's for. We add Meta AI by default for consumer clients whose customers already talk to them on Meta's apps: retail, hospitality, travel, food, beauty, local services. We add it on request for B2B clients, usually after a sales team hears a prospect mention it.

If a client already runs paid social or WhatsApp customer service, that's usually a sign. Their buyers are in those apps, asking questions.

Step 2: check the crawlers before tracking anything

Tracking Meta AI before checking access produces a month of zeros and an awkward call. So the first week is technical.

Meta publishes what each of its crawlers does (Meta web crawlers). We check the client's robots.txt and firewall for four of them:

• Meta-WebIndexer, which Meta says improves Meta AI search results and "helps us cite and link to your content in Meta AI's responses"
• Meta-ExternalAgent, for training foundation models or improving products by indexing content
• Meta-ExternalFetcher, which fetches links at a user's request and may bypass robots.txt
• FacebookExternalHit, which builds link previews when someone shares the client's URL

The problem we look for is a blanket rule: someone blocked "Meta" or "facebook" to stop training scrapers, and in doing so blocked the one crawler tied to citations. We walk the client through the trade-off: they can allow Meta-WebIndexer and still refuse Meta-ExternalAgent if they want, but that second rule isn't a clean training opt-out, since Meta describes ExternalAgent as also indexing content for its products. It's the client's call. We make sure it's a decision and not an accident.

Then we look at the logs. On Promptwatch, Agent Analytics records Meta-WebIndexer, Meta-ExternalAgent and Meta-ExternalFetcher, checked against Meta's published IP ranges so fake user agents don't count. Every Promptwatch agency plan includes crawler logs, so this is on for every client in our workspace. What we want to see in week one is Meta-WebIndexer actually fetching the pages that matter, not just the homepage. We wrote up how we explain the scale of Meta's crawling to clients in our Meta-WebIndexer brief.

Step 3: adapt the prompt set

We don't copy the ChatGPT prompt list wholesale. Meta AI questions come from inside a chat or social app, so we write some shorter, more casual versions of the client's core category questions, and add the "is this brand any good" and "where can I buy" phrasing people type into a messaging app.

The structure stays the same as every client program: category prompts without the brand, comparison prompts against named rivals, and a small set of brand prompts to catch inaccurate answers. Category prompts get most of the budget, because that's where a buyer who hasn't heard of the client can still be pointed to them.

Language and country follow the client's market. Promptwatch runs country-level tracking on every paid plan and state or city on agency plans. The exact-coordinate local targeting it shipped this month works for ChatGPT, AI Overviews and AI Mode only, so for a multi-location client Meta AI runs at city level instead.

Step 4: add it to the monitor and budget the responses

On Promptwatch, models are chosen per monitor and there's no cap on how many, only on responses. So adding Meta AI is a setting, not a new project. The cost is volume: every prompt now runs once more per day.

Agency plans have unlimited prompts and projects but a shared response pool (10,000 on Kick-off, 25,000 on Growth, 65,000 on Scale). Before we switch Meta AI on for a client, we work out prompts times engines times days and check it fits the pool alongside everyone else's. If it doesn't, we trim the prompt set rather than drop an engine the client cares about.

Step 5: report it like the other engines

Meta AI gets its own line in the monthly report, next to ChatGPT, Perplexity and Google's AI surfaces. We report three things for it:

1. How often the client is named across the Meta AI prompt set, and the trend
2. Which pages Meta AI cites for the prompts that matter, theirs or a competitor's
3. Whether Meta-WebIndexer reached those pages, from the crawler logs

The third line is what makes the first two useful. If a competitor's comparison page keeps getting cited and the client's equivalent page was never crawled, the next action is obvious, and it isn't "write more content".

We use the Action Board in Promptwatch to turn those gaps into tasks. Its Discovery agent reviews the data weekly and suggests up to five actions with the evidence attached, and we assign them to the client's team or ours.

What we don't promise

We don't promise a client they'll show up in Meta AI by a date. Nobody controls that, and an agency that says otherwise is guessing. What we commit to is the process: the crawlers can reach the pages, the prompts reflect what buyers ask, the answers are recorded daily, and every gap in the report has a next step attached.

If you want us to look at how Meta AI talks about your brand today, get in touch. And if you'd rather run it yourself, Promptwatch is the platform we'd start you on. Meta AI is on every paid plan.