Blog
Meta Muse for Brands: A 50-Site Readiness Test
Meta Muse is Meta’s personal AI agent, launched in the US on September 8, 2026 and in Canada on September 18. It manages email, books dinners and buys from Shopify stores on the user’s behalf. On October 1, 2026 we checked the robots.txt files of 50 health and low-ticket websites in the US and Spain: none of the 42 we could read named Meta-WebIndexer, the crawler Meta says helps it cite and link websites in Meta AI. Brands that want Meta’s agents to read them should make that decision on purpose, starting with robots.txt.
Muse is not a place where brands get recommended yet. It is an agent that already acts: it books, it buys, it works with the health data people choose to share. Our view is that recommendations come next, first for low-ticket purchases and healthcare. This post covers what Muse is, what it asks users for, where it already touches buying decisions, and what our test found about whether websites are ready for it.
Key findings
- No site has decided about Meta’s AI search crawler. None of the 42 robots.txt files we could read mention Meta-WebIndexer. The 41 that allow it do so by default, because they don’t name it.
- Three sites block Meta’s training crawler, Meta-ExternalAgent: the Spanish online pharmacies DosFarma and Atida, which block it along with GPTBot, ClaudeBot, Google-Extended and CCBot, and OpenTable, which blocks all bots by default.
- Meta is missing from the guest lists. Five sites name search crawlers from OpenAI, Anthropic or Perplexity on purpose. Only one of them, Ulta, names a Meta crawler.
- OpenTable allows 31 named bots and none of Meta’s three AI crawlers, even though booking dinners is one of the tasks Muse advertises.
- 4 of 15 US healthcare sites stopped a regular Chrome browser (CVS, Hims, Ro and Aspen Dental) with an “Access Denied” page or a security check.
What is Meta Muse?
Meta Muse is a personal AI agent from Meta. It launched on September 8, 2026 in the US on iOS, Android and muse.ai, and reached Canada on September 18. Unlike a chatbot, it acts for you: its App Store listing says it manages email, books dinners, tracks budgets and finds deals, and checks with you before it spends or cancels anything.
Adoption has been fast. Muse reached 5 million US downloads in 22 days, against 56 days for ChatGPT, according to Sensor Tower data reported by Forbes. The caveat: Meta put up to half of its daily house ad inventory behind the launch. Downloads are not the same as trust.
Meta has not given a date for Europe or the UK. On September 29 it also launched Muse for Small Business in the US and Canada, an agent that helps owners with campaigns, email and finances. It is a management tool, not a channel where brands get recommended.
What data does Muse ask for?
A lot, although the user chooses what to link. To be useful, Muse connects to email, calendar, Instagram, Facebook, health and fitness data and finances. In its App Store privacy details, Meta declares that the app may collect data linked to the user’s identity, including health and fitness data, purchases, financial info, location, contacts, browsing history and “sensitive info”. It may also use location when the app is not open.
Whether Muse can read private messages is disputed:
- Inc. columnist Jason Aten reported that Muse synced the Messages database on his Mac without his permission.
- Meta denies it: David Singleton of Meta Superintelligence Labs says the Messages integration is opt-in and takes three separate permission steps.
We have not found independent verification either way.
Where Muse already touches buying decisions
Two categories stand out, because Muse already acts in them.
Low-ticket retail and services. By low-ticket we mean inexpensive, frequent purchases: beauty, fashion, sports gear, food delivery, restaurant bookings. Shopify added Meta as a channel in its Agentic Storefronts program on Muse’s launch day and shares merchants’ catalogs with Meta by default (Practical Ecommerce). A Shopify product can appear in Muse, and Muse can complete the purchase with the user’s consent. Amazon went the other way and blocked Muse as an unauthorized agent. When the largest marketplace closes the door, brands that sell direct get more room.
Healthcare. “Health & Fitness” is one of the sections of Muse’s App Store listing: workouts, sleep and wellness, built from the data the user shares. On launch day, HealthEx released a connector that lets US users bring their medical records into Muse, with approval required before each action (HIT Consultant). And in April, Meta presented its Muse Spark model as able to reason through complex questions in science, math and health.
