Evidence Verification
Upload a supplement label. Eleven AI agents search PubMed, analyze clinical evidence, and produce a verification dossier with every citation traced to a specific PMID. Deterministic scoring. Zero hallucination risk on numbers.
How It Works
Upload a supplement facts panel or product label image. Our intake agent parses every ingredient, dose, and unit automatically.
Eleven specialized AI agents search PubMed, cross-reference clinical studies, and analyze dose adequacy, safety, and bioavailability.
Deterministic scoring engine calculates your Scientific Credibility Score across four clinical dimensions. No LLM subjectivity in numbers.
Receive a publication-ready verification dossier with executive summary, ingredient analysis, citations, and embeddable trust badge.
Agent Pipeline
Each agent handles a specific domain of clinical analysis. Ten run on Claude Haiku for speed. Synthesis uses Claude Sonnet for depth. The Classifier is fully deterministic.
Parses label, extracts ingredients, normalizes units and aliases
Categorizes each ingredient by functional role without LLM calls
Tiered PubMed/PMC search retrieving 30+ studies per ingredient
Evaluates dose adequacy against published clinical thresholds
Checks upper limits, contraindications, and adverse event signals
Verifies marketing claims against the clinical evidence found
Flags sub-clinical doses designed to pad the label
Checks claims against FTC Health Products Guidance and FDA rules
Validates Certificate of Analysis documents against label claims
Whole-formula narrative synthesis using Claude Sonnet for depth
Score-narrative consistency check and citation integrity audit
Scoring Engine
A composite 0-100 score calculated with deterministic Python logic. AI agents produce the narrative analysis; the scoring engine produces the numbers. No LLM involvement in any numeric output.
Study design, sample size, statistical significance, and replication across the literature
Product dose vs. clinically effective ranges from published RCTs and meta-analyses
Ingredient form quality, absorption enhancers, and delivery mechanism effectiveness
Upper intake limits, adverse event profiles, drug interactions, and contraindications
Scores are computed with pure Python math. No LLM generates, adjusts, or influences any numeric output. The same inputs always produce the same score.
Dossier Output
Every dossier is a defensible evidence package. Share it with retailers, regulatory counsel, or publish it for consumers. Export as PDF or Excel workbook.
High-level overview of product quality, composite score, grade, and key findings for stakeholders.
Per-ingredient breakdown with dose assessment, form evaluation, evidence strength, and clinical relevance.
Full citation table with PMIDs, study types, sample sizes, outcomes, and relevance scores for every ingredient.
Upper limit checks, contraindication flags, interaction warnings, and adverse event monitoring results.
Each marketing claim mapped to supporting evidence with substantiation strength ratings.
Verification status determination with embed code for your website and public dossier URL.
PDF Dossier
Publication-ready evidence report
Excel Workbook
Ingredient-level evidence data
Trust Signals
Products that pass verification earn an embeddable trust badge. One line of code adds a clinically-verified signal to your product page that links directly to the public evidence dossier.
<script src="https://tci.ai/embed.js"></script>Free to start. No credit card required.