Meta Muse
for Marketing

Muse is not a smarter chat window. It is an agent with its own machine, its own browser, and live access to the Meta accounts you connect — which is exactly why its approval rules matter so much.

Meta launched Muse in September 2026 as a personal AI agent rather than an assistant you talk to. The distinction is practical. Muse runs on a secure cloud virtual machine with a dedicated browser, so it can open pages, work through multi-step tasks, and keep going after you put your phone down. Three weeks after launch, Muse for Small Business added the connectors that matter to marketers: Instagram professional analytics, Facebook Pages, and Meta ad accounts, plus a growing list of third-party apps such as Canva, Klaviyo, Shopify, and Slack.

That changes the shape of the work. A chatbot answers the question you typed. Muse can hold a standing job — "every Monday, read my ad account and tell me which three ads are fatiguing" — and run it whether or not you are in the thread. The result is less like a conversation and more like having a junior analyst who already has read access to your accounts.

The other half of the design is the stop. Meta is explicit that nothing publishes, sends, or spends without your approval. Muse can read freely and draft freely, but the moment an action would touch your audience, your brand voice, or your money, it pauses and shows you what it wants to do. That constraint is not a limitation to work around. It is the reason you can safely point an agent at a live ad account in the first place.

The deeper walkthrough — including copy-paste prompts for diagnostics, DM mining, and campaign briefs — is in Meta Muse for marketing. The job Muse can take furthest on its own is Instagram, covered in Meta Muse for Instagram marketing.

Muse is the right tool when the job is reading a Meta account, drafting the next action, and handing it to one person to approve. It is the wrong tool when the job has to be finished inside Ads Manager, shared across a team, or run outside the US and Canada.

Most of the frustration people report with Muse comes from assigning it the wrong kind of job. Muse is excellent at the first two-thirds of a marketing task: pull the numbers, spot the pattern, write the first draft. It deliberately does not do the last third on its own, and for some jobs — building campaign structure, editing conversion tags, restructuring audiences — it cannot do the last third at all, because the built-in connectors do not operate Ads Manager.

So the question to ask before handing Muse anything is not "can Muse do this?" but "where does this job need to end?" If it ends with you reading a readout and tapping approve, Muse fits. If it ends with a change that has to land inside the ad account without you in the loop, you want an agent built for that.

When the job lives inside the ad account rather than beside it, the right tool is an agent with real account access and a governor on spend. That is what AI agents for Meta Ads in the Metaflow app are built for, and it is the hand-off this guide keeps coming back to.

Start with diagnostics and drafts, because those are the jobs where Muse is strongest and where a mistake costs you nothing. Approve actions only after two weeks of reading its work.

The most useful way to think about Muse is as three standing jobs rather than a stream of questions. The first is diagnose: read the connected accounts and surface what is actually happening. Creative fatigue — CTR falling while frequency climbs — is the classic example, because it hides inside dashboard averages and Muse can rank it by spend. Audience overlap between ad sets and the organic question of which post format earns saves versus likes are two more reads that take Muse seconds and take a person an afternoon.

The second job is strategize: turn the diagnosis into a draft. Muse can mine ninety days of comments and DMs for the exact phrases customers use to describe their problem, group them by what the person was worried about, and tell you which three would work as the first line of an ad. It can draft a seven-day campaign brief with three hook variants and a budget recommendation from your own account history. None of this ships; all of it is a better starting point than a blank document.

The third job is execute with approval. Once you trust the reads and the voice, let Muse prepare the actions: a scheduled post with a preview, a budget change showing the old and new amount, a customer reply drafted in your tone. Each one stops for a tap. A sensible rule for the first fortnight is to keep Muse in diagnose-and-draft mode only, then start approving low-stakes actions such as budget moves of ten percent or less.

Instagram is the one channel where Muse can take a job almost all the way: read the analytics, draft the caption, outline the Reel, schedule it, and wait for the tap. It still does not produce the finished visual, cut the asset for other platforms, or handle the DMs that follow. The full weekly workflow is in Meta Muse for Instagram marketing. When captions start to need a calendar and a pipeline behind them, that is content-led growth.

Muse reads freely and drafts freely. Anything that publishes, sends, spends, or pays stops for your approval every time. Learn the table once and you will never be surprised by what it does or does not do.

Every tool that touches a live ad account should come with a plain answer to one question: what can it do while I am not looking? For Muse the answer is unusually clean, and it is worth memorising, because it explains both the good surprises (it already pulled this morning's numbers) and the apparent failures (you said "launch the campaign" and nothing happened).

The pattern is simple. Reads and drafts are free. Anything that reaches a customer, changes your brand voice in public, or moves money stops and waits. Metaflow follows the same rule — it can surface a Google Ads budget change, but it cannot apply one silently.

This is also why "I asked Muse to launch the campaign and nothing happened" is not a bug. Muse built the draft and is waiting for you. Review it, edit it, approve it or kill it. The approval step is where your judgement enters the loop, and it is the only part of the workflow Muse will never automate away.

In the consumer app you do not install a marketing skill. Meta ships the skills, you authorise connectors to add data, and you write goals and instructions to shape the work. Understanding which layer does what saves a lot of wasted setup.

If you are coming from Claude or Cursor, the word "skill" will mislead you. In those tools a skill is a file you write and drop into a folder. In the Muse app a skill is a capability Meta ships — reading, browsing, planning, communicating across connected apps — and new skills appear when you connect a new app rather than when you upload a file. Connect your ad account and campaign analysis becomes available. Connect Shopify and product-level reads appear. There is no settings screen where a marketer pastes in a SKILL.md.

