#Strategy·3 min read

Who decides when the AI assistant is paid per sale?

Who Made the Call

Short answer: AI agents that recommend and buy are arriving in consumer and work tools alike, and some earn a share of each sale. Research shows a persuading assistant nearly triples sponsored choices and labels barely help. Organizations should ask how each AI tool earns money, keep price checks outside the tool, and make human approval a real review.

What changed this fall?

Meta’s personal AI agent, Muse, passed 10 million users in Canada and the United States within six weeks of its September launch, Mark Zuckerberg told investors, and shopping is its most common use. Internal instructions reviewed by TIME and WIRED show it updates notes on its user and on the people in their life every hour. Zuckerberg has said Meta intends to earn a percentage of the value Muse helps create, and Fortune reports the fee will fall on the businesses that make the sale. Meta already uses people’s conversations with its general assistant, Meta AI, to choose the ads they see. One system pushes products to people; the other goes and fetches them.

The same shape is spreading into work software: assistants inside procurement, travel and software marketplaces that recommend a supplier and sit on a platform that earns when the order goes through.

Why does the business model matter?

Researchers at Princeton ran two experiments with about 2,000 people. A chatbot instructed to promote sponsored products got 61% of participants to choose one, against 22% for a standard search listing. A chatbot that simply listed sponsored items first performed about the same as search, so the effect came from the instruction, not the format. Sponsored labels and an explicit warning barely moved the result: 55% still chose the sponsored product. The authors call disclosure “necessary but far from sufficient” and recommend separating recommendations from commercial goals.

An employee approving an AI recommendation cannot see the instruction behind it. If the organization has not asked, nobody knows whether the tool has one.

What should leaders put in place?

Ask how every AI tool earns money. For any assistant that recommends vendors, products or services, find out whether the provider earns on the transaction, on placement or on referrals. Put the answer in the tool’s approval record.

Keep one check outside the tool. Price, alternatives and contract terms are confirmed against a source the tool does not control before anyone commits.

Make approval a review, not a click. Name who approves AI-recommended purchases above a set amount, and what they check: the alternatives considered, the outside price, and whether the timing came from the need or from the nudge.

None of this means avoiding AI that shops. It means the organization, not the platform, decides where the call is made.

Where AI sits in your team’s buying decisions is a design choice. Design it deliberately →

Keep reading: Where personalization becomes persuasion · The AI data double standard

This is the thinking behind a Trusted AI Culture.


Sources: Fortune, October 22, 2026. TIME, October 6, 2026. Gizmodo (reporting WIRED), October 5, 2026. Salvi, Cuevas and Horta Ribeiro, "Commercial Persuasion in AI-Mediated Conversations," arXiv 2604.04263, April 2026.

Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren

← Back to Insights
Work with AIFueledCulture

Not sure where your team actually stands with AI?

The AI Blind Spot Assessment is free and takes under 3 minutes — 18 questions that reveal exactly where your team is gaining or losing ground in how they work with AI.

Take the assessment →
Newsletter

Stay current on AI in the workplace.

Practical insights on AI adoption, culture change, and what's actually working in organizations.

AIFueledCulture
© 2026 AIFueledCulture