Case Study — Life Sciences
An Investor Asked a Question Based on Data an AI Platform Had Wrong
A clinical-stage biopharma company discovered a narrative drift problem across four AI platforms — because an investor found it first.
- Company:
- Clinical-stage biopharma, asset in Phase 2a, awaiting data ahead of a planned Phase 3 transition
- Platforms audited:
- ChatGPT, Claude, Perplexity, Gemini
- Distinct drift types identified:
- 3 — outdated information, missing material information, inaccurate emphasis
- Platforms showing drift:
- 4 of 4
- Trigger:
- A live investor question based on outdated AI-sourced information
- Status:
- Audit complete; correction in progress under the DRIFT Framework
What triggered this narrative drift audit?
A potential investor, during a scheduled call, asked a company executive a pointed question based on information the investor had obtained from an AI platform. The information was outdated. In that moment, the executive realized someone actively forming an investment decision was working from a version of the company's story the company itself no longer recognized. That single call was enough to trigger a full narrative drift audit.
What is narrative drift?
Narrative drift happens when AI platforms describe a company differently than the company describes itself, because they're synthesizing outdated, conflicting, or unverified sources. It's the same failure mode this case study documents in practice: four AI platforms, four different versions of the same drug's story.
What did the audit find?
QuestionFuel audited how ChatGPT, Claude, Perplexity, and Gemini each described the company's Phase 2a asset, and found that no two platforms told the same story. The drift took three distinct forms: information that hadn't been updated since an earlier point in the trial, materially important information missing entirely from some answers, and at least one platform's response leaning heavily on a single aspect of the drug in a way that wasn't fully accurate. Every platform audited showed some form of drift. The investor's experience wasn't a one-off glitch — it was a predictable outcome of an unmonitored narrative.
"An investor was already forming an opinion based on a story the company no longer recognized."
What caused the drift in this case?
The audit traced the drift to six specific, identifiable causes: stale first-party content, sparse first-party content, incomplete structured data on the company's own press releases, internal inconsistency across the company's website, cross-model divergence, and shadow drift. No single cause explained the full picture — the drift was the compounding effect of all six.
Why does narrative drift matter most during a Phase 2a to Phase 3 transition?
Phase 2a to Phase 3 transitions are exactly the kind of milestone institutional investors, analysts, and partners are already researching using AI platforms, before ever picking up the phone. When the answer they get is wrong, it doesn't just create confusion. It creates doubt, at the exact moment a company can least afford it.
Did the company's investor relations team know about the drift beforehand?
No. The investor relations team — the group closest to how the company's story should be told — had no visibility into how differently AI platforms were already describing the asset, until a live investor conversation surfaced it. The gap wasn't a communications failure. It was an unmonitored blind spot, and narrative drift compounds precisely because no one is watching for it until someone looks, or until an investor asks the wrong question at the wrong moment.
What happens after a narrative drift audit like this?
The audit is complete; correction is underway. This engagement followed the Detect and Review Sources steps of the DRIFT Framework. The next phase — Identify Gaps, Fix Alignment, and Track Continuously — corrects the narrative and establishes ongoing monitoring, so the next investor conversation isn't shaped by a story the company no longer recognizes.
Under a signed NDA, company and product-identifying details are withheld.
Michael Etheredge, VP of Product Development