How to get your business found by AI agents like Muse, Gemini Spark and ChatGPT

Until about a year ago, “being found online” meant ranking in Google and keeping your Google Business Profile tidy. That still matters. But a growing share of the people who might hire you or buy from you are no longer doing the searching themselves. They hand a goal to an agent, and the agent does the searching for them.

Meta’s Muse launched on September 8, 2026 and was the number one free iPhone app in the US within ten days. In May, Google announced Gemini Spark, an agent that runs on Google’s servers around the clock and acts while the owner’s phone is in a drawer. ChatGPT has been recommending products since late 2025, and for about six months it sold them directly inside the chat. More agents are on the way.

This guide covers what those agents actually check when they evaluate a business, and what you can do about it this month.

How an agent picks a business

An agent working on “book me a haircut Saturday morning near the office” doesn’t scroll a results page. It gathers candidates from wherever it can (web search, maps data, the platform’s own index, past reviews), and then it tries to answer a handful of questions about each one:

Is this a real, specific business, and can I confirm it’s the same one across the web, maps, and social profiles?
Can I read what it offers, what it costs, when it’s open and where it operates?
Can I complete the task, meaning book, order or pay, without a human filling out a form or making a call?
Is there evidence that dealing with this business goes well?

A business that fails any of those questions usually gets dropped from consideration before the agent’s owner ever sees a name, and that’s where agents differ from search. In search results you can rank fourth and still get clicks, but an agent either recommends you or it doesn’t.

Step one: make sure agents can reach and identify you

Start with the technical basics.

Check your robots.txt. Many sites block AI crawlers by default because a plugin or a hosting provider turned that on in 2024. The bots do different jobs. OAI-SearchBot is what feeds ChatGPT’s search and shopping results, while GPTBot is OpenAI’s training crawler, and you can allow one without the other. Google-Extended controls whether Gemini can use your content and has nothing to do with Google Search. If you’ve blocked the search and retrieval bots, you’ve opted out of what those agents can find. So decide deliberately, and if you want agent traffic, allow them. Training permissions are a separate decision.

Robots.txt is only half of it. Muse doesn’t identify itself with a crawler name at all. It runs a real browser on a Meta server and visits your site the way a person would, which means that robots.txt rules don’t govern it and your bot-protection settings do. An allow rule on its own doesn’t tell you whether Muse can get through. If your host or CDN has an aggressive bot-blocking mode switched on, or there’s a CAPTCHA in front of your booking or checkout page, Muse gets turned away and neither of you finds out. Amazon reportedly blocked it that way in September. Check your server logs for challenged traffic on those pages, and ask your developer or host what the bot rules actually do before you change them.

Then look at consistency. Your business name, address, phone, hours and website should match exactly across your website, Google Business Profile, Apple Maps, Yelp, Facebook and Instagram. Agents resolve entities the way a careful person would, and mismatches make them less confident that they’re looking at a single business.

Add an llms.txt file at the root of your site. It’s a plain text page that tells a language model what your business is and where the important pages are. It won’t move your search rankings, but it gives an agent a clean summary, so the agent doesn’t have to piece one together from your homepage. We wrote a separate post on whether you need one and what to put in it.

Step two: publish information a machine can read

This is the cheapest of the four steps to fix, so it’s a good place to start.

Agents rely on structured data (Schema.org markup in JSON-LD) to read the facts about a business without guessing. For a local business, that means a LocalBusiness or ProfessionalService entry with your address, service area, hours, phone and price range. Anything you sell online needs Product entries with real prices and availability, and services need Service and Offer entries. For the questions customers ask, add an FAQPage. We cover the seven schema types that matter in detail.

Beyond markup, look at how your policies are published. If your cancellation terms, return windows, deposit requirements or service areas live in a PDF, in an image, or only in the head of whoever answers the phone, an agent can’t act on them. Put them on a plain HTML page and mark them up.

The classic failure is a menu or price list posted as a photo. Agents can run OCR on it, but they’ll trust a text page with prices far more, and they’re right to.

Step three: give the agent a way to transact

Being found gets you considered, but an agent chooses a business it can actually book, order or pay with.

If you take bookings, the question is whether your booking system exposes real availability in a way that an agent can use, or whether it’s a widget that only works with a human clicking through. Booking platforms are adding agent-facing interfaces quickly, and some of them already work with the big agent platforms. Find out where yours stands.

If you sell products, the protocols matter. OpenAI and Stripe’s Agentic Commerce Protocol (ACP) is how ChatGPT reads a merchant’s catalog and hands off a purchase. Google’s Universal Commerce Protocol (UCP), published in January 2026 and built with Shopify, Etsy, Wayfair, Target and Walmart, uses a machine-readable profile at a well-known URL on your domain so that agents can check what you support before they send a customer. If you’re on Shopify, much of this is a setting, whereas on a custom platform it’s a project. Either way, we’ve written up the difference between ACP and UCP so you can have the conversation with your developer.

Even if you can’t support the protocols yet, make sure that the path from your product page to a completed order works without JavaScript tricks, popups, or a required account. An agent will keep trying, but it won’t improvise a way around an obstacle, and it will report a failed checkout as a failed business.

Step four: mind your reputation, including the machine-readable parts

Agents read reviews, although they read them differently from the way people do. A person skims the star rating and a few recent comments. An agent can read all of them, weigh recency, check whether the specifics in reviews match what you claim, and look at whether you respond to complaints.

The usual advice still holds: ask happy customers for reviews, respond to the bad ones, and keep your listings current. You should also expect a new category of review to appear over the next year, written by agents about their own experience trying to do business with you. Did the booking go through? Did the quoted price hold? We’re building a shared record of that kind of experience at FunnelPoint, and we explain why agents will want it.

What to do this week

If you only have a couple of hours, do these in order:

Open your robots.txt and decide, on purpose, whether AI crawlers are allowed.
Check that your name, address, phone and hours match across your website, Google, Apple Maps and Yelp.
Put your prices, hours and policies on plain HTML pages if they aren’t already.
Add LocalBusiness (or the right subtype) markup to your homepage.
Try to book or buy from yourself in a private browser window without logging in, and note every place where a machine would get stuck.

If you’d rather have someone do the looking for you, request a free assessment and we’ll check your business the way an agent would and send you a short report.