A person used to ask an assistant which CRM to buy, read the answer and decide. With a personal agent they say what they want done, such as "sort out a CRM for the sales team before January", and leave it. The agent reads, compares, books a demo, maybe starts a trial, and comes back with one recommendation. The shortlist was made without them.
Brands are already being found this way, or passed over, and most of them can't see it happening.
Assistant
- 1A question"Which CRM suits a ten-person sales team?"
- 2Reads a few pagesComparison articles, reviews, trade press
- 3A short listThe person reads it and decides
Personal agent
- 1A goal"Sort out a CRM for the team before Q1"
- 2Knows the personEmail, calendar, past purchases, the tools already connected
- 3Researches and actsReads, compares, books the demo, starts the trial
- 4Comes backOne recommendation, waiting for approval
What Muse and Dots are
Muse is Meta's personal agent, launched in the US on 8 September 2026 in its own app, on the web and inside WhatsApp. It sends emails, books travel, fills in forms and negotiates, and it keeps working after the app is closed, then comes back when something changes or it needs approval. At launch its connectors were mostly consumer ones, for email, calendars, payments, shopping, dining and events.
Dots is OpenAI's, launched at DevDay on 29 September 2026 for ChatGPT Pro and Business Premium users. Each Dot pursues a goal in the background with little oversight and can connect to more than 4,000 apps, Slack and Teams among them.
Why agents find brands differently
Three things set them apart from an assistant.
They hold context. An agent has read the person's email. It knows their calendar, what they've bought before and the tools they already pay for, and when it looks for a CRM it isn't asking "what is the best CRM?" but "what is the best CRM for this team, which uses Google Workspace, has a budget of about £500 a month and complained about Salesforce in March?" Being the brand mentioned most often counts for less than being the obvious fit for that team.
They work on their own. An assistant gives a list and the person chooses. An agent narrows the list itself, often down to a single name, which means the person only ever sees the brands that survived its research and never learns which ones it dropped, or why.
They act. If your pricing is behind a "contact sales" form, or the trial needs a phone call, the agent may simply move on to a vendor whose sign-up it can complete.
Consumer first, business next
Muse is mostly a consumer product today. Dots is still early, although the direction is clear. Meta has already launched Muse for Small Business, and Dots sits on a business plan with Slack and Teams connected from day one. It's only a matter of time before a founder asks an agent to find a PR agency, or a marketing lead asks one to replace a monitoring tool, and the agent comes back having made the choice. The brands that agents trust by then will be the ones that did the work early.
How to be the brand they pick
Most of it is the work that gets you named by ChatGPT, because agents run on the same models and read the same web.
- Be on the pages they read. That means comparison articles, review sites, trade press and your own site. An agent checks more of them than a person would, so a gap shows.
- Describe yourself the same way everywhere. One plain description of what you do, for whom and at what price, unchanged across your site, LinkedIn and every listing. An agent matching you to a person's situation needs to know exactly what you are.
- Say who you're for, and who you're not for. Agents are matching, so a page that says "for B2B tech companies, not consumer brands" helps them place you correctly.
- Publish your price. An agent working to a budget can't ring for a quote.
- Make the next step possible for software. You need a sign-up that works without a sales call, a trial anyone can start, a demo booking page with real slots, and pages that don't hide key facts in images.
- Keep it current. Agents favour recent pages, as assistants do. A page that was right last year can steer an agent wrong this year.
Measuring it
Neither Meta nor OpenAI shows brands what their agents decide, so there's no direct measure yet. The best proxy is the assistants underneath them. Ask your buyers' questions each week on ChatGPT, Gemini, Claude and Perplexity, and record who is named and which pages are cited. Add a few questions written the way a person would brief an agent, with their situation in them. Then watch sign-ups and demo bookings for visits with no search or referral behind them, which is often what agent traffic looks like.
If you'd rather not do that by hand, our AI visibility tracking asks your buyers' questions on the assistants every week and checks the answers against your own facts.
Questions
What are Muse and Dots?
Muse and Dots are personal AI agents that work towards a goal in the background and check in with their user only when they need approval. Meta launched Muse in the US on 8 September 2026, in its own app, on the web and inside WhatsApp. OpenAI launched Dots at DevDay on 29 September 2026 for ChatGPT Pro and Business Premium users. Each Dot pursues a goal with little oversight and can connect to more than 4,000 apps, Slack and Teams among them, which puts it closer to work software than Muse is today.
Are they only for consumers?
Muse is a consumer product for now, and its launch connectors were mostly for everyday tasks such as shopping and dining. Meta has since launched Muse for Small Business, which helps a firm run its own marketing. Dots is nearer to business use already, because it sits on ChatGPT's Business Premium plan and connects to Slack and Teams. A B2B brand should assume agents will be shortlisting business software before long, and the groundwork takes months, so it's worth starting now.
How is this different from getting recommended by ChatGPT?
Getting recommended by an AI agent rests on the same groundwork as getting named by ChatGPT, because agents run on the same models and read the same web. Two things change. The agent knows its user's situation, including the tools they pay for and the budget they're working to, so it ranks brands by fit for that person and a strong showing on generic best-of lists counts for less. The agent also acts on its choice, which means a price it can't find or a sign-up it can't complete can lose you the sale before anyone sees your name.
Can I track whether Muse or Dots recommends me?
There's no direct way to track Muse or Dots yet, because neither Meta nor OpenAI shows brands what their agents decide. The proxy is the assistants underneath them, measured with a fixed set of buyer questions each week, which tells you whether the pages an agent would read put you in the frame. Your own analytics fill in the rest. A demo booking or a trial with no search or referral behind it is often what agent traffic looks like, and a rise in those, set beside a steady trend in the assistants, is the best evidence available today.
Try it on one brand, free
You can set up one brand in an afternoon and see what it finds. Paid plans are monthly, and you can stop whenever you like.