How we write

How we pick research, how AI writes each summary, and the automatic safety checks it must pass before it publishes.

The short version

AI writes a plain-language summary of each study. Before it goes on the site, the summary must pass a set of automatic safety checks. If it passes, it publishes on its own. If a check finds a problem, the item is held instead of published. Every summary you see was written by AI and passed these checks.

Where the research comes from

Each day we check public sources: PubMed for new studies, the National Cancer Institute for news, and ClinicalTrials.gov for trials. We only use what is public and free to link.

How we sort it

Not all research helps a patient today. We sort each item into one of these:

• Studies in people. These reach the front page.
• Early research in animals or lab cells. These are hidden by default. You can turn them on, and they carry a clear label.
• Not about brain tumors. These are dropped.

A study that has not been checked by other scientists yet is called a preprint. Preprints never reach the front page, no matter what they say.

How the summary gets written

The AI reads only the title and summary of the study. It may not add facts from anywhere else. If the source does not say something, the AI leaves it out.

The AI is told to write short sentences at about an 8th grade reading level, in plain words, the way a person would. It is banned from words like breakthrough, miracle, and cure.

The safety checks every summary must pass

These checks run on every summary. They are what make it safe to publish a summary automatically:

• Every number must come from the source. A script checks that each number in the summary also appears in the study. This catches an invented statistic.
• No hype words. Words like breakthrough, miracle, and cure are blocked.
• Reading level. The summary must read at about an 8th grade level or simpler.
• A "what this doesn't mean" section is required. Every summary has to say plainly what the research does not prove.

If a summary trips any check, it is held instead of published.

How close is this to helping patients?

Each summary also carries a readiness score from 1 to 10. A 10 means it is approved and a doctor can offer it today. A 1 means it is an early lab idea. The score can never be higher than the stage of the research allows, so a mouse study can never look close to the clinic.

What we label

Every item shows how far along the research is, with a mark for each step:

• Tested in people: the strongest kind of evidence we share.
• Review of existing research: a look across many studies.
• Early research (animals): it worked in mice, not people.
• Early research (lab cells): it worked in a dish.
• News or New or updated trial: not a finding yet.
• Preprint: not yet checked by other scientists.

When we get it wrong

We will. Every summary has a "report a problem" link. Flagging one holds it for another look, and when we change something, we say what changed and when.

What to do next

Look up a word in the glossary, or read about this site.