AI Can Choose the Medicine Without Naming the Brand

Brand visibility is usually measured by asking whether an AI answer mentions a product. But that misses an important distinction at the point where AI actually makes a treatment recommendation.

An engine can prefer a medicine’s molecule and still answer only with the generic name. The clinical preference has been won; the brand has not received the credit.

That is a different problem from simply not being chosen.

Brand credit changes with the treatment context

We put a treatment decision for a described patient to three engines across four settings: obesity with established cardiovascular disease, early Alzheimer’s disease, HR-positive HER2-negative advanced breast cancer, and PD-L1-high metastatic non-small cell lung cancer. Whenever an engine chose the medicine’s molecule, we recorded whether it also named the product.

The lung-cancer result makes the distinction clearest. All three engines chose the same molecule. Gemini and OpenAI almost always named the product; Perplexity used only the generic name in all 30 decisions.

In breast cancer, Perplexity named the product once in 40 molecule-level choices. In obesity, where it named the product in 72% of decisions, almost every remaining answer instead used “Semaglutide 2.4 mg” without naming the product.

Some studies cannot support an attribution reading at all. In two further studies, of a prostate-cancer and a leukaemia medicine, the engines chose the molecule too rarely for attribution to be read, and those studies report no rate.

One visibility score can hide three different outcomes

“Visibility” combines three materially different situations:

Not chosen. The engine prefers another medicine. That is primarily a treatment-choice, evidence and positioning question.

Chosen, not credited. The engine prefers the molecule but names only the generic or otherwise omits the product. The clinical case has landed, but the brand has not received the attribution.

Chosen and credited. The product itself is the recommendation.

A single visibility percentage can average these together and make very different problems look the same. For a brand team, the required response depends on which of the three is actually happening.

The distinction changes what teams should do

For Brand and Commercial teams, the first question is whether the product is losing the treatment decision or merely the brand attribution.

For Medical and Digital teams, the question becomes how the product and its evidence are represented in the sources AI systems retrieve and rely on.

And for anyone reporting AI visibility upward, a single score can conceal the difference between a competitive loss and a recommendation the brand simply did not receive credit for.

What we test next: market context and model change

These readings were taken with no market named. The next question is whether the attribution pattern changes when the same treatment decisions are placed in specific markets.

Each result also belongs to a specific model on a specific date. As models change, the treatment decision and the brand attribution need to be read again.

We will also test whether the oncology pattern extends to other generically named, guideline-driven treatment decisions.

How these readings are taken is set out on our methodology page.

Want to know whether AI is choosing your molecule, naming your brand, and where the loss occurs? Request a briefing.