Loss of Exclusivity Happens Twice Now: The Second Cliff Is in the Answer Layer

Loss of exclusivity now happens twice: once when a pharmaceutical product's legal and commercial protections end, and again in the AI answer layer, where decades of molecule-level authority begin supporting every approved generic or biosimilar that enters behind the originator. Managing that second loss — ensuring the complete, substantiated decision context survives AI synthesis — is answer-layer governance, and almost nobody has assigned an owner to it.

Every pharmaceutical brand team knows the traditional loss-of-exclusivity playbook: pricing strategy, rebate programs, formulary defense, payer negotiations, provider education and, where appropriate, consumer communications. Notice something about that list. Every classic lever is organized around human decision-makers and institutions.

Meanwhile, nearly one in three American adults says they turned to AI chatbots for health information in the past year — as many as used social media for it, per KFF. Some ask before deciding whether to see a provider at all; many never follow up with a clinician. And at the moment an originator franchise is most exposed, one of the highest-value questions in the category becomes: is the generic — or, for biologics, the biosimilar — as good as the originator?

Increasingly, an answer system responds in seconds, assembling a reply from sources it retrieves, ranks and synthesizes. The strategic question is not whether your brand name appears in it. It is whether the complete, substantiated decision context survives into the answer. In our early audits, the striking pattern was not hostility toward the originator. It was absence — absence from the very evidence set the AI system uses to frame substitution.

The sentence that changes sides

There is a structural reason for this, and it is older than AI. The most trusted references in medicine — the manuals, the databases, the university health pages, the clinical compendia — describe molecules, not brands. That is their job. That is precisely why they are trusted. For the entire life of an exclusivity period, this arrangement quietly works in the brand's favor: every neutral, authoritative sentence about the molecule has exactly one product it can possibly describe. The reference work says "the only approved treatment," and the reader hears your brand name, even though it isn't on the page.

Then competition enters. Not one word changes at the reference layer. Nothing needs correcting; nothing is wrong. But decades of accumulated, molecule-level authority — sitting at the most-retrieved sources in your category — now support every approved product carrying the molecule. The source does not have to change. Its commercial meaning does.

The information advantage was never protecting the brand. It was protecting the molecule.

Humira showed what ownership looks like

When adalimumab biosimilars arrived, the payer layer moved with breathtaking speed — CVS Caremark removed branded Humira from its major national commercial template formularies effective April 1, 2024, and 97% of adalimumab prescriptions had converted to preferred biosimilars within weeks. The payer layer moved because it had owners, budgets and dashboards.

The answer layer may move differently — by engine, by question, by date, by user context. But here is the uncomfortable truth: most organizations cannot tell you whether it moved at all, because they never established a baseline. Ask your favorite assistant, right now, whether a biosimilar is as good as the originator in any major category. Don't just read the answer. Read the citations. Payers, competitor manufacturers, trade press, patient forums. Ask yourself who supplied each one on purpose — and who just left the room empty.

One important distinction before going further: generics and biosimilars are not the same thing. Generics are bioequivalent copies of chemically synthesized drugs; biosimilars are highly similar biologics with no clinically meaningful differences from their reference product. What they share is the moment that matters here — the moment a AI system is asked whether switching is safe.

And the front door for that question is becoming more consequential by the month. When OpenAI introduced ChatGPT Health in January, reporting noted 40 million people were already asking ChatGPT health questions daily; OpenAI itself says over 230 million people globally ask health and wellness questions on ChatGPT every week, and in July it began rolling Health out broadly to U.S. adults, with optional connections to supported medical records. Anthropic introduced Claude for Healthcare in January, letting subscribers securely connect lab results and health records. In May, Google made its Gemini-powered Health Coach broadly available through the rebuilt Google Health app. Three of the largest consumer AI platforms are no longer merely answering health questions — they are building persistent, personalized health interfaces around them.

When the question is framed as equivalence alone, the lower-cost alternative has the structural advantage.

To be clear, this is not an argument for teaching AI to discount an accurate equivalence conclusion. FDA-approved generics are expected to provide the same clinical benefit as the brand; biosimilars are expected to match their reference product's safety and effectiveness. If equivalence is the complete, clinically relevant answer, the AI system should say so. The communications opportunity is narrower and more honest: to ensure that equivalence on one dimension does not erase other substantiated considerations — formulation, delivery, approved uses, continuity, patient services, supply reliability, evidence depth — when those considerations genuinely matter to the decision.

Which is why the goal is not to make the originator the citation behind every answer. Often it should not be. The goal is decision-frame fidelity: ensuring the full, substantiated decision context survives synthesis — whether the AI system cites the label, FDA, a journal, a guideline, an independent expert, or the brand itself.

Why this belongs to Communications

Answer systems do not judge credibility exactly as people do — they retrieve, rank and synthesize signals of relevance, accessibility and quality at AI system scale, in ways that vary by engine and are not fully disclosed. But what that machinery rewards is much of what Communications has always been responsible for: precision (a claim stated exactly, with the number and source attached, is something a system can carry; a vibe is not), consistency (one canonical voice compounds; a brand speaking from four websites in four voices fragments), and attached proof (named, credentialed people; citations; independent verification).

The opportunity is not to manipulate a model. It is to orchestrate an accurate, coherent, verifiable public evidence system that both people and AI systems can use. Answer engines simply make the consequences of that work visible — in the answer, and in its citations.

Before any tactics, assign an accountable owner. Communications should own the answer-layer operating loop — not every input to it. Medical owns scientific integrity. Regulatory and Legal own the boundaries. Market Access owns payer evidence. Digital owns technical discoverability. Communications owns cross-source coherence, trusted-intermediary strategy, and the measurement of what actually survives into the answer.

Three things to do before your next patent cliff

1. Write down the fifteen questions. The ones your customers ask at the moment of substitution — in their words, not your brand team's. Run them monthly, across more than one AI surface. Date-stamp everything. You cannot manage an answer you've never actually read.

2. Find your sentence. Somewhere in the most trusted reference in your category, there is a sentence that has quietly protected your brand for years — and it describes the molecule, not you. Go read it the way a AI system reads it. Then decide, calmly and factually, what the reference layer should say now that the world has changed.

3. Audit your evidence for retrievability. Much of a brand's most decision-relevant evidence is hard to discover, extract or verify at the moment of inquiry — stranded in PDFs, paywalled journals, or unstructured pages. PDFs and paywalls don't automatically make evidence invisible, and structured data doesn't guarantee inclusion. But evidence that cannot be reliably retrieved and cited at the moment of the question has little power in the answer.

If evidence cannot be retrieved and cited at the moment of the question, it has little power in the answer.

Proof of process is respect. Showing your work is credibility. The brands that win the next decade of substitution battles won't be the loudest ones. They'll be the ones the AI systems can quote.

This essay also appears in AI in PR, Signal Fidelity Group's LinkedIn newsletter on AI's collision with communications, trust, and regulated industries. Next week: the buyer isn't human either — agentic optimization and the purchasing decisions AI systems are beginning to make.