Agent discoverability: can a buying agent find, understand and buy your offer?
When media planning starts in an assistant or a buying agent, an offer that people know well can stay invisible to machines. Seven levels to work through, and what to keep private.
More and more media planning will start with a prompt: “Find me premium video inventory in French-speaking Belgium for an electric SUV launch in April.” The assistant or buying agent that receives that brief will look for sellers it can understand and query. An offer that every Belgian media planner knows can still be invisible to it.
For publishers, being found, understood and chosen by agents becomes a commercial and technical capability, one that has to be designed, tested and measured like any other.
Why SEO is not enough
A search engine can find a commercial page. It cannot reliably deduce from it that a product is available in French-speaking Belgium, accepts a given video format, reaches a given audience and can be booked for April. Agents need structured descriptions of what you sell and, step by step, interfaces they can query.
Seven levels to build
| Level | The agent’s question | What the publisher provides |
|---|---|---|
| Discoverable | Who can answer my brief? | Canonical pages, brands, markets, clear descriptions. |
| Understandable | What does this publisher sell? | A consistent taxonomy of formats, audiences and contexts. |
| Comparable | How does this offer compare? | Structured attributes, KPIs and standard units. |
| Selectable | Is it relevant to my brief? | Capabilities, restrictions, geography, languages, use cases. |
| Queryable | Can I ask for reach or availability? | Authenticated APIs or agent tools. |
| Buyable | Can I request a proposal or create a deal? | Identity, pricing policy, a transaction workflow. |
| Measurable | Can I follow and analyse delivery? | A reporting API, defined metrics, provenance. |
What to make machine-readable
- Publisher: name, markets, languages, brands, domains, contacts, capabilities.
- Product: stable ID and version, name, description, channel, format, device, geography, language.
- Audience: taxonomy, method, reach, consent constraints, freshness.
- Context: taxonomy, brand suitability, editorial environment.
- Commercial: pricing model, currency, minimum budget, availability, restrictions.
- Measurement: KPIs, methodology, reporting frequency, verification.
- Transaction: booking rules, creative specifications, lead time, required authentication, endpoints.
Two rules make a catalogue usable. First, separate stable attributes (description, format, market) from dynamic ones (availability, reach, indicative price), and serve the dynamic ones from a separate endpoint. Second, say when each piece of information was last updated and which source is authoritative, so that an agent knows whether it can trust it.
The channels
- The public web: stable, text-based, structured, indexable and dated pages about your brands and products.
- Structured data: existing web standards where they apply, rather than proprietary schemas.
- A machine-readable catalogue: JSON or an API with stable identifiers and versioning.
- An advertising API: product search, audiences, authorised pricing, forecasts, proposals.
- Agent protocols: AdCP, where buying agents call tasks such as product discovery and media buy creation; and IAB Tech Lab’s AAMP, with its seller agent SDK and the OpenProposal draft for proposals.
- Registries: IAB Tech Lab’s Agent Registry lists MCP and A2A agents, free of charge. Companies are validated against their GPP and TCF IDs and their domain, and the registry can itself be queried as an MCP server. Register a seller agent once it is mature enough to be called.
- Developer documentation: examples, scopes, authentication, errors, rate limits and service levels.
The two protocol families overlap on most business functions, and nobody knows yet whether they will merge or coexist. A clean internal catalogue lets you serve both without doing the work twice.
Not everything is public
Discoverability is not the same as openness. We recommend access levels:
- Public: brands, formats, capabilities, contacts, public case studies.
- Identified: indicative audiences, detailed products.
- Authenticated: availability, forecasts, authorised prices.
- Trusted: private marketplaces, discounts, private products.
- Transactional: proposals, deals, bookings, reporting.
And some things should never be exposed: prices negotiated with other clients, internal margins and full discount rules; reserved inventory without authentication; personal identifiers or segments without a legal basis; direct endpoints to the ad server, SSPs or CRM; system prompts, secrets, tokens and detailed anti-fraud rules.
Test it like a product
Every month or quarter, run a fixed set of typical briefs through the main assistants and agents. Record whether your company is found, which brands are cited, whether the information is correct and current, and whether the agent can move on to a request or a proposal. The goal is not to be cited once, but to be found and understood correctly, every time.
- Audit how assistants describe you today.
- Normalise product IDs, taxonomies and one source of truth.
- Publish public pages and a machine-readable catalogue.
- Expose an API or agent tools for the right use cases.
- Authenticate with public, partner and transactional scopes.
- Benchmark with your fixed set of briefs.
- Improve what is inconsistent or missing.
We are doing exactly this with our two demo publisher sites, whose offer we express as products that a buying agent can read: see the demo sites.
Sources
- IAB Tech Lab’s Agent Registry hits 10 with Amazon and new deployment types, PPC Land, March 2026.
- IAB Tech Lab gives industry until October 22 to weigh in on AAMP 3.0, PPC Land.
- AAMP and AdCP overlap on 13 functions, PPC Land, August 2026.