Three participants: the AI front door and what it changes for HCP engagement

The We Are Social 2026 mid-year digital report came out last week. It’s a massive data source of 695 slides and covers a wide range of digital topics. The best thing about it? It is entirely free.

Most of it will end up in strategy decks. And honestly, most of it probably won't change how anyone acts. But we spotted three charts, read together, that describe a structural shift for life sciences companies that we don't think is getting anywhere near the attention it deserves.

Here's the summary.

ChatGPT is the AI front door to the internet. 460 million unique monthly web visitors, 592 million mobile MAUs, and roughly 80% of all AI referral traffic globally. “Find information” is the number one reason people use AI, cited by nearly 60% of users.

Health has been a digital-first decision long enough to stop being a trend. 24% of internet users check symptoms online every week. 80% say online resources are important when choosing treatments for everyday ailments. Both figures have been flat to rising over the past eight consecutive quarters. Women lead men by 3 to 10 points across all age brackets. The household caregiver is usually the one doing the lookups.

Trust lags usage. The top concern about AI is unreliability (27%), followed by misinformation (14%) and sycophancy (a nice term for AI telling you what you want to hear) at 11%. People are using AI for health decisions despite not fully trusting the answers.

If you put the three together, you get an implication that is hard to avoid:

Every HCP consultation now has three participants. The patient. The HCP. And an AI that has already been consulted before the appointment and will be consulted again afterwards.

The reality is that most pharma engagement models still assume two.

The unreliability paradox

27% of people worry AI is unreliable. 80% use online resources to decide what to take for a headache anyway. Both are true at the same time, and that gap is where brand preference is now being formed. Not in the exam room. Not on the detail aid. In the private exchange between a patient and a language model at 11pm the night before the appointment.

And pharma is almost entirely absent from that exchange. Not because regulation forbids it but because the content infrastructure was never built for it. Medical content sits behind HCP portals, not in formats LLMs can surface. Patient-facing material is written for websites, not for extraction by a model. The walls designed to protect approved content are the same walls that keep LLMs out of the source material that could anchor them in science.

And it's not just patients. Recent US data suggests roughly two-thirds of doctors are now actively using LLMs in clinical practice to summarise literature, draft patient communications, pressure-test differentials. The “Digital-native HCP” isn't a cohort any more.

The part most teams are getting wrong

The reflex we're seeing in most life sciences teams right now is either to treat this as a 2027 problem, or to reach for familiar marketing logic and optimise for visibility in the new channel. Get the brand mentioned. Climb the LLM equivalent of search rankings. Call it GEO.

It's the wrong first move.

Being mentioned more isn't the goal if you're being mentioned wrong. Alex Jakobsen, founder of CAIRO and one of our partners at Fractal Force, puts it cleanly: if you're selling candy, you just want to be visible. If you're selling a regulated therapy, visibility without accuracy is a liability. You don't want your brand surfaced more. You want it surfaced correctly.

CAIRO sits in a category that doesn't really have a name yet. You could call it AI-layer medical information governance. What they do: monitor how the major LLMs (ChatGPT, Claude, Gemini, Grok, Perplexity, Mistral) reason about specific drugs, track where model responses diverge from approved information, surface pharmacovigilance signals appearing in AI outputs, and — longer term — integrate directly with model providers so that when a brand is mentioned in a conversational AI, the response pulls from pharma-approved content and routes adverse events back to the manufacturer.

And the divergence between how different LLMs describe the same drug isn't marginal. Different models, asked the same question about the same on-market product, can omit major risks, include indications that aren't on the approved label, or skip contraindications entirely. Today. On drugs being prescribed this week. And we did not even mention how language impacts the results.

Where this leaves life sciences leaders

Three things we'd push back on, for anyone telling us this can wait.

Your brand is being represented in LLM responses right now, whether anyone on your team is looking or not. Engaging with this doesn't add risk, it adds visibility to risk that already exists.

Visibility is the wrong first objective. Accuracy is. Chasing share-of-voice inside LLMs before you've solved for correctness is a regulatory incident waiting to happen.

And you can't rebuild what you can't see. Monitoring is the first move, not the last.