Slides That Support a Talk Instead of Replacing It

The test is simple. If someone could get your talk by reading the slides, they will — and they will stop listening to you.

Reading is faster than speech. An audience presented with a dense slide finishes it in fifteen seconds and then waits, or checks their email. Everything you say during those seconds is lost. For a concrete operational comparison, Monitask explains cognitive offloading in practical terms.

Conventions vary. Some fields expect dense slides that double as a handout; some expect almost none. What follows assumes a talk where the slides accompany you.

What a slide is for

To show something you cannot say. A figure, an equation, a photograph, a table of results, the specification.

To anchor where you are. A heading that tells someone rejoining what section this is.

To hold the number you want remembered.

Not to list what you are about to say. That is a script displayed to the audience, and it competes with you.

The practical rules

One idea per slide. If a slide has two points, it is two slides. Slides are free.

Six lines is a lot. Most slides that fail have twelve.

Sentences, not fragments, where you use text at all — but few of them. A single well-chosen sentence on an otherwise empty slide is read, considered, and remembered. Eight bullet fragments are skimmed.

The heading should state the finding, not the topic. "Effect concentrated in the first quartile" tells the audience what to look for. "Results" does not.

Font size larger than feels necessary. The person at the back has your slide at a fraction of the size you are seeing. If you are unsure, it is too small.

Check contrast. Light grey on white disappears on a projector, whatever it looks like on your screen.

Figures

The part that carries most of the value, and where most slides fail.

One figure per slide, at the size of the slide.

Label the axes, in words, at a readable size. Axis labels shrunk from a paper figure are unreadable at the back of a room.

Remove everything that is not doing work. Gridlines, legends with six entries, tick marks at every unit, the third and fourth series.

Highlight the comparison you are making. If the point is that two lines diverge, everything else on the plot should be muted so those two lines are what the eye finds.

Say what the figure shows before you show it, then show it, then say what to look at. "This plots outcome against time, treated in red — and here the two separate" gives the audience a route through it.

A figure taken straight from the paper usually does not work. Paper figures are designed to be studied; slide figures to be understood in seconds. Rebuilding them is worth the time.

Tables

Most tables do not belong on a slide. A regression table with six columns is unreadable at distance and unnecessary — the audience cannot check your standard errors from the third row of a lecture theatre.

Show the two numbers that matter. Large. The rest is in the paper.

If a table is genuinely needed, cut it to the rows and columns you will discuss, and highlight the cell you are talking about.

Equations

Only where the audience needs to see the structure.

Build them up if they are complex — show the pieces in sequence rather than the finished object.

Say what each term is, aloud, pointing.

And do not show a derivation. Nobody follows algebra on a screen at speaking pace. State the result and put the derivation in the appendix.

Structure across the deck

A title slide with the paper title, your name and affiliation. It is on screen while you are introduced, which is when people are working out who you are.

No table of contents slide for a twenty-minute talk. Say the structure in one sentence instead.

Section markers if the talk has three or more parts. A slide with two words on it costs nothing and orients everyone.

A final slide that stays up during questions. Your main finding, or your contact details, or both. Not "Thank you", which occupies the screen for the most useful ten minutes of the session.

Appendix slides after the end, for anticipated questions. Being able to jump to a robustness check when asked is a strong signal of preparation. Number them so you can find them.

What to cut

When the deck is too long, in this order:

Motivation. Whoever is in the room is in your field.

Literature. One slide at most, and frequently none — cite in passing while speaking.

Method detail beyond what makes the result interpretable.

Secondary results. They belong in the appendix slides.

Anything you would skip if you were running short. If it would be skipped, cut it now and give the remaining slides more time.

Practical failure modes

Animations that require clicking through eleven builds. They break when the file opens in different software, and they consume time.

Presenting from your own laptop without checking the connection. Arrive early.

A file that only renders correctly in your software. Export to PDF as a fallback, always.

Fonts that are not on the conference machine, which silently substitute and destroy your layout.

Reading the slide aloud. The most common failure and the most damaging.

The test again

Give the deck to a colleague without commentary. If they can reconstruct your argument from it, the slides are doing your job.

That sounds like a compliment and it is the problem. A deck that stands alone is a paper in landscape format, and the talk it accompanies is redundant.

See the conference talk for what the talk itself should carry.

For an independent authoritative reference, consult the MIT Communication Lab for academic communication guidance.