A PubMed scan that respects your time.
Free setup prompt. Pastes into Claude, interviews you about your specialty, then builds a daily digest that runs while you sleep — deduped against the papers you've already read, filtered through a quality rubric you control. Set up in about ten minutes.
It runs your queries — you pick six to twelve, narrow ones like carotid stenosis or AI in surgery, not vascular surgery alone — and dedups against the papers you've already filed. The rest goes through a quality rubric you write and edit. You see three to eight papers a day, not eighty. Case reports under twenty patients get dropped. "We asked ChatGPT" papers get dropped. The signal-to-noise ratio is the whole point.
The interview asks what else you check every morning. Common picks for clinicians: institutional news, society announcements (SVS, ACS, your board), conference deadlines, deliverables from your own task tracker, FDA approvals, market data. You get three to six sections — pick what's actually useful, skip the rest.
Not an email. Not a chat thread. A single file at a fixed path that overwrites daily. Open it in your browser, AirDrop or sync to your phone, read it over coffee.
Go to claude.ai/download. Free to install. To actually run the briefing on a schedule you need the Claude Pro tier ($17/month) — that's what unlocks Claude Code, which is the part that does the work. The briefing's daily token use sits comfortably inside Pro's monthly allowance, so there's no extra cost on top.
Open the Claude desktop app and click either the Cowork tab or the Code tab. Both run the same engine — pick whichever surface feels more comfortable. The prompt works identically in either.
It tells Claude to interview you about your specialty, your reading habits, and your daily routine, then build the briefing system from your answers.
It will ask about six things, in plain English. Your specialty. Which journals you trust. What study designs you weight up. What other sections you want. What time it should run. Roughly ten minutes of conversation.
Claude summarizes the plan before writing any code. Say "go." It writes the scripts, sets up the scheduled task, and runs it once so you can see the first briefing. If anything looks wrong, tell it — it fixes and re-runs.
After day one, it runs on its own.
Copy this whole thing into a new Claude session. Or download it as a markdown file.
If you have the $17/month Claude Pro subscription, the briefing's daily tokens sit inside your existing Pro allowance — no extra cost on top. If instead you're running through the API with token credits, it costs roughly $4/month. Retrieval is done by a Python script that doesn't use any LLM tokens; the model only gets paid to triage and summarize.
The first day is the noisiest because the dedup list starts empty. Every paper the briefing shows you (and every paper it considered for the digest) gets auto-appended to your "seen" list at the end of the run — so by day two those won't reappear. The list builds itself just by your reading the briefing daily; you don't need to file or curate anything.
Yes — that's the design. At the end of every run, the dedup list is automatically updated with the PMIDs and DOIs from today's candidates. That's how tomorrow's digest stays fresh. The list is just a flat record of IDs you've already been shown; it doesn't store summaries, tags, or notes.
The PubMed queries run from your machine to the NCBI API (same as any literature search). The triage and summarization run through Claude. Your dedup state, your rubric, your knowledge base, and the HTML output all stay local on your computer. Nothing about the briefing itself is sent anywhere else. (StatUp the site uses standard Google Analytics like most websites — page views only.)
Yes — that's the point. The rubric lives in a plain markdown file the orchestration re-reads every morning. If a paper survived that shouldn't have, you sharpen the rubric and the next run does better.
This prompt was written for Claude Code, but I suspect their systems can handle it as well. It may take some tweaking. Please try it out and let me know!
The architecture works for any "I need a curated morning digest" use case — the PubMed section becomes whatever your domain feed is (arXiv, SSRN, a Substack list, GitHub releases). The interview questions are tilted toward clinicians, but you can redirect them.
I'm Marissa Famularo, DO — a vascular surgeon and health data science trainee. I built this for my own morning routine first, then realized the pattern transfers. If you set it up and want to compare notes, message me on LinkedIn.