OpenAI Dots: How AI Agents Browse Online Stores and What Ecommerce Sites Should Fix

On 29 September 2026 OpenAI launched dots: always-on AI agents that each run on their own cloud computer, with their own browser and access to the apps a person connects. Three weeks earlier Meta launched Muse, a personal agent that can browse and, Meta says, check out. For a store this is a new kind of visitor. It does not announce itself, robots.txt does not reach it, and when the sale happens inside an AI app the analytics tag never fires.

This guide covers how dots reach a store, what an agent reads on a product page, what to fix first, and how to explain it in a client report without overclaiming.

What are OpenAI dots, and how is Meta's Muse different?

OpenAI dots are always-on personal AI agents introduced on 29 September 2026, powered by GPT-6 Astra and running on their own cloud computer and browser. The dots launch post says they can connect to "over 4,000 apps" through its plugin ecosystem and work toward a person's goals around the clock. People message them in ChatGPT, Slack or Teams.

Dots are rolling out to Pro and Business Premium users in eligible markets, and Enterprise, Edu and Healthcare workspaces can try a beta once an admin turns it on. OpenAI's help center says the Pro plans cover users over 18 outside the European Economic Area, the UK and Switzerland.

The launch post is about work. Neither it nor the dots help pages describe shopping. Meta's Muse, launched on 8 September 2026, is the agent built with buying in mind. Meta's Muse security post says Muse asks for approval with the exact details of a purchase every time, and pays new merchants with single-use card numbers tied to one merchant and one amount. Treat a dot as a browser a buyer may point at your store, and Muse as the agent wired for checkout.

How do OpenAI dots reach an online store?

OpenAI dots reach a business in two ways: through a plugin the person has connected, or by opening the store's website in the dot's own cloud browser.

• Plugins: a dot can use supported plugins that are installed and enabled for the account, under the person's existing ChatGPT app permissions. A business that wants a structured route submits its own plugin through OpenAI's developer platform, which asks for a verified identity, a public remote MCP server, domain verification and a review with test cases and a video walkthrough.
• The cloud browser: without a plugin, the dot visits the site like a person. OpenAI's help page on dots computers and apps says the cloud browser keeps its own sessions, does not inherit the person's logins, and handles sign-in through a private form or a takeover.
• The person's own computer: if a site blocks the cloud browser, OpenAI tells users to ask the dot to try their connected computer.

Shopify stores may already sit in an AI channel without anyone choosing it. Shopify's help page for the Meta channel says direct checkout on Meta surfaces, including Muse when available, is on by default for eligible stores selling to customers in the US, Canada or Mexico. The setting sits under Sales channels > Agentic: turn off "Allow Shopify to manage for me", open Meta AI and Muse, and change Direct checkout. Check it for every Shopify client before the next report.

Can robots.txt block AI agents like dots and Muse?

No: robots.txt cannot target dots or Muse, because neither company publishes a user agent for them. OpenAI's bots page lists OAI-SearchBot, OAI-AdsBot, GPTBot and ChatGPT-User,, and dots are not on it. Even for ChatGPT-User, OpenAI says robots.txt rules may not apply. Meta's web crawler page names five crawlers and Muse is not one of them. Meta's own post says that when Muse browses, it "will appear as your activity".

What does stop them is bot protection at the firewall or CDN; OpenAI itself notes that some sites block cloud browsers. GeekWire reported on 20 September 2026 that Amazon cut Muse off from shopping on Amazon.com, with a notice citing its Conditions of Use. Amazon's reasons, as GeekWire reported them: Meta never told it, the agent does not identify itself, and it appears to store customer credentials.

For most independent stores the right call is the opposite. Amazon earns ad revenue from people browsing its pages; a small store just wants the order. Before tightening bot rules, check whether challenge pages turn away real shoppers' agents, and remember a blocked dot can retry from the person's own computer.

What does an AI agent read on a product page?

An AI agent reads the rendered page, often through its accessibility tree, rather than the hidden markup a search crawler parses. Meta says Muse's browsing sub-agent sees "an accessibility tree snapshot of the page" rather than the raw DOM, and cannot run JavaScript in the page. Google's AI optimization guide says browser agents may analyze screenshots, inspect the DOM and interpret the accessibility tree. OpenAI has not published how dots read a page.

The accessibility tree is the version of the page a screen reader uses: each element's role, name, state and text. An unlabelled icon button, a size picker built from bare divs, or a sold-out state shown only by grey styling can be missing or ambiguous in it. Neither OpenAI nor Meta says its agent reads llms.txt or Markdown copies of pages, and Google's guide says those files are not needed for Google Search.

How should an ecommerce site prepare for OpenAI dots?

