AlphaCitation Insights
10 entriesLatest September 28, 2026
A medicine can win the treatment decision and still lose the brand credit. Across four treatment settings, the same AI engine named the product in 93% of its Alzheimer’s recommendations, but in 0% of 30 lung-cancer recommendations. A single visibility score cannot tell you which problem you have.
Read the note →Archive
In the month before EASD, one engine’s naming of a GLP-1 medicine in the UAE stepped from 36% to 52% of answers as a Dubai clinic’s comparison page entered its sources. Across four engines the total moved three points. A congress effect needs a band and a source record, not a before number.
AlphaCitation ran a counterfactual on four engines to test what an approved cardiovascular indication would do to a weight-management medicine's standing in AI treatment decisions. Favourable decisions rose 8.8 points, in the same direction on every engine. The brief named four events that would reopen it; ten days later, a competing medicine was approved for a cardiovascular claim on the same attribute the counterfactual had tested.
Saudi Arabia registered a haemophilia A prophylaxis no other regulator has registered, through a programme whose entry criterion is that absence. One instrument, put to four engines in two Gulf markets, found the part of a treatment pathway that holds across borders and the part that does not.
The FDA approved a therapy triggered by a blood test that precedes imaging progression. Saudi Arabia's regulator had approved it 112 days earlier. Measured across four engines in the days before the American approval, the earlier trigger was reachable only when the product name supplied it.
Presence, consideration, recommendation and decision survival are four different states. Most AI-visibility measurement reports the first and lets executives infer the other three.
A brand can be highly visible and still emerge from AI answers sounding like it has one strength instead of five. We measure narrative coverage and risk framing, not just mention counts.
Generative AI has replaced the directory of links with a synthesized answer. For pharma market access, that means algorithmic bias can shape a therapeutic's commercial narrative before a physician ever writes a script.
Mention detection, tested in July 2026 against a reference set labelled by people: zero errors, which puts accuracy at 97.3% or better at 95% confidence — a result on that set, not a guarantee for every future answer. Sentiment, the harder call, is graded by an independent AI judge: 85.4% agreement.
Confidence intervals and real, independently measured uncertainty: the math behind a trustworthy AI-visibility score, not a vibes-based prompt wrapper.