The strength of clinical evidence is recoverable from language model representations but not from their stated grades
For clinicians and patients relying on LLM-summarized evidence, the models' stated confidence grades are unreliable even though the underlying signal is recoverable, highlighting a critical gap in LLM transparency.
The study shows that clinical LLMs encode evidence strength in their internal representations (median AUROC 71.8) but fail to express it when asked (stated grades at chance, 25-27 percentage points below estimator). The signal is lexical, topic-specific, and distinct from truth, yet still flags weakly supported claims (AUROC 69.2).
Large language models (LLMs) increasingly summarize clinical evidence, where a claim's weight depends on how strongly it is supported. Yet these models convey confidence poorly, and properties they never state, such as truth, are often readable from their activations. Whether a clinical model registers evidence strength, distinct from truth, and states it when asked is untested, and any such signal could be lexical. We compiled 45,134 clinical claims from six public sources, harmonized 20,611 into a four-level evidence grade under three independent frameworks, and tested 22 local, open-weight LLMs from several developers (0.6-70 billion parameters; general, medical, and reasoning), with lexical, truth, and cross-framework controls. A linear estimator recovered the grade in every model (median AUROC 71.8), yet decodability did not rise with scale and was weakest in reasoning models. The grade the models stated fell to chance, 25-27 percentage points below the estimator. The recoverable signal was largely lexical and did not transfer across topics or frameworks, yet it was distinct from factual truth and still flagged weakly supported claims (AUROC 69.2). Clinical LLMs thus carry an ordered evidence-strength signal they do not express, so their stated grades fail to convey a claim's support even when it is recoverable from their representations and text.