What you do control is the goal and the instruction. A goal is the standing outcome you give Muse — "keep my ad account healthy." An instruction is the procedure you write in the thread that tells Muse how to do it, what to produce, and where to stop. For a marketer, the instruction is the stand-in for a skill file, and it is where almost all of the leverage lives.

The shape of a good instruction

A vague instruction produces a vague agent. The instructions that hold up week after week have the same five parts, borrowed from the weekly ad review in how to create Meta Muse skills:

1. The task. One job, stated plainly: review last week's ad performance.
2. The trigger. When it runs: every Monday at 8am, or whenever you ask.
3. The approval gate. What must stop for you: any pause, any budget change, anything a customer would see.
4. The output. The exact table you want back — ad, spend, CTR change, frequency, recommendation — so you can scan it in a minute.
5. The boundary. What Muse must not do even if it seems helpful: do not change budgets, do not create campaigns, do not reply to anyone.

A procedure you need to keep, version, and reuse across tools belongs in a real skill file rather than a chat instruction. The skills directory is where those live for agents that load them.

Muse cannot see Google Ads or Search Console on its own. Do not paste a third-party connector. Open Metaflow. Muse can ask. Metaflow reads the connected account and waits for you before spend or a live page changes.

Muse has no first-party Google Ads or Search Console integration. There is no Metaflow URL to paste into Muse yet. Open AI agents for Google Ads or SEO automation. Muse can still draft the question. The live read and the approval live in Metaflow.

The limits are structural, not temporary. Metaflow exposes what the Google Ads API exposes. Campaign structure, conversion tracking, audience building, Performance Max configuration, and account settings stay in the Google Ads UI, and no amount of prompting changes that.

SEO follows the same shape. Muse cannot see Search Console on its own. Open SEO automation. Metaflow can read clicks, impressions, positions, and indexing status, flag pages that lost traffic, and draft a refresh brief or a new title. The five jobs worth giving Muse as questions are in Meta Muse for SEO. What Metaflow will not do is crawl your site, change a meta tag on its own, or manage dozens of client properties without you.

The same logic applies to paid search. Muse is a good analyst sitting beside the account. When the work has to happen inside the account on a cadence, that is AI agents for Google Ads.

Knowing the edges up front saves a week of frustrated trial and error. Most of these are launch constraints Meta has signalled will move; a few are design decisions that will not.

Muse is three weeks old as a business tool, and it shows in the edges. Some limits are geography and packaging. Others are deliberate: Meta chose a personal agent with a hard approval gate over an autonomous one, and that choice is unlikely to reverse. Here is the honest list.

Meta named retail partners at Connect and has pointed at transaction fees rather than subscriptions as the long-term model, so expect the connector list and the commerce surface to grow quickly. The approval gate and the personal-account model are the parts to plan around as permanent.

01/

Is Meta Muse good for marketing?

For a specific set of jobs, yes. Muse is genuinely useful for reading live Instagram, Facebook, and ad-account data on a schedule, flagging creative fatigue and audience overlap, drafting short-form copy in your voice, and mining customer DMs for language. It is not a replacement for a strategist, it does not produce long-form content, and it does not run multi-channel campaigns. Treat it as a 24/7 junior analyst and copywriter that needs your approval to ship anything.

02/

Can Muse change my Meta ad budget?

It can prepare the change and show you the old and new amounts, but it cannot apply it on its own. Spend always stops for your approval. If you want budget changes to happen inside the account on a policy without a tap each time, that is a job for a dedicated Meta Ads agent, not Muse.

03/

How much does Meta Muse cost?

Muse is sold as a $20 and a $100 monthly plan. The higher tier unlocks more daily agent actions and earlier access to new app connectors. Google Ads and Search Console jobs run in Metaflow. There is no Metaflow URL to paste into Muse.

04/

Can I upload our Claude skill into Muse?

Not in the consumer app. Muse does not take a SKILL.md file. The equivalent is a goal plus a written instruction with a task, a trigger, an approval gate, an output format, and a boundary. Muse Code is the developer surface where a skill file can load, but it is not the path for a marketer working in the app.

05/

Can Muse post to Instagram and Facebook?

Yes, with your approval. Muse drafts the post, shows you a preview, and publishes once you confirm. The same gate applies to DMs, replies, and ad changes. Nothing reaches your audience without a tap from you.

06/

Is Muse Image the same thing as the Muse agent?

No. Muse Image is part of Meta's Advantage+ creative tooling and generates ad imagery. The Muse agent reads accounts, drafts text, and runs tasks. They share a brand name and nothing else, and the agent will not render your creative for you.

07/

Can my team or agency share one Muse?

Not at launch. Muse is a personal account with a single owner who approves every action. An agency managing several clients would need one Muse per person and still would not get a shared audit trail across accounts. For that, use a platform designed for multiple owners and a governed book of accounts.

• Muse is an agent, not a chatbot: it reads your Meta accounts on its own machine, drafts the next move, and stops for your approval.
• Give it diagnostics and drafts first. Approve low-stakes actions only after two weeks of reading its work.
• Reads and drafts are free. Publishing, sending, spending, and paying always wait for you.
• In the app, skills belong to Meta. Your leverage is the instruction: task, trigger, gate, output, boundary.
• Ads Manager, Google Ads structure, SEO at scale, team logins, and anything outside the US and Canada are exits to a different tool.