Amazon has blocked Meta’s Muse personal AI agent from shopping on Amazon[dot]com, extending its fight against third-party shopping agents from Perplexity to one of its largest technology partners.
GeekWire reports:
“The problem, Amazon says, is that it never agreed to any of it. Meta didn’t tell Amazon that Muse would access its store, the agent doesn’t identify itself when it browses, and it appears to capture and store customer credentials, which the company says could create privacy and security risks.”
Left unsaid: This customer behavior could threaten Amazon’s high-margin advertising business, which was running at an annualized rate of $70 billion as of Q1 2026.
Read: “Amazon blocks Meta’s Muse AI assistant in new standoff over agentic shopping” (Sept. 21) – GeekWire
Adweek adds that Amazon expects outside agents to operate transparently, allow merchants to opt out, provide a “mutual value exchange” and improve the customer experience. The reference to a value exchange suggests that agent access may ultimately depend on commercial terms, not merely technical compliance.
Read: “Amazon Locks Out Meta’s Muse in Agentic Shopping Standoff” (Sept. 21) – Adweek
The legal framing is notable. After the Ninth Circuit ruled in August (tipsheet) that the customer, not Perplexity, was accessing Amazon’s computers, Amazon is now moving from federal anti-hacking law toward contracts and terms of service. Its warning to Muse users explicitly invokes those terms.
From tipsheet: Why this matters for advertising:
- Agents can bypass sponsored listings. An agent can move from a consumer’s request directly to a recommendation or purchase without showing Amazon’s search ads.
- The sale can leave Amazon. Muse can compare Amazon’s products and prices with competing stores, threatening both Amazon’s advertising revenue and the underlying transaction.
- Attribution becomes harder. If an agent encounters recommendations across several services before buying, marketers may struggle to connect the conversion to a particular ad, click or platform.
- Agent access becomes an advertising decision. Retailers that control whether outside agents can enter their stores also influence which recommendations, ads and merchants can reach their customers.
LLMs & CHATBOTS
Muse moves into the mainstream
- “Meta’s New Muse AI App Tops Charts, Draws Strong Early Reviews” (Sept. 21) – Bloomberg
- “Meta’s Muse TV ad is the latest sign that AI agents are going mainstream” (Sept. 20) – Business Insider (subscription)
- “Meta’s stock is enjoying its best month in 13 years thanks to the company’s hot new AI assistant” (Sept. 21) – MarketWatch
EVENTS
From tipsheet: Meta Connect’s advertising story
Meta Connect begins tomorrow. Although its published agenda includes no sessions dedicated to advertising, the two-day event could offer important clues about Meta’s commercial AI strategy.
Four developments deserve marketers’ attention:
- Business agents could move messaging closer to the point of sale.
- Meta’s Muse models are designed not only to understand but to decide and act.
- AI glasses could create a new interface for product and business discovery.
- Horizon’s tools could tighten the connection between content creation and distribution.
Together, these initiatives could give Meta greater control over the path from discovery to transaction. The most immediate question is whether Meta will connect Business Agent directly to its advertising system.
Read the entire article on tipsheet’s website. (Sept. 21)
SELL-SIDE
Curation moves audience matching to publishers
Audience data platform Permutive is expanding its existing relationship with TransUnion by making TransUnion audience segments directly available across its network of more than 150 publishers.
The companies argue that moving audience matching closer to the publisher can improve match rates, expand reach and reduce data fees. In one test, Permutive says a financial services advertiser generated 12x more addressable impressions, reached 5x more consumers and exceeded its target KPI by 170%. It also says bid requests without an ID signal produce 41% lower CPMs.
Julie Clark, SVP of diversified markets, media and entertainment at TransUnion, explains what distinguishes the partnership:
“The unique element here is having one key to many different sources, and being able to have that be scaled out across publishers. This is one of the first times that we’ve scaled in this way where our identity becomes incredibly portable.”
Read: “Can Sell-Side Curation Solve The Cookieless Audience Problem For Advertisers?” (Sept. 21) – AdExchanger
From tipsheet: Sell-side curation belongs to the same architectural shift as containerized bidding and SSP-side decisioning.
- Containerization: Moves bidding and optimization closer to the auction.
- Sell-side curation: Moves identity matching and audience construction closer to the publisher.
- Shared logic: Brings computation to the data and inventory instead of sending fragmented signals through the programmatic chain.
