How Super Founder works

Super Founder answers startup questions from evidence. It shows its sources, says what it checked, and says what it could not.

How a Research answer is built

Search. The AI turns your question into several search phrasings and gathers up to about 100 candidate papers from OpenAlex. When semantic ranking is on, papers are also ranked by meaning and by a reranker, and papers from earlier searches can be recalled.
Screen. The AI rates the 40 best candidates for how directly they answer your question and labels each study's design (meta-analysis, experiment, survey and so on). The best 20 are analyzed.
Read. For the top papers that are open access, the app reads the results section. The rest are read from their abstracts.
Weigh. Each paper is marked Yes, Possibly or No for your question. The evidence split counts stronger designs more: a meta-analysis outweighs a survey.
Check. Every key finding must come with a quote from the paper. The app confirms the quote really appears in the abstract or excerpt, and drops any finding it cannot confirm.

What is checked, and what is not

Checked by code: that each finding's quote exists in the source text; the counts, percentages and the evidence split; that citations point to real papers in the result.

Not checked: the AI's judgment of a paper's relevance, design and stance; anything found by web search (official data, market gaps, company lists), which is linked but unverified; whether papers the search never retrieved would change the answer.

Evidence confidence

High needs at least 12 papers, at least 3 strong-design studies and at least 70% agreement. Medium needs at least 8 papers and a clear lean. Anything less is low, and the result panel says what is missing. A low rating is the app being honest, not a failure.

The other tabs

Unmet needs: groups complaints from Hacker News, Stack Exchange and GitHub into themes. Percentages are shares of the posts sampled, not of all users.
Patents to products: turns recent granted US patents into product concepts. It is not a legal or freedom-to-operate opinion.
Market gaps: uses live web search and labels its findings as unverified.
Idea check: weighs an idea you describe across demand, competition, feasibility, regulation and business model. Every reason must cite gathered evidence. If fewer than three of the five areas have source-backed evidence, the answer is "not enough evidence" instead of a guess. It is not a prediction of success.
Startup ideas: proposes ideas that must cite the evidence gathered; each shows how many sources of how many types support it.

Limits

Answers come mostly from abstracts, search is not exhaustive, and the AI can misjudge a paper. Treat results as a well-sourced starting point, not professional, legal, financial or medical advice.

Your data

See the Privacy Policy. Searches are private unless you create a public link.