Meta's Muse Is Coming for Your Contact Form. Don't Lock It Out.

Within two years, a real share of the quote requests and bookings a local business gets will be filed by a customer's AI assistant. Meta's Muse is the clearest sign yet. It launched September 8 as a personal agent that can open a browser and fill out forms for the person using it, and within two weeks Amazon locked it out, saying Meta never asked for permission, Fortune reported.

Amazon can afford to turn agents away. A roofer or an HVAC shop can't. When a customer asks an assistant for quotes from three roofers, it tries each contact form the way the customer would, and the shop whose form fails never hears about the job. Our call for local service businesses: let agents in, and verify what they send.

TL;DR

• Meta's Muse is an AI agent that fills out forms and makes purchases on ordinary websites, and Sensor Tower projects it passed 5 million downloads in about three weeks.
• Amazon and Insurify blocked Muse in September, the same month Shopify and Stripe announced agent-ready checkouts and OpenAI launched always-on agents of its own.
• Our bet: by the end of 2027, assistants will file a real share of the quote and booking requests local businesses get, so let them in, verify what they send, and test your own form this week.

What is Meta's Muse?

Key fact: Sensor Tower projected that Muse passed 5 million downloads on September 30, about three weeks after launch. By the firm's count, ChatGPT took 56 days to reach that mark (9to5Mac).

Muse is a separate app from the Meta AI app (TechCrunch), and it's free for most uses in the US and Canada. Ask it to find patio furniture or order dinner, and it clicks around websites the way a person would in its own browser, then reports back in a chat. Meta says it "checks with the person before sensitive actions" like making a purchase.

On September 29, Meta added Muse for Small Business, which connects the agent to an owner's Facebook and Instagram business accounts and to tools like QuickBooks and Shopify. Meta says nothing goes out or gets spent without the owner's approval.

What did OpenAI launch at DevDay?

At DevDay on September 29, OpenAI said ChatGPT reaches more than 1.2 billion weekly users and introduced Dots, "always-on agents" that each get their own cloud computer and can connect to over 4,000 apps. They're available on ChatGPT's Pro and Business Premium plans. OpenAI also released GPT-6.1 Sol, which it says nearly matches GPT-6 Astra, the model behind Dots, on computer use at a fifth of the price. That makes agents cheaper to run.

If you're looking for ChatGPT Atlas, OpenAI's AI browser, it shut down on August 9, and its agent browsing moved into ChatGPT itself (Techzine).

Trades and home services should watch a beta Get Quote button that OpenAI's developer docs say can appear on a local business's card in ChatGPT, opening a partner's quote-request form inside the chat. It's being tested with approved partners and only shows when the business works with one.

Why are Amazon and Insurify blocking Muse?

By September 20, Muse users shopping on Amazon were hitting a pop-up that said "continued access by an unauthorized AI agent violates Amazon's Conditions of Use" (Gizmodo). Amazon said Meta didn't warn it and that the agent doesn't identify itself while shopping.

Insurify, an online insurance marketplace, blocked it too. Co-CEO Giorgos Zacharia tried Muse after a few customer complaints and found it easy to use, but it botched quotes and clicked on the site's ads. He told the Boston Globe it's "not ready for prime time."

Not everyone is blocking. Wayfair told the Globe it's testing agent shopping with guardrails, and Muse can order from restaurants that use Toast's online ordering.

Palo Alto Networks CEO Nikesh Arora wrote on X that every app selling goods or services will need to "existentially decide" whether to open up to consumer agents (Fortune). That includes a two-truck HVAC company, even if nobody calls its website an app.

What changed at checkout?

Agentic commerce, where an agent buys on a customer's behalf, now reaches the checkout pages small stores already use. Stripe said on September 22 that every Stripe Checkout page, used by over 7.8 million businesses, lets agents finish through built-in actions instead of reading the screen. Shopify did the same for its checkout and Shop Pay on September 28, with the order placed after the buyer authorizes it.

If you sell on Shopify, check one setting. Google has been emailing merchants that native checkout in AI Mode and Gemini is on for eligible products (Search Engine Roundtable). If you'd rather buyers finish on your own site, Google says to turn it off in your Shopify admin under Sales channels > Agentic.

Can an AI agent complete your contact form?

Find out before a customer's agent does. This week:

1. Ask Muse or ChatGPT's agent to request a quote or book a time on your site, using your own name and a note that it's a test.
2. Watch where it stalls: a CAPTCHA, a pop-up, a date picker, or a step that needs a phone call.
3. Read the lead that arrives as your office would, and check that it reached the right inbox.
4. Run the same request on a competitor's site.

Then decide on purpose, and tell whoever answers your leads. For most local service businesses, our call is the second row: route them.

If you block, Cloudflare now lets every customer, Free plan included, block agents as their own category of AI traffic. If you welcome them, it helps to know what agents look for when they read a website, which we covered last week.

How can you tell an agent's lead from a person's?

Not reliably, from the submission alone. An agent that doesn't announce itself sends the same fields a person would.

So ask, with that optional checkbox, and confirm. Call or text about the job and the address before you send a truck or schedule a showing. Then tag those leads so you can see in a month whether they close as well as the rest.

What are the risks?

Agents make mistakes, as Insurify found. For a local business, a wrong address or service in an agent's request costs a callback or a wasted trip.

Agents also hold a lot of access. Two weeks after launch, Meta patched a flaw in Muse's Mac app that a security researcher showed could let an attacker with code already on the computer hijack the agent (Gizmodo). Meta's David Singleton called the practical risk "quite low." Weigh it before you connect any agent to your books or your store.

You won't always know who's calling, either. Meta tested having call-center workers place some Muse calls to businesses, 404 Media reported, before rolling that back.

Blocking has a cost too. If a customer asks an agent to book the first plumber who can come Tuesday and your form turns it away, the job goes to the next site on the list.

Where is this headed?

We think the blocking phase will be short. Here's what we expect next.

• By the end of 2027, assistants will file a real share of the quote and booking requests local businesses get. Muse passed a projected 5 million downloads in about three weeks, and ChatGPT already reaches more than 1.2 billion people a week. When that many people have an assistant, it becomes a channel.
• Big platforms will trade blocks for rules. Amazon's complaint was that Muse didn't ask and didn't identify itself. Both are fixable, so expect agents that announce themselves and sites that set terms for what they're allowed to do.
• A form that needs a phone call to finish will look as dated as a fax number. The shops that let an assistant book a slot at 10 p.m. will win jobs the others never see.
• Agent-made leads will get their own line in the CRM. The businesses tracking them now will know what those leads are worth before their competitors do.

Testing forms and setting up a path for agent leads is part of the AI and automation work we do, if you'd like a hand.