Together, these models can improve signal fidelity, reduce intermediaries and fees and expose inventory that buy-side systems might otherwise overlook. The sell side is evolving from a passive inventory pipe into a computational layer. For AI agents, that means cleaner, more complete audiences they can evaluate and act on.
LLMs & CHATBOTS
Anthropic builds out its marketing machine
Anthropic is adding to the marketing infrastructure behind Claude.
A new “Product Marketing Manager, Platform” will develop campaigns designed to drive API adoption and platform usage, while a new developer copywriter will create work across launches, social, paid and events.
Both roles are listed on Anthropic’s job board:
The hires add an execution layer to the marketing measurement operation tipsheet highlighted in August. Anthropic is bringing marketing mix modeling in-house and using geo experiments, synthetic controls and incrementality testing to determine which marketing investments drive growth.
From tipsheet: The bigger picture: Anthropic still says Claude will remain ad-free. But it is building a more sophisticated advertising and marketing operation of its own, connecting paid media measurement with campaigns designed to drive adoption and usage.
How long until they drink the Kool-Aid on Claude ads?
LLMs & CHATBOTS
Why ChatGPT fanouts may matter
A “fanout” is an additional search query an AI generates to help answer a prompt.
Josh Blyskal of GEO firm Profound says ChatGPT is now generating roughly 70% more fanouts than before, based on an analysis of 3.5 million citations.
Citing data from a presentation he gave at the Shenzhen SEO Conference last week, Blyskal writes:
“ChatGPT is becoming an evaluator, doing more of the work of comparing brands, products and services itself….
ChatGPT uses these fanouts to feel around and path through the internet rather than doing an A → B content retrieval mechanism.”
Blyskal found that 64% of ChatGPT’s second fanouts carried forward at least two concepts from the first, which he interprets as a shift toward a more iterative research process. Read more on LinkedIn. (Sept. 21)
From tipsheet: Citations show the sources that appear in the final answer. Fanouts show the questions ChatGPT generated before making its selection, but the data does not tell us how ChatGPT weighs those searches. Instead, it suggests that marketers should look beyond citation counts to the evidence and attributes the system seeks before making a recommendation.
LLMs & CHATBOTS
Developments
- Upgrade: Introducing Grok 4.7 (Sept. 21) – SpaceXAI
- State of AI: “Frontier Overhangs” (Sept. 21) – Analyst Ben Thompson on Stratechery (subscription)
- Shopping: Shopify to Use Meta’s Muse for Agentic Checkout (Sept. 21) – The Wall Street Journal (subscription)
AGENCIES
Opinion: AI weakens holding companies’ moat
AI is making two traditional holding-company advantages, production and insight at scale, cheaper and more widely available. Consultant Ian Whittaker argues on LinkedIn that the industry’s scale advantage is therefore “moving, not disappearing.”
Holding companies are responding by building proprietary AI platforms, but Whittaker questions whether a platform clients can rent for themselves constitutes a durable moat. (Read tipsheet about WPP Open, for example.)
He says that independent agencies cannot match that investment or compete in principal media at the same scale, but founder-owned firms can differentiate themselves through direct client relationships and a deeper understanding of their businesses.
Those relationships only matter if agencies can explain their work in terms CFOs recognize: risk, return and timing. Whittaker concludes:
“The trusted adviser role to the CEO that agencies used to hold is vacant. My view is it will not go to the largest supplier. It will go to whoever the CFO believes.”
Read more. (Sept. 21)
From tipsheet: AI makes agency capabilities easier to replicate, but not client understanding or trust. As execution becomes automated, control moves to whoever helps define what the systems should optimize for.
MORE
- Remedies reality: “The Practical Effects of the Google Antitrust Remedies” (Sept. 21) – Ari Paparo on Marketecture
- Bots: “Paywall violation ‘eviscerates’ fair use defense: What OpenAI-NYT twist means for publishers” (Sept. 21) – Digiday (subscription)
- Pitch wins: “Publicis Media Tops In New Business For The First Half Of 2026” (Sept. 21) – MediaPost
- Tag repair: Multi-source conversions in Google Ads (beta) – Google Ads Help
- Bidder logic: Why AI model Jev looks familiar to ad tech (Sept. 20) – Sergio Serra, product leader at Amazon Web Services on LinkedIn