Every Professional Category Will Get Its Own ChatGPT

01 /The Concept

OpenEvidence just raised $200M in new funding – only three months after its previous round of $210M. In that span, its valuation doubled from $3B to $6B.

Adding $3 billion in valuation in three months is impressive on its own. But a $6B valuation for a company only three years old is something else entirely.

OpenEvidence built an AI assistant for licensed physicians – a tool doctors can query and get professional-grade answers from.

The professional caliber of the platform was validated in benchmark tests run in August, where AI systems answered questions from the US Medical Licensing Examination. OpenEvidence scored 100%. GPT-5 scored 97%.

A 3% difference might sound minor – until those 3 percentage points affect life-and-death decisions. Then it becomes very significant.

From what's publicly available, OpenEvidence's AI is built on a foundation model enhanced with a retrieval-augmented generation (RAG) layer that draws on content from licensed professional medical journals. Every answer the AI provides includes citations to the professional sources it used.

Licensed physicians with verified credentials can use OpenEvidence for free – the startup monetizes through advertising.

The platform is officially in use at more than 10,000 hospitals and clinics in the US, and 40% of primary care physicians visit it daily. OpenEvidence's AI now conducts more than 15 million consultations per month – double the rate from July, when the previous round was raised.

In late August, OpenEvidence launched a new tool called Visits – which helps physicians take notes during patient appointments and then analyze those notes with the platform's AI. Patients can also bring in files, such as lab results, for AI-assisted review.

Visits effectively becomes the physician's notebook – a full history of every patient. The AI can summarize patient histories on demand and answer any clinical question using both the recorded patient history and the broader OpenEvidence knowledge base.

02 /Why It Matters

The New York Times described OpenEvidence, right in its headline, as "ChatGPT for medicine."

That framing recalls a recent review on a completely different topic, titled "Like ChatGPT – But for Professionals." The argument there was that professionals in every field need their own specialized AI – one that provides more complete, current, and reliable information than general-purpose ChatGPT or any other universal AI chatbot.

OpenEvidence is further confirmation that these platforms are genuinely wanted by professionals.

What's also interesting: even within a single professional domain, different subtopics may benefit from their own distinct AI.

In August, Elion ([related review](/review/novoe-rozhdenie-drugih-marketplejsov)) raised $9.3M on a platform that helps healthcare institutions select and purchase the software they need. Choosing the right enterprise software requires the same things: completeness, currency, and reliability – none of which you'll reliably get from review sites or a general-purpose AI.

And a specialized AI in the same medical domain can be packaged into entirely different products depending on who's facing it.

A week ago, Counsel ([related review](/review/novaja-model-dlja-marketplejsov-uslug)) raised $25M on a patient-facing medical AI. Users get unlimited consultations with the AI – but can connect with a live physician within minutes if they want or need to. Those live consultations cost $29 each, or $199 per year on a subscription. The connecting physician sees a summary of the user's AI conversation and full visit history – allowing them to pick up exactly where the AI left off.

Zooming out further, it's worth revisiting Dappier ([covered here](/review/kak-zabit-sebe-mesto-v-novom-internete)), which argues that question-answering AI systems should now become the primary entry points into any specialized topic area. Since ChatGPT's emergence, people have come to prefer asking questions and getting direct answers over receiving a list of links to read through and synthesize themselves. Dappier provides content projects with the building blocks to assemble their own Q&A AI for their specific vertical. It placed healthcare first on its list of verticals – and estimates the total market for such specialized Q&A systems at $75–100 billion.

03 /Opportunities

Put simply: the internet has become an information landfill – somewhere it's genuinely difficult to find current and reliable information, let alone distinguish reliable information from noise.

Any universal layer built on top of that landfill risks becoming landfill itself. Google is struggling with this. General-purpose AI systems face the same structural problem.

The trend is obvious: build "subsets" of the internet where current, reliable information on a specific topic is curated and then packaged into a specialized product.

This specialization is necessary because no one has the resources to curate the entire internet. Every player will instead manage their own domain – and within that domain, they can bring at least relative order.

And within this trend, the products don't have to be ChatGPT clones.

VuMedi ([related review](/review/prostoj-sposob-otkusit-kusochek-ogromnyh-bjudzhetov)) raised $80M in May on what amounts to "YouTube for physicians" – a video platform where only licensed doctors and pharmaceutical company representatives can publish content.

For a completely different domain, XOi ([related review](/review/tema-v-kotoroj-mozhno-i-horosho-zarabatyvat-i-horosho-prodatsja)) raised $230M in February on an app for home appliance repair technicians. Point your phone's camera at a piece of equipment – the app identifies it and gives the technician a step-by-step diagnostic and repair guide.

So – in what topic area could you build a specialized AI or other specialized product for professionals, or for people who want professional-grade information on a subject they care about? Why would it be better than a general-purpose AI? How would you measure and prove that?

Don't you want to build something that beats ChatGPT? Even just in one area.