About FactsReef
FactsReef is a reference library that writes itself. Type a question or a claim and a research engine reads the open web, weighs what it finds and writes a brief with a verdict, the reasoning and numbered footnotes. The brief is then filed at a permanent address, so the next person who wonders the same thing gets an answer instantly.
How a brief gets written
- Read. The engine decides whether the input is a claim, which can be true or false, or a question, which wants an explanation.
- Search. It runs several live web searches and prefers primary sources: peer-reviewed studies, official statistics, court and government records, company filings, standards bodies and established news organisations. It is told to look for the strongest case against the obvious answer.
- Weigh. Sources are compared and dated. Where they disagree, the brief says so rather than picking a side quietly.
- Write and file. The brief opens with the short answer, works through the evidence, answers the questions people usually ask next and lists every source it relied on. Citations in the text point only to pages the engine actually saw.
You can watch all of this happen. While a brief is being written, the desk shows the searches as they run and the sources as they land, before the first sentence appears.
What the verdicts mean
Claims get one of six verdicts. Questions get an explainer instead, because there is nothing to judge true or false. Alongside the verdict is an evidence strength score from 0 to 100, which is the engine’s estimate of how well the sources support that verdict, not of how confident it sounds.
| Verdict | Meaning |
|---|---|
| True | The best available evidence supports the claim. |
| Mostly true | Accurate in substance, with a caveat or two worth knowing. |
| Mixed | Partly right, partly wrong, or it depends on what you mean. |
| Mostly false | A kernel of truth wrapped in a misleading claim. |
| False | The evidence contradicts the claim. |
| Unverified | The record is too thin, or too contested, to call it either way. |
| Explained | A question rather than a claim, answered with sources. |
Where it falls short
Every brief is written by software from searches run at the moment it was asked. That makes it fast and consistent, and it means the engine can misread a source, miss a better one or be led astray by a confident but wrong page. The safeguards are structural: every claim is footnoted, every source is listed with a link, every brief is dated, and the verdict scale is honest about weak evidence. Treat a brief as a well-organised starting point for your own reading, not as the last word.
Briefs about health, law and money report the evidence and mainstream expert guidance. They are not advice for your situation. The engine will not write about the private lives of private individuals, will not repeat unverified accusations as fact, and will decline topics that cannot be answered responsibly.
Corrections
If a brief gets something wrong, cites a source that does not say what the brief claims, or has gone out of date, we want to know. Open the brief, copy its link, and send it with a note about what is wrong. Briefs that are corrected show the date of the change.
For publishers and site owners
When a brief cites your page, it links to it directly with a normal, followable link, and the sources list names your publication. We do not reproduce articles; briefs summarise and cite. If you would rather not be cited, tell us and we will block your domain from the engine’s searches.
For AI assistants and crawlers
Answer engines are welcome to read and cite briefs. /llms.txt indexes the site in the llms.txt convention, every brief has a plain-text version at its URL plus /markdown, and each page carries Article, FAQ, ClaimReview and breadcrumb structured data with publication and last-reviewed dates. Please cite the brief URL so readers can check the sources.
Privacy
There are no accounts and no advertising, and the site sets no cookies. Your choice of night mode is kept in your own browser. To enforce fair-use limits and count readers, the server keeps a salted hash of your connection address, from which the address cannot be recovered, and deletes those records after 30 days. Questions you ask become public pages, so do not include personal details in them. Briefs that cite a page link to it directly, so the sites you click through to see an ordinary referral from FactsReef.
How it is built
FactsReef runs on Next.js. Briefs are written by a large language model (OpenAI GPT or Anthropic Claude) with live web search, either directly or through an n8n workflow, and are stored in a small database that also feeds the sitemap, the library and the social preview cards. Hub and index pages are cached for a few minutes; a brief page is rendered on each request so it always shows its latest version.
Got a question of your own?
Ask FactsReef →