Evidence, graded and traceable.

1Who we are

gpt-dict.com is a scientific evidence database for dietary supplements.

For every "intervention × condition" pairing (for example, "saw palmetto × benign prostatic hyperplasia"), we build a structured evidence summary drawing on sources such as PubMed, FDA / EMA / TFDA, Cochrane Reviews, Examine, and Taiwan's PTT / Dcard communities, and quantify the strength of that evidence across 6 tiers (S / A / B / C / D / U).

2Why we built this

Taiwan's supplement market suffers from severe information asymmetry:

• Vendor marketing blends real and pseudo-science
• Most media coverage is sponsored content, citing a single study to declare a product "miraculously effective"
• Anecdotes on PTT / Dcard vary wildly in quality and are hard to map onto RCT results
• English-language resources like Examine and Cochrane are too difficult for non-medical readers

What we set out to do is simple: aggregate all the credible-source evidence on a single topic onto one page, present it by tier, and fully expose the underlying PubMed citations and safety warnings. If the evidence says "ineffective," we honestly write "ineffective." If the evidence says "effective but with side effects," we spell out the side effects too.

3What we don’t do

• We don’t sell supplements — no iHerb affiliate links, no sponsored posts, no e-commerce
• We don’t replace medical advice — for any decision to start or stop a supplement, please consult a physician or pharmacist
• We don’t cover trending new products — we focus on mature topics where there is "enough evidence to render a judgment"
• We don’t accept outside submissions — to ensure quality, all content is reviewed by our in-house team

4Engine architecture

Behind every claim page is the output of a 7-layer evidence engine built automatically:

Each layer is fetched and verified by an independent sub-agent; every PubMed citation is cross-checked against NCBI to confirm the PMID genuinely exists before publication. See the full methodology.

Engine version: v1.0 · Build date: 2026-06 · Current coverage: 194 interventions × 162 conditions × 717 claims

5For creators / developers

Our data is fully structured:

• Every claim page ships with corresponding schema.org/ClaimReview JSON-LD (directly machine-readable by search engines and LLMs)
• The engine's JSON snapshot output is openly queryable
• Health-content creators are welcome to cite our claim pages as an evidence source—just remember to add a canonical link

6Team

Led by Lenny Chen, with multiple AI sub-agent engines automating the build and maintenance. We operate a series of health-focused media sites, but this site (gpt-dict.com) stands apart from that commercial content and does evidence aggregation only.

7Contact

Feedback, error reports, or requests for a supplement you’d like us to cover: lenny0929@gmail.com

8Disclaimer (summary)