In Europe the bar is higher: health data is a special category under the GDPR. In healthcare, privacy is not a detail. It is the price of entry.
Our test: do health and low-ticket sites let Meta’s AI crawlers in?
An agent can only cite or recommend what it can read. Muse only works in the US and Canada, so we could not run prompts on it. What we could measure is the step before: whether websites let Meta’s AI crawlers in.
Meta’s three AI crawlers
Meta documents three crawlers related to its AI:
| Crawler | What Meta says it does | Follows robots.txt |
|---|---|---|
| Meta-WebIndexer | Crawls to improve Meta AI search results. Meta says allowing it helps it cite and link your content in Meta AI’s responses. | Yes |
| Meta-ExternalAgent | Crawls to train AI models or to index content directly. | Yes |
| Meta-ExternalFetcher | Fetches specific links when a user asks, including to help AI navigate websites and complete tasks. | Not always: it may bypass robots.txt |
Method
On October 1, 2026, between 9:15 and 9:35 UTC, we opened the robots.txt files of 50 websites in Chrome: 25 in the US and 25 in Spain. In each country, 15 are healthcare sites (online pharmacies, dental chains, hospitals, insurers, opticians, doctor booking, online therapy, labs) and 10 are low-ticket retail and services. For each crawler we applied the rule that governs it: its own User-agent group if the file names it, otherwise the * group. We counted a crawler as blocked when that rule disallows the whole site. We could read 42 of the 50 files.
Results
| Group | Sites read | Name Meta-WebIndexer | Allow Meta-WebIndexer | Block Meta-ExternalAgent |
|---|---|---|---|---|
| Healthcare, US | 10 of 15 | 0 | 10 | 0 |
| Low-ticket, US | 10 of 10 | 0 | 9 | 1 |
| Healthcare, Spain | 13 of 15 | 0 | 13 | 2 |
| Low-ticket, Spain | 9 of 10 | 0 | 9 | 0 |
| Total | 42 of 50 | 0 | 41 | 3 |
Allowed by omission, not by choice. The 41 sites that let Meta-WebIndexer in do so because their rules don’t mention it. The only one that keeps it out, OpenTable, does it with a block on every bot it doesn’t list.
The sites that decide treat Meta like everyone else. DosFarma and Atida block Meta-ExternalAgent along with GPTBot, ClaudeBot, Google-Extended and CCBot, the training crawlers, and leave the search crawlers open. It is a consistent “no to training, yes to showing up”, and Meta gets the same treatment as the rest.
On the guest lists, Meta is missing. Five sites name search crawlers from OpenAI, Anthropic or Perplexity on purpose: Topdoctors, Hawkers, Zocdoc, OpenTable and Ulta. Only Ulta names a Meta crawler, and it is Meta-ExternalAgent, not the search one.
The OpenTable case
OpenTable’s robots.txt, generated on September 30, 2026, blocks all bots by default and then allows 31 by name, including ChatGPT-User, OAI-SearchBot, Claude-SearchBot and PerplexityBot. From Meta it only admits two veteran bots, Facebot and FacebookExternalHit. None of its three AI crawlers is on the list.
OpenTable’s robots.txt, October 1, 2026. It blocks everyone by default and then allows a list that does not include Meta-WebIndexer, Meta-ExternalAgent or Meta-ExternalFetcher.
Booking dinners is one of the tasks Muse advertises. We don’t know whether Meta has a deal with OpenTable, and Meta-ExternalFetcher may bypass robots.txt when a user asks Muse to open a specific page. But the crawler Meta uses to cite and link content in Meta AI cannot read OpenTable, while the search crawlers of ChatGPT, Claude and Perplexity can.
The sites we couldn’t read
Of the 8 sites that didn’t load, 5 are US healthcare sites, and 4 of them (CVS, Hims, Ro and Aspen Dental) showed a regular Chrome browser an “Access Denied” page or a security check. An agent that books and buys for users has to be able to open your pages. If an anti-bot wall stops a real browser, check what it does to an agent.
