How to structure a pharma pipeline page so AI can read it

There is no schema.org type for a drug pipeline, so an AI engine often cannot tell a clinical-stage asset page from ordinary marketing copy. You make a pipeline page machine-readable by composing the types that do exist over a clean, labeled HTML table, so an assistant can extract each asset, its indication, and its phase without guessing. The pipeline is the single most-queried fact about a clinical-stage company, and most companies publish it in a form a machine cannot read.

Why pipeline pages are hard for AI to read

What a partner, an investor, or a scientist asks an assistant about a clinical-stage company is almost always the same: what is in development, at what phase, for what indication. Yet the pipeline is usually published as a styled graphic or an image, with the actual facts locked inside the pixels. Even when it is a table, it often carries no machine-readable structure. And schema.org, which offers Drug, MedicalCondition, MedicalWebPage, and Dataset, has no dedicated type for a pipeline or a clinical stage. There is no standard for the exact thing everyone asks about.

Start with the HTML, not the schema

Structured data only helps if the underlying content is already present and readable in the HTML. Build the pipeline as a real HTML table with explicit column headers, asset, modality, indication, phase, status, and real text in every cell, not an exported image and not a chart drawn in the browser. If a crawler that ignores scripts cannot read the table, no amount of markup will save it.

Compose the types that do exist

With the table in place, layer the schema.org types that fit, as a deliberate composition rather than a single tag:

• MedicalWebPage to type the page itself.
• Drug for each asset, with its name and, where public, its mechanism.
• MedicalCondition for each indication.
• An ItemList wrapping the rows, so the set of assets is explicitly enumerable rather than an undifferentiated block.

There is no phase property anywhere in the vocabulary, so the development phase has to live in a clearly labeled field and in visible text. This is A39A's recommended composition, not an official standard, because no official standard exists. That gap is exactly why structuring it well is a differentiator.

Make every row self-describing

Each row should state, in text a crawler reads without surrounding context, the asset, the indication, the development phase, and the date the status was last confirmed. The visible "as of" date matters more than it looks: AI engines favor content whose recency is clear, and a pipeline with no date reads as possibly stale, which is the one thing a pipeline must never be.

Keep it within what was approved

Pipeline pages carry forward-looking and investor-sensitive language, and structure changes none of that. The markup describes the approved text; it never adds a claim, a phase, or an implication that regulatory, legal, and investor relations did not sign off. Done right, the page reads the same way to a partner's analyst, a scientist, and an AI assistant, and all three are reading the content you actually cleared.

The same thinking carries into retiring the clinical PDF library, and it depends first on the page being visible to AI crawlers at all.

Source: schema.org medical types (Drug, MedicalWebPage, MedicalCondition, Dataset).