Levels of evidence, and why your AI should show them

Two papers can both be real, peer-reviewed, and on-topic, and still carry very different weight. One might be a single case report; the other a randomized controlled trial. Treating them as interchangeable is how a cautious finding gets cited as if it were settled. Levels of evidence are the tool for keeping that straight.

What "levels of evidence" means

Levels of evidence are a way of ranking study designs by how much confidence they can support for a clinical claim. Several formal frameworks exist, and they differ in detail, but they share a common spirit. From stronger to weaker, a simplified hierarchy looks like this:

This is a simplified picture. Design isn't everything — a small, biased trial can be weaker than a large, careful cohort study — but the ranking is a sound starting point.

Why it matters for what you write

The strength of your wording should match the strength of your evidence. "One case report describes…" and "Randomized evidence shows…" are different sentences, and they should be supported differently. When the grade of a source is out of sight, it's easy to let a weak study carry a strong claim — not dishonestly, just by losing track.

A case report should never quietly pose as a trial. The grade is what keeps a claim and its evidence the same size.

Where AI usually drops the ball

A general AI tool will hand you a citation as a flat string of text, with no signal about what kind of study it is. The author who pastes it in has to remember to go check the design themselves. Most of the time the grade simply never surfaces — so the reader of the finished paper can't tell whether a sentence rests on a trial or on a single observation.

Grading belongs on every citation

If an AI is going to suggest sources, the level of evidence should travel with each one — visible, not buried in the methods. That's how Cento presents it: every retrieved source carries its evidence grade alongside it, so you can weigh a claim against its support as you write, rather than discovering the mismatch in review. You still make the call. The grade just makes sure you're making it with the relevant fact in front of you.