Start here

You have ten minutes and a journal article. Here's how that's done.

Reading an academic paper front to back is the wrong move and also not what anyone in the field actually does. There is an order, it is not the order the paper is printed in, and it takes about ten minutes once you know it.

Pick your team

Each team has a different study. You'll be walked through yours.

The method

Five stops, in this order.

Not abstract-then-intro-then-conclusion. That order tells you what the authors want you to think. This order tells you whether they earned it.

Why not just read it?

Because you are not going to, and pretending otherwise is how people end up summarizing an abstract and calling it a literature review. Ten focused minutes on the five things that determine whether a finding is believable beats forty distracted minutes on all of it.

You have the green light to use AI to help you summarize. It is very good at telling you what a paper says. It is much worse at telling you whether the design supports it, which is stop two, and which is the stop that actually matters.

Distill

Your cheat sheet.

This is what you wrote, assembled. If a line is thin, go back and fix it — you'll be presenting from this.

The test

Read your five lines out loud to someone on your team who has not read the paper. If they can then answer "how do they know that?" — you're done. If they can only repeat the finding back, your design line is doing no work and the summary will not survive a question.

Design

Now one slide. You are going to prompt an AI to build it.

One slide, for classmates who have not read your paper, who get about forty seconds.

Read this before you scroll

Nothing on this page makes your infographic. There is no template here and no generator. What follows is a set of guiding principles — the judgment calls that separate a slide that lands from one that doesn't.

You make the infographic by prompting an AI yourself. Claude, ChatGPT, Gemini, whatever you use. Your job is to carry the judgment below into that prompt, and then to audit what comes back against it.

This is the whole skill. An AI will cheerfully hand you a slide with eight colours, a pie chart and a truncated axis, and it will look professional. Knowing why that slide is wrong, and saying so in advance, is the part you are being graded on — and the part the AI cannot do for you, because it does not know which number is your number or what your study actually showed.

Decision 1 — is it even a chart?

This is where most study slides go wrong. If your finding is one number, the honest form is that number, large, with a label. Not a bar chart with one bar. Not a pie with two slices. A one-bar bar chart is a number wearing a costume, and it makes the reader work to extract something you could have just told them.

Don't

One bar. The axis, the gridline and the category label are all doing nothing. The reader still has to look up the value.

Do

0.39

percentage points more likely to vote
out of 61 million people

The number is the chart. The label carries the unit and the comparison, which is what the bar chart was hiding.

Decision 2 — if it is a chart, the job picks the type

Not your taste. The job.

One headline number

A large figure with a label. No axes.

Comparing sizes

Bars, one colour, sorted. More is darker only if the categories have a real order.

One thing is the point, the rest is context

Emphasis — colour the one, grey the others. The most underused form and usually the right answer.

Something over time

A line. If the question is "how long until it's gone," put the zero line in and label where it crosses.

Above or below a baseline

Diverging bars from zero — two opposed hues, neutral grey at the middle.

More than about seven categories

A table. Not more colours.

Never two y-axes. Two scales on one plot invent a relationship that isn't in the data. Two measures means two charts.

Decision 3 — make the size honest

You are summarising an effect. The single easiest way to mislead with a real number is the axis.

Don't

Axis starts at 59.8. A 0.39-point difference becomes a tripling. Every number on this chart is true.

Do

Zero baseline. Now you can see the real story: the effect is real, and it is tiny. That is the finding.

Rules that follow from this: bars start at zero, always. Put the unit on the chart, not in your head — percentage points and percent are different things. And if you truncate a line chart's axis, say so on the chart.

What to cut

Forty seconds buys you about four things. Pick them:

The claim

One sentence, in words, at the top. Not the paper's title.

The number

With its unit. Large.

One picture

Either your own, or the paper's — redrawn, not screenshotted.

The catch

One line. This is the line that makes people trust the other three.

Everything else goes: the literature review, the sample-size table, the authors' institutional affiliations, the second finding, your logo, and any sentence beginning "this study seeks to."

If two of you disagree about what to cut, you have found the interesting thing about the paper. Put that on the slide instead.

Writing the prompt

A prompt that produces a usable slide carries six things. A prompt that produces a generic one is missing most of them. Before you type anything, have an answer for each:

1The claim, in your words. One sentence. Not the paper's title, not its abstract. If you paste the abstract, you get a slide about an abstract.

2The number and its unit. Percentage points or percent. Standard deviations or raw scores. Say which. An AI will not catch the difference and will happily label it wrong.

3The form you have decided on, and why. Don't ask it to choose. "A single large number, no chart, because the finding is one estimate" is an instruction. "Make an infographic" is a wish.

4The audience and the time. Graduate classmates who have not read the paper, forty seconds, projected.

5The constraints you want enforced. Zero baseline. One accent colour. No pie charts. No dual axes. Unit on the chart. State them — they are not defaults.

6The catch, and that it must appear. One line. Models drop caveats unless told to keep them, because caveats make a slide look weaker and the model is optimising for looking good.

Then audit the output. Ask yourself: does the axis start at zero? Is the unit visible? Is there more than one colour doing work? Did the caveat survive? Is there a chart where a number would do? If any answer is wrong, say so specifically and ask again. Two rounds of that is normal.

If you find yourself accepting the first thing it gives you, that is the tell that you skipped the judgment and let the tool decide. Your slide will look like everyone else's, and in the debrief it will be obvious which teams did this.

What you should have in hand

Your cheat sheet below has the raw material. Copy it, then write your own prompt around it — do not paste the sheet in and ask for a slide.

Paste into your slide notes.

When the slide is done

Export it as a single PNG or JPG and put it on the wall — all fifteen slides land in one place, and the ten minutes we spend looking across them is the actual lecture.

PowerPoint: File → Export → PNG, current slide only. Google Slides: File → Download → PNG.