How to Make Journal Club Slides With AI

Journal club is the one presentation format where summarising the paper well is a failure.

Residents learn this the hard way. You are handed an article, you spend Sunday evening building fifteen careful slides walking through what the authors did, you present it, and the attending's first question is "so, did you believe it?" The slides had no answer, because summarising and appraising are different jobs and only one of them was done.

This matters for AI tools, because summarising is precisely what they are good at, and appraising is precisely what they are not.

The search results tell the story

Of the eight organic results Google returns for "journal club presentation ai," exactly one comes from a medical school. The rest are tools. That ratio is the whole problem in miniature: the supply of software that will generate journal club slides now vastly exceeds the supply of guidance on what those slides are supposed to contain.

What a journal club deck owes the room

A working structure, in the order the room needs it:

• The clinical question, in PICO form if your programme uses it. Population, intervention, comparison, outcome. One slide.
• Why this paper, now. Does it change anything? If the honest answer is no, say so early and the discussion gets better, not worse.
• Study design, stated plainly, with its known weaknesses named rather than glossed.
• The threats to validity. Randomisation and allocation concealment, blinding, loss to follow-up, whether the analysis was intention-to-treat, whether the primary outcome was the one they registered. This is the heart of the presentation.
• The results, with effect sizes and confidence intervals rather than p-values alone. Absolute risk reduction alongside relative, because relative risk on its own flatters almost every intervention.
• Applicability. Would this change what you do on Monday, for the patients you actually see?

Notice that only one of those six is a summary of the paper. Five are judgments about it.

Where AI helps, honestly

Extraction. Pulling the trial characteristics, sample sizes, outcome measures and confidence intervals out of a PDF and into a clean table is exactly the sort of accurate-but-boring work that machines do better than a tired resident at midnight. This is real time saved, and it is most of the time saved.

Structure. Getting a first-pass deck that already has the standard journal club skeleton means you start from a scaffold instead of a blank file. Tools built for the format, such as ChatSlide's AI journal club presentation maker, start from the paper itself rather than a generic slide template, which is the difference between a draft you edit and a draft you rewrite.

Reference handling. If your programme wants citations on the slides, having them pulled from the source rather than retyped removes a whole class of error.

Where it does not help at all

The appraisal. An AI reading a trial report will generally reproduce the authors' framing, because the authors' framing is the document. It will tell you the intervention reduced the primary outcome by the stated amount. It will not spontaneously tell you that the primary outcome was changed after the trial began, that the confidence interval brushes against no effect, or that the control arm received a treatment nobody has used since 2019. Those observations are the presentation.

Knowing your service. Applicability is local. A paper conducted in a tertiary referral centre with unlimited imaging access may be irrelevant to a community hospital in northern Ontario. No model knows your denominator.

The uncomfortable conclusion. The most valuable journal club presentations end with "this paper is weaker than its abstract suggests, and I would not change practice." AI drafts almost never land there, because the source document is written to persuade.

A workable method

1. Feed the paper in, and take the extracted trial characteristics table. Keep that.
2. Delete the generated summary slides. Most of them are restating the abstract.
3. Write the validity section yourself. This is the presentation and it cannot be outsourced.
4. Add absolute numbers wherever the paper gives you only relative ones.
5. End on your own verdict, stated plainly.

Broader academic tools that handle papers, posters and defence talks from the same source, like ChatSlide's research presentation toolset, are useful if journal club is one of several formats you present in. But the division of labour stays the same regardless of tool.

The verdict

AI has eliminated the tedious half of journal club preparation and cannot touch the half that the session actually exists for — which makes it a genuine improvement for anyone who understands the distinction, and a quiet degradation for anyone who does not. The residents who use it to skip the extraction and spend the reclaimed hour on critical appraisal will give better presentations than they used to. The residents who use it to skip the appraisal will give worse ones, and the attending will notice within about ninety seconds.