One stealable ChatGPT/Claude app idea per demand-map vertical built from real usage data.

Companion to two editions of The Diff: the demand map (what people actually ask AI for help with, from Anthropic's study of 1M conversations) and this week's post on the chat-window app store. The map told you what to build. This library gives you a running start on each vertical.

How to use it: each entry has six parts. Skim for the vertical that matches your skills or interest, then take the whole entry into the Launch Kit and build. These are starting points, definitely not finished specs.

How each entry works

• The demand signal — what people are actually asking, with the numbers from the study
• The micro app — the atomic, one-job-per-turn product that answers it
• Tools to define — the 2–3 MCP tools, named in problem-language (this is your ranking surface)
• Widget format — inline card / fullscreen / picture-in-picture, and why
• Golden prompts — real-user sentences that should trigger your app (steal their phrasing, then write 15 more)
• The funnel + watch out — how it converts before in-chat payments exist, and the honest risk

1. Health & wellness — the Lab Decoder 🩺

The demand signal: The biggest domain in the study — 27% of all guidance conversations, 81% rated high-stakes, and the lowest pushback rate anywhere (2.9%): people trust health answers and act on them. Sub-topics named in the study: interpreting test results, chronic condition management, injury triage, calorie/macro tracking. Macros is taken (Cal AI → MyFitnessPal, already on the ChatGPT shelf). Test results are not.

The micro app: Paste your bloodwork or test results. An inline card renders each marker against its reference range in plain English, flags what's off, and generates the questions to ask your doctor. Connect an account and it remembers your history — the trend line between two lab panels is worth more than either panel alone.

Tools to define:

• interpret_lab_results — "Explains blood test and lab results in plain language, flags values outside reference ranges, and suggests questions for your doctor."
• compare_lab_history — "Compares new lab results against the user's saved history and shows trends per marker."

Widget format: Inline card for the decode; fullscreen for history trends.

Golden prompts: