How to Turn a Research Paper into a Presentation That Doesn't Lose the Science

By the Kinsy team · Published 25 July 2026

The paper took a year. The talk is on Thursday. You have twenty pages of carefully hedged argument, four experiments and sixty references — and a fifteen-minute slot in front of people who will remember exactly one thing. Turning a research paper into a presentation is not compression; it is translation. Same science, rebuilt for a room. Here is how to do it without flattening the findings.

How many slides does a research paper become?

Twelve to fifteen, for a standard fifteen-to-twenty-minute conference or seminar slot — and that number barely moves whether the paper runs eight pages or forty. The arithmetic is unforgiving: about a minute per content slide, minus the opening, minus questions. What changes with a longer paper is not the slide count but the cut — a 20-page paper holds perhaps an hour of material, so most of it must go, deliberately. That is not dumbing down. The talk's job is not to replace the paper; it is to make the room want to read it, and to leave behind one finding they can repeat at dinner. The paper holds the detail. The talk sells the argument.

Why IMRaD doesn't survive contact with a room

A paper and a talk fail for opposite reasons, because they are read in opposite ways. IMRaD — introduction, methods, results, discussion — is a filing system for readers with random access: a reviewer jumps straight to the methods, a rival to table 3, a student to the discussion. Nobody reads a paper straight through, so the structure doesn't need to persuade in order; it needs to be findable.

An audience has none of those privileges. Listening is strictly linear, happens once, at your pace, with no re-reading — so a talk gets one pass at working memory and has to spend it on a single argument. That inverts the structure: the contribution moves from the discussion to minute two; the literature review shrinks to one sentence of context; the methods compress to a diagram; the robustness checks retreat to backup slides. The common failure — one slide per section, every subsection dutifully represented — produces a deck that is faithful to the paper and useless to the room.

The manual method: six steps from paper to slides

1. Write the one-sentence contribution. Before opening any slide software, finish this sentence: "We show that X does Y under Z." If it takes three sentences, you have three talks; pick one. Every slide you make from here either supports that sentence or gets cut — it is the cheapest editorial tool you own.
2. Storyboard 12–15 beats on paper. Problem and why it matters; the gap; your contribution, stated plainly by minute two; how you did it (one beat); the evidence (three to five beats); limitations; the so-what. Note that this is the discussion's logic, not the paper's page order — you are organising for persuasion, not for retrieval.
3. Make the figures the slides. One figure per slide, and let the slide title state the finding — "Treatment effect holds across all three cohorts" — so the figure proves the title rather than decorating it. Re-set axis labels and legends at 18 point or larger; type that works on an A4 page is invisible from the back row.
4. Compress methods to one slide. A diagram of the pipeline or design, not paragraphs. The audience needs to trust the method, not reproduce it live. Park the sample sizes, parameter tables and preprocessing detail on backup slides after the end — that is where Q&A lives.
5. Give limitations their own slide. One honest slide, placed before the conclusion, saying what the result does not cover. It reads as confidence, not weakness — and it answers the hostile question before anyone gets to ask it.
6. Move the hedging into speaker notes. The slide carries the claim; the notes carry the confidence interval, the exact n, the caveat about the second cohort. Precision you speak lands better than precision the audience has to squint at — and the science survives, one layer down.

Citations on slides, done properly

Slides need citations for the same reason papers do — claims you did not generate need sources the audience can check — but a bibliography pasted onto a slide is unreadable and reads as nerves. The working convention: a numbered superscript beside the specific claim it supports, a small sources line at the foot of that slide (first author, venue, year), and the full reference list on a backup slide rather than on screen. Numbered markers beat author–year on slides because they take three characters, not thirty. Two more rules: any figure you reproduced or adapted from someone else's paper gets its credit on the slide where it appears, not buried at the end; and your own paper's full citation belongs on the title and closing slides, because the talk's whole purpose is to get it read. If an AI tool is drafting the deck, the same standard applies to it — we wrote up what real, checkable citations look like in AI presentations with citations that actually hold up.

The AI route: upload the paper, keep the citations

The six steps above are a few evenings' work, and before a job-market talk they are evenings well spent. For the Tuesday lab meeting or the third retelling of the same paper to a new audience, there is a faster route: an AI deck generator that treats the paper as source material. We build Kinsy to do exactly this, so read what follows as the makers' description — and verify it on your own paper.

The flow: upload the paper PDF (uploads are free), write a one-line brief — "15-minute seminar for a methods-savvy audience" reads very differently from "10 minutes for a general-science session" — and answer a couple of clarifying questions. Kinsy plans the narrative first, from the document: which findings carry the argument, what order they should land in for a listening audience, and what belongs in speaker notes rather than on screen. A fact-check pass then verifies the slides' claims against your PDF as the source of truth — the guard against the number-drift that plagues chatbot-made decks — and where it researched beyond the paper, it adds numbered, dated citations so you can see exactly what came from where. Every slide gets speaker notes.

The export matters as much as the draft in academia, because a deck gets edited right up to the session: the output is a native, editable .pptx in which every text element opens as a real editable object — never a flattened picture — with the designed fonts embedded so the file survives the lectern machine, and the preview you approve matching the download to the line break. (Why that is rarer than it should be: AI presentation makers that export truly editable .pptx files.) Honest limits: five curated themes rather than journal-specific templates, decks only, and a build costs 40–90 credits — the free plan's 100 credits cover about one full deck, enough to test this section against your own paper. The step-by-step walkthrough is in document to deck with Kinsy.

The rehearsal checklist

1. Run it out loud, timed, standing. Silent read-throughs lie by about thirty per cent. Aim to land at ninety per cent of the slot — questions and the previous speaker's overrun will eat the rest.
2. Do the back-row test. Put the deck on a screen and stand three metres away. If any axis label, legend or superscript is illegible, it is invisible in the room — fix the type, not the projector.
3. Rehearse the limitations answer. You already know the two hardest questions. Say your answers out loud once, and check the backup slides that support them are actually in the file.
4. Say the contribution cold. If you cannot state the one-sentence finding without looking at a slide, neither will anyone else. It should open the talk and close it.
5. Open the .pptx on the machine you will present from. Fonts, notes view, clicker, the works — five minutes in the empty room beats any amount of hoping. If the file was exported from a web tool, this is the step that catches swapped fonts and flattened text.

The bottom line

A research paper and a conference talk are different genres that happen to share an author. The paper is complete, hedged and random-access; the talk is one argument, told once, in order. Getting from one to the other means choosing the twelve slides the room actually needs — contribution first, figures as evidence, limitations owned, citations numbered — and letting the paper keep everything else. Do that by hand when the stakes justify the evenings, or let a generator draft the structure and spend your time where judgement lives: deciding what the room must not be allowed to forget.

Keep reading

• How to turn a PDF into a presentation
• How to turn rough notes into a presentation
• The best AI deck makers in 2026, compared