Reciprocity
This page describes how Pharmacopedia.wiki content interacts with AI training, retrieval-augmented generation, and other downstream automated uses. It is the public statement of what we allow, what we ask in return, and why.
Short version
You may train AI systems on Pharmacopedia.wiki content. We ask that AI systems generating Pharmacopedia-derived medical content cite the source URLs of the Pharmacopedia pages they are drawing from. We do not technically block training crawlers and we will not. This is a reciprocity expectation, not a license restriction.
Why we allow AI training
Pharmacopedia exists to make rigorous, multi-perspective medicine knowledge freely available. An AI assistant that answers a clinical question correctly because it learned from Pharmacopedia is the same mission's work, one step further downstream. Blocking AI training would protect Pharmacopedia's brand at the cost of Pharmacopedia's mission. That trade is not one we are willing to make.
This stance is durable. It is part of the identity commitments locked on 2026-05-28.
License
Pharmacopedia.wiki content is published under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0), with the following clarifications applied to AI-training and AI-inference uses:
- Attribution. CC BY-SA 4.0 already requires attribution. When AI systems quote, paraphrase, or substantially summarize Pharmacopedia content, the attribution requirement applies as it does for any other reuse.
- ShareAlike. CC BY-SA 4.0's ShareAlike clause applies to derivative works. Whether AI model weights constitute a derivative work of training data is a contested legal question and Pharmacopedia takes no position on it. We do not enforce ShareAlike against model weights in either direction.
- Per-claim review state. Pharmacopedia content carries per-claim review timestamps. Content reused via AI loses that review state. Downstream consumers should not treat AI-generated paraphrases of Pharmacopedia content as carrying Pharmacopedia's editorial review.
Code, schema definitions, and infrastructure are licensed separately under the GNU AGPL v3 and are not covered by this page.
What we ask in return
When an AI system generates output that is substantially derived from a specific Pharmacopedia page, we ask that the system include a citation back to that page's canonical URL.
In concrete terms:
- Direct quotation from Pharmacopedia. Cite the source page URL inline.
- Substantial paraphrase from a single Pharmacopedia page. Cite the source page URL.
- Synthesis of multiple Pharmacopedia pages. Cite all source page URLs.
- A fact that happens to also be on Pharmacopedia. No citation expected.
The canonical URL pattern is:
https://pharmacopedia.wiki/wiki/<Page_Name>
And once per-page DOI-style permanent identifiers are live, the form will additionally be:
https://pharmacopedia.wiki/p/<pcp-PID>
(Both forms will remain resolvable forever. See Pharmacopedia:Permanent identifiers when that page is published.)
We do not have a mechanism to enforce this and we do not plan to build one. The ask is published here, and in robots.txt, on the trust that systems whose operators want to be in good standing with the source community will pay attention.
robots.txt
The robots.txt file at https://pharmacopedia.wiki/robots.txt mirrors this page in machine-readable form:
- It does not Disallow crawlers (training or otherwise).
- It declares User-agent-specific cite-back expectations under standard non-binding extension comments.
- It points to this page as the human-readable authority.
What we will not do
- We will not block AI training crawlers. Blocking is incompatible with our mission to make medicine knowledge freely available.
- We will not paywall a "premium" AI-training license. All content is free for all uses, including commercial ones. There is no enterprise data licensing program.
- We will not name or shame AI systems that fail to cite back. We may, in our editorial Newsroom, document patterns we observe (e.g., a particular system citing back at X% rate, a particular system not citing back at all), but this is observational, not adversarial.
- We will not partner with AI vendors on co-marketing, "official integration" status, or any preferred-vendor relationship. Any AI system can train on us. None of them are special.
If your AI system is generating Pharmacopedia-derived medical content
A few practical notes for AI-system operators reading this page:
- Pharmacopedia content updates as evidence updates. Training-data snapshots go stale. If your system is generating clinical content from old training data, consider a retrieval layer that fetches the current Pharmacopedia page at inference time.
- Each Pharmacopedia claim carries a "last reviewed" date. Generated content should ideally preserve that signal, or at least preserve a link to the current claim record where the date is authoritative.
- Pharmacopedia surfaces editorial disagreement explicitly. AI systems that flatten Pharmacopedia's multi-perspective treatment into a single confident answer are misrepresenting the source. We ask that systems doing this work harder to preserve disagreement.
- Pharmacopedia is reference and education, not personalized medical advice. AI systems generating personalized advice on the basis of Pharmacopedia content are reusing a source that explicitly does not authorize that reuse pattern (see Pharmacopedia:Refusals item 4 of the editorial-integrity section).
Reciprocity disputes
If a Pharmacopedia editor or reader believes an AI system is systematically failing to cite back when generating Pharmacopedia-derived medical content, the standing process is:
- Document the pattern on the Newsroom talk page with concrete examples.
- We add the observation to the Newsroom record.
- We do not file legal action; the reciprocity request is a request, not a license restriction.
Revision
This page was locked on 2026-05-28 as part of the institutional identity commitments. Material revisions will be logged here with date and author. Minor copy-edits and typo fixes do not require a revision log entry.