Better Rank

Inspiration

Every day, millions of people open an AI and type "I need a therapist." In seconds, the AI picks a winner. That winner gets the click, the signup, the person who needed help. BetterHelp has spent years building the best product in mental health market but in 2026, having the best product isn't enough. You have to be the AI's answer. We built BetterRank because no tool existed to tell BetterHelp whether they were winning that moment, losing it, or why.

What it does

BetterHelp is great at helping people find therapy. But when someone asks an AI "what's the best therapy app?" — nobody knew if BetterHelp was the answer the AI gave back. BetterRank fixes that. It works in three steps: Measure — We ask an AI thousands of questions that real people type when they're looking for therapy. We track how often BetterHelp comes up versus competitors like Talkspace and Cerebral. Think of it like a scoreboard. Diagnose — When a competitor wins instead of BetterHelp, we find out why. Is their website structured better? Do they use certain words the AI trusts more? We don't guess — we find the actual reason. Optimize — We turn that reason into a fix. A clear, prioritized to-do list that BetterHelp's team can act on right away to improve their score.

How we built it

Gemini 2.5 Flash simulates thousands of high-intent mental health queries at scale Braintrust tracks Share of Model as a reliable, comparable metric over time Tavily reverse-engineers why an AI favored a competitor using live web data A custom Optimization Roadmap engine translates findings into ranked, actionable fixes tailored to BetterHelp's competitive position

Challenges we ran into

We started on Emergent AI but ran into serious compatibility issues — getting our tracing service, LLM, and search tool to talk to each other cleanly proved harder than expected. The stack wouldn't play nice together. We moved to Vercel to solve that — but hit a new problem. The model started hallucinating, generating confident-sounding recommendations that simply weren't accurate. For a product built on trust and precision, that was a non-starter. The breakthrough came when we swapped Arize AI out for Braintrust. It integrated far more smoothly with Gemini and Tavily, gave us the observability we needed, and finally let the full pipeline run the way we designed it. Every wrong turn taught us something. The final stack didn't come from a plan — it came from pressure testing until something held.

Accomplishments that we're proud of

BetterHelp can see its real-time competitive position inside AI models — and trace every gap directly to a fixable signal. We built the full loop: Simulate → Observe → Diagnose → Optimize → Repeat. That's what turns GEO from a concept into a repeatable competitive advantage.

What we learned

Got to learn a lot from eachother since we came from different backgrounds. BetterRank helps BetterHelp understand how AI models recommend them compared to competitors like Talkspace and Cerebral. Measure — We run a large set of therapy-related queries through an AI and track how often BetterHelp is recommended versus competitors. We call this Share of Model. Diagnose — When a competitor ranks higher, we investigate why. We look at content, website structure, and other signals the AI is picking up on. Optimize — We turn those findings into a clear action list — specific content and technical fixes BetterHelp can make to improve how AI models recommend them. We also felt the weight of the use case. When someone asks an AI for therapy help, they're often in a vulnerable moment. The platform that wins that recommendation carries real responsibility.

What's next for BetterRank

Right now, BetterRank measures how one AI sees BetterHelp. But people are searching across ChatGPT, Claude, Perplexity, and more — all day, every day. The next step is simple: expand BetterRank to run the same audits across every major AI model simultaneously. So BetterHelp doesn't just know their score on one platform — they know their score everywhere, and can see exactly where they're winning and where they're losing ground. One dashboard. Every AI. No blind spots.

Built With

• braintrust
• claude
• gemini
• react
• tavily
• typescript
• v0