AI for Conference SlidesWhat to Delegate, What to Guard Yourself

AI can propose a slide structure, compress your text, translate it, and polish the phrasing. In other words, it's good at building the "surface" of a slide deck.

But the substance on that surface — your data, your analyses, your conclusions — is something AI cannot vouch for. Speed up the surface; guard the substance. Every technique in this article rests on that principle.

Give AI your presentation time and a summary of the study, and ask: "Draft a slide outline for a 7-minute oral presentation, ordered background → objective → methods → results → discussion → conclusion." In minutes you'll have a scaffold, slide counts included.

A scaffold makes every decision faster than facing a blank page. But whether to adopt it is your call. If the structure buries what makes your study worth presenting, rearrange it without hesitation.

When a slide's text runs long, requests like "turn this into terse bullet points" or "cut this to ten words" are where AI is unbeatable.

It's also good at register shifts for presentations — unpacking jargon for a general audience, or tightening prose into compact academic phrasing. Just make sure the meaning didn't drift during compression — above all, that "suggests" didn't quietly become "demonstrates." The strength of a conclusion is the researcher's responsibility, not the editor's.

If you're presenting at an international meeting and need slides in English, AI translation of your draft is genuinely serviceable.

Verification is still required. Do the technical terms match the standard usage in your field? Did the units survive translation intact? Prioritize checking terminology and numbers over how natural the prose sounds — that's the trick.

The numbers, graphs, and tables on your results slides must come directly from your own analysis output. Asking AI to "turn these results into a graph" and pasting in what it generates is dangerous — axes and values can shift subtly.

Export figures from your own analysis software, and check them against the source data before they go on a slide. A wrong number in a conference talk gets almost no chance of correction. This is the one place where saving time is off the table.

Here's a check that's easy to miss but important: some conferences and academic societies have begun requiring disclosure of — or placing limits on — generative AI use in presentation materials.

Read the submission guidelines and presenter instructions, and when in doubt, ask the conference office or your supervisor. The right posture is "check the policy, then use AI openly." It beats using it quietly and regretting it later.

The numbers and conclusions of prior studies cited on your background slides should be verified in the original papers — not recalled from an AI summary. That's the same principle that governs every encounter with the literature.

With Paperfy, each candidate citation keeps its PDF, AI summary, and figures on one page, and a summary you've reworked for your talk becomes the first draft of your background slide. Edit the radio script to follow your presentation flow, and you can rehearse by ear during your commute. Build the surface fast with AI; keep the evidence solid in your library.