What is the AI open web?

The argument

Every previous digital advertising era settled into a small number of closed systems. Search consolidated around one company; social around two. In each case the platform owned the inventory, the buying interface, the measurement and the identity layer, and advertisers accepted terms they could not negotiate because there was no comparable alternative.

The AI open web is the argument that this outcome is not inevitable, and that the deciding factor is whether standards arrive before consolidation does. If any buying agent can discover and transact with any selling agent over a shared protocol, then a small AI publisher is reachable on the same terms as a large one, and the market stays contestable.

What it requires

• A transport standard for agents calling tools — MCP is the current answer.
• An advertising vocabulary on top of it, so discovery, negotiation and reporting mean the same thing everywhere — this is AdCP.
• Comparable measurement, so delivery on one surface can be read against another without translating between proprietary metrics.
• Adoption by surfaces large enough to matter, which is the part that is neither technical nor solved.

Why it is contested

The incentives do not point the same way for everyone. A surface with dominant attention has little reason to make itself substitutable, and interoperability is precisely the property that makes it so. The open web's advocates are therefore mostly the parties who would otherwise be locked out: independent AI publishers, buyers who want cross-surface reach, and the infrastructure layer between them.

That does not make the argument wrong. The open web in display survived alongside walled gardens for two decades, and independent supply retained real value. But it does mean the outcome is a commercial question rather than a technical one, and it will be settled by adoption rather than by which specification is better.

What it means for a buyer today

Practically, very little changes about how you buy this quarter. What it changes is what you should avoid committing to: single-surface integrations that cannot be reused, measurement that only exists inside one platform's reporting, and creative assets that cannot be expressed as structured data. Keeping those portable costs almost nothing now and preserves the option later.