How AudioScholar is made

AudioScholar turns the week’s peer-reviewed medical literature into short audio briefings: a ~10-minute podcast for every specialty, and tools for papers you choose yourself. The scripts are written by AI, and we say so plainly, in every episode and on this page. This page explains the checks between the source papers and what you hear.

Who runs it

AudioScholar is built and run by Dr. Nikhil Shah, a practicing academic nephrologist in Edmonton, Canada, who is also one of the service’s heaviest listeners. There is no content team and no sponsor: one physician, an engine he supervises, and the checks described below.

Where the papers come from

Every Specialty Weekly starts from PubMed, the U.S. National Library of Medicine’s index of the biomedical literature, filtered through a hand-curated catalogue of about 285 journals selected per specialty. Only papers added to PubMed since that specialty’s previous episode are eligible, and a paper without a usable abstract is excluded outright. A summary written from a title alone would be guesswork, so it is not allowed to happen. Every Specialty Weekly and Topic Update is written from each paper’s PubMed abstract. The full text is not read, so use an episode to choose which papers to read in full. Every specialty’s journal list, and each step of the pipeline, is published here.

What the AI does, and what checks it

A model selects the week’s most significant papers and writes the narration, and the episode publishes without a human listening first. These checks make the output trustworthy:

  • Citation verification. Every paper cited in a script is verified against the week’s actual PubMed results. A script whose citations cannot be verified is regenerated, and if it still fails, the episode is not published. We would rather skip a week than misattribute a paper.

  • Numbers from the source only. Scripts are constrained to the figures present in each paper’s abstract, and a negative or non-significant result must be reported as exactly that. Because every episode publishes its full transcript with linked references, any claim can be checked against its source in one click.

  • Quiet weeks stay short. When a week’s literature is thin, the episode says so and stays short. Nothing is padded to fill ten minutes, and a week with nothing worth your time tells you that in under two.

  • The AI notice is spoken. Every episode ends by stating that it is an automated summary that can make mistakes, and that the original sources are the reference. This is said in the audio, not only in small print.

When something is wrong

AI systems make mistakes, and some will reach publication despite the checks. We make two commitments. First, reported errors are investigated against the stored record of how that episode was generated. Second, every confirmed defect becomes a new permanent, automated check. Each safeguard above exists because a physician caught a real problem in a real episode, and the fix was built so that class of error cannot recur silently. If you hear something wrong, please say so: nikhil.shah@audioscholar.cc.

Listen or read

Every episode has its own page with the full transcript and the reference list, each paper linked to PubMed. If audio is not your medium, or not your colleague’s, you can read the same briefing. Papers you bring yourself (a PDF, a link or pasted text) can be turned into audio in any of 31 languages. A PDF or pasted text is summarised from that document. A link uses the paper’s open-access full text when there is one, and its abstract when there is not.

What AudioScholar is not

AudioScholar is a current-awareness briefing. It tells you what was published and why it might matter, so you can choose which papers to read. It is not medical advice and it is not a systematic review. It is for published literature only: uploading patient-identifiable information is prohibited, every upload asks you to confirm there is none, and summaries of your uploads stay private to you unless you share them.

Cost

The weekly specialty shows carry no advertising and require no account: pick a specialty and follow it in your podcast app.