An ecommerce site prepares for AI agents by making every fact and purchase step visible, labelled text, because that is what an agent can read and act on. Check the top ten product pages and the checkout:

1. Name every button. Add to cart, quantity steppers and gallery arrows need an accessible name, not just an icon.
2. Make variants real controls. Size and colour pickers should be labelled radio buttons or buttons with names such as "Size 10", not unlabelled swatches.
3. Show stock in words. Write "Out of stock" or "Only 3 left" in visible text. A crossed-out swatch or schema markup alone is not enough.
4. Put prices and thresholds in the HTML. Price, free shipping threshold and delivery estimate belong in the raw and rendered HTML, not inside an image.
5. Label every form field. Checkout fields need real labels. Placeholder text disappears once someone types.
6. Do not hide the answers. Return terms and size guides tucked into display:none tabs or closed accordions may never be read. Keep the key line near the buy button.
7. Keep structured data, then repeat it. Keep product markup, and repeat the same facts in visible text.
8. Let guest checkout finish. Every forced account, CAPTCHA or pop-up is a step the agent hands back to the person, or where it stops.
9. Test it like a screen reader would. Use VoiceOver, NVDA or the Accessibility pane in Chrome DevTools. If it is confusing there, it is confusing to an agent.

This is accessibility work, so it helps shoppers using assistive technology too.

What do OpenAI dots do to a client's analytics report?

Dots and similar agents make part of a store's activity invisible or mislabelled in analytics, in three ways.

• Agent visits look like ordinary visits. GA4 added an AI Assistant channel on 13 May 2026, and it assigns the medium ai-assistant only when the referrer matches a recognized AI assistant. OpenAI has not published a referrer, user agent or IP range for the dots cloud browser, so nobody outside OpenAI can say which channel those sessions land in.
• Purchases inside an AI app skip the tag. Shopify says Google Analytics and client-side pixels do not fire in Meta's direct checkout; only server-to-server pixels for checkout started and completed do. Its page for the Google AI Mode and Gemini channel says the same. Those Meta orders show in the Shopify admin attributed to Meta.
• Totals stop matching. Once some orders arrive through AI channels, GA4 purchases and the store's order count drift apart.

Bot noise has a known fix (see GA4 spam traffic filters). Agents are harder: there is no published identifier to filter on.

How should an agency report AI agents to a client?

Report AI activity as three separate lines, each with its source named, and say plainly what cannot be measured yet.

• AI assistant referrals, from GA4: people who clicked a link in ChatGPT, Gemini or a similar assistant.
• AI channel orders, from the store platform: orders Shopify attributes to Meta or Google AI channels, which GA4 will not show.
• AI visibility, from prompt tracking: whether the brand is named in answers from ChatGPT, Gemini, Perplexity and Google AI Overviews.

Then add one honest sentence. Suggested wording: "Personal AI agents such as dots and Muse can browse the site like a normal visitor. Neither company publishes a way to identify them, so their visits are counted inside our normal traffic and cannot be separated today. Orders placed inside AI apps are shown from Shopify's order records." Annotate any change to bot protection or Shopify's agentic settings, so a jump or dip has a cause on file.

For clients also buying ads inside ChatGPT, see ChatGPT ads for agencies. AgencyReportr does not detect agent visits, and without a published identifier no tool can separate them reliably. Its SEO reporting tool includes an AI assistant traffic section built from GA4, and AI visibility tracking covers ChatGPT, Gemini, Perplexity and Google AI Overviews in the same branded report.

What has OpenAI not said about dots?

OpenAI has not published usage numbers for dots, a user agent or IP range for their cloud browser, or how a dot chooses which business or plugin to use for a task. Meta's crawler page names no Muse crawler either. Until that changes, any tool claiming to count dot visits precisely is guessing.

What an agency controls now: labelled pages, a deliberate bot policy, a checked Shopify agentic setting, and a report that separates what is measured from what is not.

OpenAI dots FAQ

Can robots.txt block OpenAI dots?

Not reliably. OpenAI's bots page lists OAI-SearchBot, OAI-AdsBot, GPTBot and ChatGPT-User, and dots are not on it, so there is no user agent to name in robots.txt. Bot protection at the firewall or CDN is what blocks cloud browsers, and OpenAI tells users to try their connected computer when a site blocks the cloud browser.

Do OpenAI dots show up in GA4?

Not as a separate source. GA4's AI Assistant channel only labels visits whose referrer matches a recognized AI assistant. OpenAI has not published a referrer, user agent or IP range for the dots cloud browser, so those visits are counted inside normal traffic and cannot be separated today.

Can an AI agent buy from my store?

Meta says Muse can complete purchases after asking the person to approve the exact details, using single-use card numbers for new merchants. OpenAI's dots pages describe signing in to websites and approval checks but do not describe shopping. On Shopify, Meta and Google AI channels can also take orders through direct checkout.

Do OpenAI dots read llms.txt?

OpenAI has not said that dots read llms.txt or Markdown copies of pages, and Meta has not said it for Muse. Meta says Muse reads an accessibility tree snapshot of the rendered page, so visible, labelled content matters more than extra files.

Should an online store block AI agents?

Most independent stores should not. An agent usually acts for a real buyer, and a blocked dot can retry from the person's own computer. Amazon blocked Muse, citing its Conditions of Use, but Amazon also earns ad revenue from people browsing its pages, a different position from a small store that simply wants the order.