Most cannabis sites are anonymous, unsourced, and unverifiable. Some are written by AI but pretend they aren't. Some are written by humans copying from each other.
Weedpedia is built to be checkable. The generation history of every article — drafting model, fact-check pass, and re-audit record — is visible from the article itself.
The stages
Drafting assistance creates the first version
Anthropic's Claude is given a topic and the Weedpedia editorial guidelines. It produces a structured draft in which every factual claim is attached to a citation tier (peer-reviewed, government, reported, book, practitioner) and every uncertain claim is labeled as such (strong / weak / anecdote / disputed / no-data).
The draft is not published. It goes directly to stage 2.
An independent fact-check pass reviews the draft
OpenAI's GPT reads the Claude draft as an independent reviewer and produces a structured fact-check report: which claims look unsupported, which sources look hallucinated, which evidence labels look overstated, and whether the article is safe to publish.
If the fact-check flags critical problems, the article is held as a draft until a human spot-checks it. Otherwise, it publishes automatically.
Old articles get re-audited on a schedule
Every published article gets re-read by the fact-check agent roughly every 90 days. The agent looks for outdated facts (especially legal status), newly contested claims, dead source links, and internal inconsistencies. New issues are added to the article's flag list, visible to the operator.
A human operator spot-checks weekly
One human (the site operator) reviews the previous week's newly-published articles and the auto-generated flag reports. The operator can approve held drafts, regenerate broken articles, retire articles that turned out badly, and add new topics to the queue.
The operator does not write articles. The operator is a quality control gate, not an author.
What this means in practice
For most articles — strains, terpenes, cultivation, history — this pipeline produces something noticeably more rigorous than the typical cannabis-website article, because it cites sources and labels its evidence honestly.
For medical articles — anything involving dosage, drug interactions, or health claims — the pipeline holds the draft for human approval. Hallucinated medical claims are the kind of mistake that hurts people. This is the liability firewall.
Things we don't claim: that this is as reliable as a textbook reviewed by domain experts. It is not. We are honest about that on every page. What we do claim is that it is more transparent, better-sourced, and more calibrated than the unsigned anonymous content that dominates this topic on the web today.
If you find an error
Click the orange "Flag an error" button on any article and tell us what's wrong. Operator response within one week. Confirmed errors trigger regeneration; the old version stays visible in the generation history.