How does the AI matching work?

GPT-4o extracts medical concepts from your description; we run BM25 + MedCPT hybrid search over the full ClinicalTrials.gov index; GPT then ranks candidate trials for relevance with a one-sentence explanation per trial.

MedCPT is a biomedical neural embedding model that captures medical-concept similarity. It surfaces trials whose terminology differs from yours but whose clinical meaning is the same.

We do not require an account, do not store personally identifying information, and do not sell or share search queries. Condition descriptions are sent only to the LLM providers and the trial-search backend powering matching.

Home › How Find a Trial Works

How Find a Trial Works

You describe your condition in plain English. Our system extracts the relevant medical concepts, runs hybrid keyword + semantic search across every active study on ClinicalTrials.gov and NIH RePORTER, and uses GPT to score each candidate trial for relevance to your specific situation. Results are returned in seconds and explained in patient-friendly language. The service is free; we do not sell patient data.

The matching pipeline, step by step

You describe your situation. Diagnosis, stage, prior treatments, age, location. The more detail you give, the better the match.
Medical concept extraction. GPT-4o reads your description and pulls out the medically meaningful entities — condition, modifiers, anatomy, biomarkers, prior therapies.
Hybrid retrieval. We query our index of every ClinicalTrials.gov study using two complementary methods:
- BM25 — a classical keyword-relevance algorithm that surfaces trials whose protocol text contains your terms.
- MedCPT — a domain-specific neural embedding model trained on biomedical literature that surfaces trials that are conceptually similar even when the wording is different.
AI ranking. The top candidates are scored by GPT for clinical relevance to your case, with each trial's verdict explained in one or two sentences.
Filtering. Location filter (worldwide, US, US&Canada, Europe), study type filter (clinical trials, NIH research, both), and recency are applied to the ranked list.
Results in plain English. Each trial card includes a simplified summary, eligibility highlights, study locations, sponsor, phase, and a direct link to the full ClinicalTrials.gov record.

What makes our matching different

Semantic understanding. "Heart attack" finds trials labeled "myocardial infarction" or "STEMI" because the embedding model understands medical synonymy.
Whole-protocol search. We search the full eligibility criteria, study description, and outcome measures — not just trial titles.
Patient-language summaries. Every trial gets an AI-generated plain-English summary so you do not have to parse medical jargon to decide whether a study is relevant.
Updated daily. New trial listings and status changes are indexed within 24 hours.

What we do not do

We do not provide medical advice. We connect you to trials; your doctor and the research site decide eligibility.
We do not enroll you in a trial. Enrollment is between you and the research site.
We do not share your search queries or contact information with third parties.

Privacy

Find a Trial does not require an account, does not require you to enter personally identifying information, and does not transmit your condition description to anyone except the LLM providers (OpenAI / Google) and the trial-search backend that power matching. See our privacy policy for details.

Try it now →

Last reviewed 2026-05-23 by the Find a Trial editorial team. Information on this page is for educational purposes and is not medical advice. Always consult qualified healthcare professionals about clinical trial participation.