Most health claims on the internet are not evidence-based. A supplement that “boosts immunity,” a diet that “detoxifies,” a practice that “restores balance”—these claims sound good but rarely withstand scrutiny against actual research. The gap between marketing confidence and scientific evidence is vast.
Checking whether a health claim is supported requires knowing where to look for reliable information and how to evaluate it. This seed catalogs the best tools and databases for distinguishing marketing from evidence.
Why This Matters
The Default is Marketing, Not Evidence — Google’s top health results are often from supplement companies, wellness blogs, and health gurus with financial incentives to convince you. Peer-reviewed research is often buried. The incentive structure rewards confident claims, not honest uncertainty.
Medical Authority Doesn’t Guarantee Accuracy — An MD or “naturopath” or “wellness expert” expressing confidence in a claim is not evidence the claim is true. Appeal to authority is common in health marketing. Evidence is what matters.
Research is Messy — Individual studies contradict each other. Statistical significance differs from practical significance. Correlation is not causation. Understanding a single study requires context: How does it compare to other research? What are its limitations? What does the preponderance of evidence show? This is why systematic reviews exist.
Your Time is Valuable — Searching PubMed yourself and reading papers takes hours. Evidence synthesis tools do this work so you don’t have to.
The Five Essential Tools
1. PubMed (pubmed.ncbi.nlm.nih.gov)
The National Library of Medicine’s search engine for biomedical literature. Over 37 million citations, most with free abstracts.
How to use it: Search any health claim. PubMed returns peer-reviewed research on the topic. You can filter by study type (clinical trials, systematic reviews are higher quality) and date.
Limitations: The sheer volume can be overwhelming. Reading abstracts requires some scientific literacy. Individual studies are often small, contradictory, or limited. Don’t rely on one study; look for consensus across many.
Best for: Finding what researchers have actually studied on a topic. Answering “has this been studied?” and “what do researchers think about this?”
2. Cochrane Library (cochranelibrary.com)
Systematic reviews synthesizing the best available evidence on specific health questions. Widely considered the gold standard for clinical evidence summaries.
How to use it: Search a health question. Cochrane returns systematic reviews that synthesize dozens or hundreds of studies on that topic, with a plain language summary of findings.
Quality: Each review is conducted by experts, follows rigorous methodology (PRISMA guidelines), and includes a risk-of-bias assessment. A Cochrane review saying “insufficient evidence” is more trustworthy than a website confidently claiming effectiveness.
Access: Every review includes a free plain-language summary. Full-text access varies but is often available through public library databases.
Best for: Understanding what the totality of evidence shows on a clinical question. When you want the consensus, not individual studies.
3. NIH’s National Center for Complementary and Integrative Health (nccih.nih.gov)
Evidence-based summaries of supplements, herbs, and alternative health practices. Each entry details what research shows, what’s inconclusive, and safety concerns.
How to use it: Search a supplement or practice. NCCIH provides a summary of the evidence: which conditions it’s been studied for, what the research shows, what’s still unknown, and what safety issues exist.
Honesty about uncertainty: NCCIH is good at saying “there’s not enough evidence yet” rather than either endorsing or condemning. This reflects reality better than most sources.
Best for: Supplements, herbs, acupuncture, meditation, and other complementary practices. A balanced source that takes research seriously without dismissing traditional practices out of hand.
4. Examine (examine.com)
An independent database synthesizing research on supplements, nutrition, and health topics. Free tier covers the core evidence summaries.
How to use it: Search a supplement or nutrient. Examine provides a research summary: the proposed mechanism, what studies exist, effect sizes, and practical takeaways.
Strength: Examine synthesizes evidence in a digestible format without corporate bias (no supplement company owns it). Effect sizes show whether a supplement helps a little or a lot, not just whether research shows any effect.
Transparency: Every claim links to the studies supporting it. You can drill down from “this supplement may improve focus” to the actual research.
Best for: Supplements and nutrition. If you want to know whether a supplement is worth taking, Examine is faster than PubMed and more thorough than marketing.
5. MedlinePlus (medlineplus.gov)
The NIH’s consumer health library. Written in plain language, reviewed by medical professionals, and free of advertising.
How to use it: Search a health condition or treatment. MedlinePlus explains the condition, how it’s diagnosed, treatment options, and related resources, all in accessible language.
Quality baseline: MedlinePlus is a better starting point than Google for health queries. It’s not trying to sell you anything.
Limitations: It’s a reference, not a substitute for professional diagnosis. But for understanding a condition or treatment, it’s reliable.
Best for: Getting oriented on a health topic. Understanding what a condition is, how it’s diagnosed, and what treatments exist. A trustworthy first source.
How to Use These Tools Together
You have a health claim you want to check: “Vitamin C prevents colds”
- Start with MedlinePlus or NCCIH — Get oriented. What does the evidence summary say?
- Check Cochrane Library — Is there a systematic review on this? What does it conclude?
- Search Examine (for supplements) or PubMed (for general questions) — Dive deeper into the evidence.
