AI clinical evidence search that shows its work
ClinicalEvidence searches an index of registered trials and published studies, writes an answer only from the passages it retrieves, verifies every citation and tells you plainly when the evidence is not there. Here is each step, in the order it happens.
An index of registered trials and published studies, refreshed every day
Every answer starts from our own index, not from what a language model remembers. The index brings together the public trial registries and the biomedical literature in one place, so a single question reaches trial records and papers at the same time.
- ClinicalTrials.gov registry records, including status, phase, arms, endpoints and posted results
- PubMed citations and abstracts
- The open access subset of PubMed Central, with full text where the licence allows
- The EU Clinical Trials Register, covering EudraCT and CTIS records
- The WHO International Clinical Trials Registry Platform, where the licence allows
The index is refreshed daily. New registrations, status changes, posted results and newly indexed publications reach your answers within a day. Each record keeps its original identifier, an NCT number, an EU trial number or a PMID, so every source you see links back to the registry or literature database it came from. Read more about the clinical trials database behind the answers.
Semantic and full text retrieval for clinical questions
You type a question the way you would ask a colleague, for example whether a class of drugs reduces hospitalisation in heart failure with preserved ejection fraction. ClinicalEvidence runs two searches in parallel. A semantic search finds passages that mean the same thing even when the wording differs, such as "HFpEF" and "heart failure with preserved ejection fraction". A full text search finds exact drug names, gene names, outcome terms and identifiers.
The results are merged and ranked, then narrowed by the filters you set: condition, intervention, phase, status, study type, date range, sponsor type, country and whether results are available. Only the best matching passages move on to the next step.
Answers written only from retrieved passages
The language model receives your question and the retrieved passages, nothing else. It is instructed to write each sentence from those passages and to attach the identifier of the source it used. It is not allowed to add facts, figures or studies from its general training.
This is the core design decision. A general chatbot answers from memory and can produce a convincing reference to a study that does not exist. ClinicalEvidence can only cite what it has just retrieved from the index, so every claim has a record behind it.
Citation verification before you see the answer
Before an answer reaches your screen, an automatic check reads it sentence by sentence.
- Every cited NCT number or PMID must exist in the index
- The cited passage must actually support the sentence it is attached to
- A sentence that fails either check is removed
- An answer without any verified source is not shown at all
What you read is what passed verification. Each sentence carries one or more identifiers, and one click opens the registry record or the abstract so you can judge the study yourself. See the detail in cited AI answers.
When there is no evidence, the answer says so
Some questions have no trial or publication behind them yet. In that case ClinicalEvidence does not improvise. You get a plain statement that the index holds no evidence that answers the question, together with the closest matching studies when there are any, so you can see what does exist.
This behaviour is deliberate. An empty answer is useful information for a protocol designer scanning the landscape, for a medical information specialist replying to a request, and for a researcher deciding whether a systematic review is feasible. A fluent answer without sources would be worse than none.
The sources searched panel
Every answer comes with a sources searched panel. It lists only the sources that were actually searched for that question, together with the filters that were applied. If you limited the question to registered trials, the panel shows the registries and not the literature databases. If you set a phase or a date range, the panel shows that too.
The panel gives you the audit trail a reviewer or a colleague will ask for: where the answer looked, with which limits, and which records it found. You can open the full list of matching studies in the clinical trial table.
Limits of the tool you should know
ClinicalEvidence is evidence support for professionals. It is not medical advice and it is not a diagnostic device.
- It reports what registered trials and published studies say. It does not tell you what to do for an individual patient.
- It covers the sources listed above. Evidence that exists only in conference abstracts, internal reports or paywalled full text outside the open access subset may not be in the index.
- Registry records are entered by sponsors and investigators. If a record is incomplete or out of date, the answer reflects that record.
- Verification checks that a cited passage supports a sentence. It does not judge study quality for you, so the level of evidence is shown next to each source to help you weigh it.
- Answers are limited by your plan, as described on the pricing page.
Always open the sources before you rely on an answer in clinical, regulatory or commercial work.
Questions about AI clinical evidence search
Can ClinicalEvidence answer from its own knowledge if the index has nothing?
No. Answers are written only from passages retrieved from the index. When nothing relevant is found, the answer says so plainly and shows the closest matching studies if there are any.
How do I know a citation is real?
Every NCT number and PMID is checked against the index before the answer is shown, and the cited passage must support the sentence. Each identifier links to the source record so you can open it yourself.
How current is the index?
The index is refreshed every day. Newly registered trials, status changes, posted results and newly indexed publications appear in answers within a day.
Do I need to enter patient data?
No. You ask questions about published research and registered trials, so no patient data is needed. Read more on the security page.
Ask a clinical question and check every source
Try the one-time demo on the homepage, or create your account and choose a plan.