On Friday OpenAI formally announced that conversational ad units are coming to ChatGPT (no MCP required)

"we plan to test ads at the bottom of answers in ChatGPT when there’s a relevant sponsored product or service based on your current conversation"—Fidji Simo

At PA, we've long believed commerce eats advertising as discovery and conversion converge everywhere an ad can appear.

Fidji is CEO, Applications at OpenAI so there is likely a link worth identifying

OpenAI adopted Model Context Protocol (MCP) as the standard for building integrations (now called "Apps") for ChatGPT. Apps, however, have had a distribution problem:

..filtering all apps [functions] based on contextual utility (your chat) in real-time hasn't been economically viable at scale.

Advertising changes this

Ads introduce funding that makes selective, intent-driven invocation of functionality viable.

If functionality can be invoked inside an ad or brand chat, installs become optional.

At that point, 'ads' stop being messages and become interfaces:

• inventory becomes callable
• pricing becomes queryable
• availability becomes real-time
• checkout becomes a tool, not a redirect
• order status becomes checkable
• support becomes accessible
• even negotiation becomes feasible

Ads stop pointing to software and start being the software.

This is where commerce and advertising finally collapse into a single system.

It lets brand chat move from:

—“Here’s some information”

to:

—“Here’s the outcome you were already trying to achieve”

The implication is of course bigger than ad formats.

- Ads stop being distinct from discovery,

- 'Conversion' stops being a downstream metric (CTR and ROAS are proxies because the decision engine isn't the transaction engine),

- Conversion becomes the ad product.

If you’re in retail or retail media the questions I would be asking are:

1. do we have an MCP-compatible architecture for agentic interoperability?

2. how do existing merchandising and monetisation rules survive real-time agent filters and ranking?

3. what differentiated rich experiences justify selection when interfaces are callable?

Google and Shopify's Universal Commerce efforts are adjacent and important.

So is Stripe's headless transaction stack.

But transaction is only one function.

MCP architectures need to govern decisioning, not just payment.

With PA sub brand Retail MCP we are working on multiple pilots with leading retailers to port functionality they offer into agentic workflows using MCP.

If MCP is the standard, then a brand needs build their MCP tooling once

They can then traffic this into ChatGPT, Claude, and others.

It solves the fragmentation problem. Brands need not build a 'ChatGPT App', a 'Claude Tool' - they build one MCP Server and pay for distribution across different AI surfaces, tap their loyalty base for installs and can support owned channel (e.g. onsite) interfaces.