- Form a conclusion — What does the preponderance of evidence show? What’s inconclusive?
Key principle: If five sources agree on a claim, it’s probably solid. If they disagree or say “inconclusive,” be skeptical of confident marketing.
Red Flags in Health Marketing
- Extreme confidence with no uncertainty. Real medicine and nutrition involve trade-offs and unknowns.
- Testimonials instead of evidence. “It worked for me” is not evidence it works in general.
- Conspiracy thinking. “Big Pharma doesn’t want you to know…” is usually a sign the claim isn’t evidence-based.
- Appeals to nature. “Natural” doesn’t mean safe or effective. Arsenic is natural.
- One study makes the case. Good evidence is replicated across many studies by independent researchers.
- No discussion of limitations. All research has limitations. If none are mentioned, be suspicious.
Project Ideas
1. Health Claim Fact-Checker Website
What it does: User enters a health claim. The tool searches PubMed, Cochrane, Examine, and NCCIH simultaneously, synthesizes findings, and presents a verdict: “Well-supported,” “Mixed evidence,” “Insufficient evidence,” or “Contradicted by evidence.”
Technical approach:
- Backend: Python/Node.js with APIs to PubMed, Cochrane (both have free APIs)
- For tools without APIs: web scraping or RSS feeds (Cochrane publishes summaries)
- LLM layer: Use Claude to synthesize findings across sources into plain English
- Frontend: React or Vue for search interface and results display
Key features:
- Shows the actual evidence, not just a verdict
- Links to source documents
- Explains what “sufficient evidence” means (how many studies, effect size, etc.)
- Flags when a claim hasn’t been researched
- Shows conflicting findings across sources
Challenge: Building accurate evidence synthesis is harder than it sounds. An LLM can hallucinate or oversimplify. Validation against human experts would be essential.
Impact: A tool that transparently shows what the evidence actually says would be genuinely useful—cut through marketing noise in health queries.
2. Claude Skill: “Health Evidence Checker”
What it does: A Claude extension that users can invoke during a conversation to fact-check health claims in real-time.
How it works:
- User: “Is vitamin C really good for preventing colds?”
- Claude uses the skill to query health evidence sources
- Claude reports back: “Here’s what the research shows…”
Technical approach:
- Build a Claude skill using the Claude API and tool use
- Tools available: PubMed search, Cochrane Library queries, NCCIH lookups, Examine summaries
- Claude orchestrates the queries, interprets results, and synthesizes into conversational response
Key features:
- Integrated into conversation flow (no need to leave chat)
- Cites sources with links
- Explains limitations (“only 3 studies,” “small sample size,” etc.)
- Flags when the evidence is genuinely inconclusive
Advantage over website: Lower barrier to entry (lives in Claude), no new interface to learn, faster iteration
Challenge: Skill tools have latency (waiting for API responses); need to handle slow searches gracefully.
Impact: Makes Claude more reliable for health questions; users can verify claims within conversation.
3. Browser Extension: Evidence Badge
What it does: When you’re reading a health article or claim online, the extension queries evidence databases and shows a badge indicating research support.
How it works:
- User visits a wellness blog claiming “probiotics cure IBS”
- Extension detects the claim, queries Cochrane + Examine
- Badge appears: “Mixed Evidence” (3 small studies show mixed results)
Technical approach:
- JavaScript extension (Chrome, Firefox)
- NLP to detect health claims in page text
- Background API calls to evidence sources
- Injects badge into page
Challenge: Automatically extracting claims from text is hard. False positives (flagging things that aren’t claims) would be annoying.
Impact: Helps readers immediately check claims without leaving the page.
4. Evidence Database + RAG System
What it does: Build a local knowledge base of health evidence (using web scraping and APIs), then use RAG (Retrieval-Augmented Generation) to answer health questions grounded in research.
How it works:
- Scrape and index Cochrane reviews, NCCIH summaries, and Examine findings
- User asks a health question
- RAG system retrieves relevant evidence documents
- Claude reads the documents and answers based on evidence
Technical approach:
- Backend: Vector database (Pinecone, Weaviate, or local Chroma)
- Embeddings: Use Claude’s embedding API
- Pipeline: Scrape sources → chunk text → embed → store in vector DB
- Frontend: Simple chat interface
Advantage: Fully grounded in evidence; can cite specific studies and summaries; works offline if cached
Challenge: Keeping evidence updated (health research moves fast); handling conflicting evidence across sources
Impact: A dedicated health evidence chatbot that’s more reliable than general-purpose AI for health questions.
Which Project to Start With?
Easiest (fastest to MVP): Claude skill (no frontend, API integration only)
Most useful: Health claim fact-checker website (directly addresses the problem; no Claude dependency)
Most ambitious: Evidence database + RAG (most work, but most powerful and grounded)
Best if you enjoy: Browser extension (UX-focused; lower backend complexity)
All of these would solve a real problem: making health evidence accessible and checkable in a world of marketing noise.
Related Notes
- Scientific Research — How science builds knowledge and evaluates evidence
- Health — Broader health and wellness information