Future of Marketing Briefing: Imagining the ad business hiding inside Meta’s Muse

Meta’s AI personal agent launched earlier this month (Sept. 8), and by day five the app had 730,000 downloads in the U.S., analytics firm Sensor Tower found. That’s more than ChatGPT, Claude, Polymarket or Kalshi pulled in over the same stretch. Whether that pace holds through the coming weeks and months is an open question.

If it does, though, advertising could follow close behind. That’s what happened with Facebook, Instagram, WhatsApp and Threads. Should the same logic apply here, history won’t repeat itself so much as rhyme. Because unlike on any of those social networks, where a person is scrolling, lingering and clicking on whatever catches the eye, Muse is a piece of software executing a task and moving on. A banner won’t work on something that does that.

What might work instead is a placement inside the data an agent actually reads. Agents rarely browse a page the way a person does. Most pull a stripped-down version built for machines to parse cleanly without the clutter a human interface carries. Some publishers are already selling into that layer. Time converted its site to markdown for exactly this reason and now sells a balled ad inside those files, built for a bot to read rather than a person to see. What none of these early moves have determined is what an agent actually does with an ad once it reads once, whether it gets factored in, ignored or lands somewhere in between. Solve that, and the scaling case is obvious. 

Picture asking Muse to book a hotel for the weekend. It comes back with one pick, a boutique place near the river that matches the budget and vibe you wanted. Underneath it sit a couple of other options marked “sponsored”. Nobody can buy their way into being Muse’s actual answer this way. They can only buy their way into being an alternative to it. Cross that line and there’s no reason to trust anything Muse tells you again since you’d never know if it picked the hotel or the highest bidder. 

It is, in other words, a fundamentally different kind of ad slot for marketers who would want their product to influence what a Muse agent does on a person’s behalf. That’s a long way from settled, though. Meta says Muse carries none and nothing coming out of the company suggests otherwise. What’s taking shape instead is subscriptions plus a cut of commerce. Muse already charges $20 and $100 a month for heavier use on top of a free tier. CEO Mark Zuckerberg has floated something bigger on top of that: a small fee, paid by the merchant rather than the user, whenever Muse helps someone buy something, sell something or save money. It is, in other words, a toll on transactions not an ad auction. 

The real question is whether that holds once Muse gets big enough to make an ads business worth building.  

“Our world could suddenly become very small,” said Farhad Divecha, group CEO of digital marketing agency AccuraCast. “With Google Ads and Facebook/Instagram Ads, our clients had multiple opportunities to get in front of their customers. Soon, they may have just one shot (or very few). They’re either in with Muse (or other personal agents) or not. Advertising to ensure our brand is the one that gets picked by Muse to complete a task would be critical to get business.”

Divecha’s right to put emphasis on the “may” since Amazon’s response to Muse suggests the funnel he’s describing doesn’t automatically shrink to a single gate that Meta controls. Rather, it shrinks to whichever gate still has leverage and that’s not obviously Meta’s.

Then there’s Meta’s own promise sitting on top of all that. It has said Muse doesn’t share a person’s conversations or browsing sessions with Meta’s ad systems. Zuckerberg doesn’t make a promise like that unless he knows the alternative would spook people and therefore threaten to slow adoption of the one product Meta needs to take off. The man, after all, has never had a captive audience he didn’t want to sell.

How long he can keep those baser instincts at bay remains to be seen. 

His company’s 2026 AI capex guidance sits at $135 to $145 billion, roughly double what it spent the year before and by the second quarter that spending was eating 98% of Meta’s operating cashflow. Almost all of that bill is being paid by the advertising business Muse is supposed to sit apart from. Let that math hold for a few more quarters and the app may not stay a product Meta can afford to keep walled off, especially one sitting on a gold mine of data spanning everything someone shops for, negotiates over and wants via an agent. 

