Sign in

How summaries are made

What this site is for

Information for health professionals about publicly registered clinical trials, published research and research sites, and about how to contact them. It does not assess whether a patient is eligible, and it gives no diagnosis or treatment advice.

Where the data comes from

Trials come from public registries: ClinicalTrials.gov and EU CTIS today; WHO ICTRP is coming. Research comes from PubMed and OpenAlex. Every page links to the record it was built from and shows when it was last checked.

What the machine writes

Short plain-language summaries are written by a language model from the registry and publication text. Numbers, trial status, places, phases and dates never come from the model: a template fills them in from the registry record. Texts written this way carry the label “Machine summary”.

The checks before publishing

A machine summary is published only if it passes every check below. If one fails, the page stays out of search engines and a person looks at it.

  1. No numbers in model text. Counts, status, places, phases and dates come from templates.
  2. Every statement has a source, and each value taken from the source must really appear in it.
  3. Two separate automatic auditors check that the sources support the text. If they disagree, the page is hidden from search until a person decides.
  4. Words that judge or promise are blocked, such as “eligible”, “suitable for you”, “promising”, “breakthrough” or any promise of a cure.
  5. Patient pages use plain language: at most 15 words in a sentence, frequencies as “3 out of 10”, and a readability score check.
  6. Translations get the same checks and a medical glossary. A language is offered to search engines only when it is complete.
  7. A page for one disease in one country needs at least 3 open trials. Thinner pages are not published.
  8. Every machine-made text shows the “Machine summary” label, a link to this page and links to its sources (EU AI Act, Article 50).

What we do not do

  • We do not score or rank people for trials, and we never say a trial is right for someone.
  • We give no diagnosis and no treatment advice. Please talk to your doctor.
  • We do not store health data. Your choices stay in your browser and in the page address.
  • We use no tracking cookies. We count page views per page type and day only, without IP address or browser details.
  • Money never changes the order of trials on patient pages.

When something is wrong

Every page has a “Report an error” link. A person checks each report. Reports about health content hide the page from search engines at once, and it is fixed or removed within 72 hours. Every real fix is listed in the corrections log.

Who checks

The editors check reports and failed checks. Every patient text is based on published research and shows its sources and dates; it says where research stands today. Machine-made summaries stay labelled as machine-made.