Limits and sample
robots.txt says who may come in. It does not say who actually does, or whom Muse recommends. This is one morning’s measurement on 50 sites, not an industry study. The sites we checked (* = could not be read):
- Healthcare, US: CVS*, Walgreens, GoodRx, Zocdoc, Hims*, Ro*, Teladoc Health, One Medical, Aspen Dental*, Warby Parker, 1-800 Contacts, BetterHelp, Talkspace, Quest Diagnostics*, Labcorp.
- Low-ticket, US: Allbirds, Glossier, Chewy, DoorDash, OpenTable, Sephora, Ulta, Target, Gymshark, Liquid Death.
- Healthcare, Spain: Mifarma*, DosFarma, PromoFarma, Atida, Vitaldent, Sanitas, Quirónsalud, HM Hospitales, SegurCaixa Adeslas, DKV, General Óptica, MultiÓpticas, Doctoralia, Topdoctors, ifeel*.
- Low-ticket, Spain: Druni, Primor, Notino, TheFork, Just Eat, Decathlon, PcComponentes, Hawkers, Mr. Wonderful*, Singularu.
The founder’s view: where Muse goes next
This section is my opinion, not something Meta has announced.
I think Muse will start recommending businesses soon. Today it is a consumer agent for everyday tasks, but an agent that already books the table and completes the purchase is close to choosing the restaurant and the brand. Once people trust it, the businesses it suggests will become their first choice.
Its appeal is to mass consumers, which is why I expect it to matter most for low-ticket goods and services. I would also go as far as saying it will be quite relevant in healthcare, for the reasons above. A business version is coming too: Muse for Small Business is the first step, even if today it manages campaigns and email rather than recommending anyone.
The brake is privacy. Muse needs your calendar to be useful, and the claim that it could read your messages, unverified as it is, has not gone away. In healthcare, and in Europe, that will decide how far it goes.
We will run real prompt tests on Muse as soon as we can do it from a market where it works, and publish the results here. If it starts naming brands, it becomes one more engine to track next to ChatGPT, Gemini and Perplexity.
What should a health or low-ticket brand do now?
-
Decide about Meta on purpose. If your robots.txt already names OpenAI, Anthropic or Perplexity, add Meta-WebIndexer, or leave it out knowing why. A “yes to showing up, no to training” policy looks like this:
User-agent: Meta-WebIndexer Allow: / User-agent: Meta-ExternalAgent Disallow: /A named group replaces your
*rules for that crawler, so repeat any Disallow lines (checkout, account pages) you still want it to follow. -
Check Shopify if you sell in the US. Shopify shares your catalog with Meta by default. You can change it in Shopify’s agentic settings.
-
Test your anti-bot protection on product, booking and appointment pages, the ones an agent would open.
-
In healthcare, be careful about what you publish and what you ask for. Clear information on services, prices and practitioners, with no sensitive data in plain view.
-
Work on the engines people already use. Muse is new. ChatGPT, Gemini and Perplexity already recommend brands every day, and the same groundwork applies. If you sell online, start with our ecommerce playbook or with how brand mentions in ChatGPT work.
Frequently asked questions
Is Meta Muse available outside the US?
Only in Canada so far. Meta launched Muse in the US on September 8, 2026 and in Canada on September 18. It has not given a date for Europe or the UK, where health data and combining data across Meta’s services face stricter rules.
Can Meta Muse read your messages?
It is disputed. Jason Aten, a columnist at Inc., says Muse synced the Messages on his Mac without permission. Meta denies it and says the integration is opt-in and takes three permission steps. We have not found independent verification.
Does Muse recommend brands, and how does it choose them?
It can surface products and complete purchases, for example from Shopify stores, but Meta has not published how Muse chooses businesses or products. What is known is where it can get information: Shopify’s catalog, shared with Meta by default, the apps the user links (email, calendar, Instagram, Facebook, health, finances) and the websites it reads. That is why your website needs to be readable by Meta’s crawlers and your catalog needs to be where Muse shops.
How do I allow or block Meta’s AI crawlers?
In robots.txt, with a User-agent group for Meta-WebIndexer (Meta AI search) and another for Meta-ExternalAgent (training and indexing). Meta-ExternalFetcher may ignore robots.txt when a user asks it to open a link. In our October 1, 2026 test, none of the 42 websites we could read mentioned Meta-WebIndexer.