“Meta says Muse conversations are kept separate from its ad systems, so the obvious route is not mining private prompts for targeting,” said Dan Gee, chief strategy office for media agency MFM. “Rather it’s identifying routes through which genuine incremental value can be attributed.” 

Let’s say all these moving parts align and advertising to someone’s Muse agent becomes as real as advertising to them, Meta would still need enough people to hand over that permission first, and that’s no small ask given who’s doing the asking. Cambridge Analytica happened. So did the Instagram teen mental health hearing as did the years of misinformation nobody caught in time, plus a $17 billion settlement last month over claims it built apps to hook kids and hide what that was doing to them. Trust has never been the thing Meta sells.

And yet people forgave it anyway. They kept using Facebook through Cambridge Analytica. They kept using Instagram through the hearings. Outrage, it turns out, has a shelf life. That said, an agent asks for something different. Facebook, Instagram et al want your attention whereas Muse wants your authority to act while you’re not looking. It’s a much bigger leap of faith and whether people make it anyway is the one part of this story nobody’s written yet. 

“Muse is a really interesting development as it represents the first truly mainstream argentic tool that users can activate to help manage their lives,” said James Mortimer, paid social director at digital marketing agency Roast. “From an advertising perspective, this could be transformational as it creates a whole new ecosystem of data and user behavioral understanding. These signals will be able to help inform delivery of ads to better and more accurately deliver relevant ads to the right people when they are likely to be most interested.”

Numbers to know

108.2 million: Total unique U.S. visitors to Temu in June 2026, indicating that the brand’s U.S. traffic is back, and bigger than it was before tariffs hit.

$2 trillion: The expected valuation of Anthropic when it goes public in November.

65%: Percentage by which the weekly advertiser volume in ChatGPT ads grew between July 9 and August 10.

11.4%: Percentage of which AI is now part of the purchase research process. 

What we’ve covered

In Graphic Detail: LLMs keep citing YouTube in search results

YouTube has become the most-cited source across major AI platforms, overtaking Reddit, with its share climbing fastest in Google’s own AI products. Long-form videos of 10 to 20 minutes with chapters and timestamps get cited most, though those citations fade fast, dropping by around 43% within a month.

From brand deals to equity deals: creators want a stake, not just a fee

Creators are moving from brand deals to taking equity in startups, with new platforms and agencies building the pipes to make it happen.

Some creators want to negotiate deals like TV networks

Creator upfronts are catching on, like Arcade’s first European one in London, but because creator inventory is effectively unlimited, the model is settling into agencies pitching small rosters of top talent.

How women athletes are betting on the creator playbook

Women athletes are treating content creation as a core income stream, using programmes like TOGETHXR’s Snapchat cohort to grow audiences that close pay gaps, lift their leagues and set them up for life after sport.

What we’re reading

Anthropic Shifts Planned IPO to November

Anthropic is reportedly pushing its IPO to November so it can show strong third-quarter numbers, aiming for a record valuation of around $2 trillion, while investors weigh whether Dario Amodei’s call to slow AI development will dent its prospects, per The Wall Street Journal.

German court rules Meta liable for fake ads on Instagram, Facebook

A German court has ruled Meta liable for scam ads that misused a financial portal’s branding, ordering removal damages and revenue disclosure, after finding its algorithmic control over content strips it of the Digital Services Act’s lack-of-knowledge defence, per Reuters. Meta disagrees and may appeal.

Google, OpenAI and Anthropic AI Safety Group Takes Shape

Google, OpenAI and Anthropic plan to launch their own frontier AI safety standards body by early 2027, filling the gap left by stalled government efforts, though critics worry it could box out rivals like Meta and open-source developers, according to The Information.

Amazon blocking Muse is a gift to Meta

Amazon blocking Meta’s Muse agent nudges Meta towards an ad partnership with Amazon, like the one Amazon struck with OpenAI, since its valuable first-party data means agentic commerce only works on its terms, per Mobile Dev Memo.