Infographics with AI

What an infographic is (and why it works)

An infographic is a visual representation of information or data — a chart, a flowchart, a timeline, a comparison. Done well, it lets someone grasp a relationship at a glance that would take a paragraph to state and several readings to hold.

One claim to retire while we are here. You will often read that the brain processes images tens of thousands of times faster than text, usually with a very specific-sounding number attached. That figure has no identifiable source and is not something anyone measured. Visuals are genuinely better than prose for showing relationships — bigger, sooner, more, grouped-with — and that is a real and sufficient reason to make one. You do not need the invented statistic, and repeating it in a piece about honest data presentation is an awkward way to start.

First question: is this actually an infographic?

Most things published under the label are a list of bullet points with an icon next to each one. That is a decorated list. It is not necessarily bad — a nicely set list is fine — but it does none of the work an infographic does, and it will not repay the effort of building it as an image.

The test is whether the layout carries meaning. If you could read it out loud as a list and lose nothing, it did not need to be a picture. An infographic earns its format when position, size or direction is doing something words would struggle with: a shape over time, a proportion, a flow with branches, two things compared on several axes at once.

Work this out before you generate anything, because it determines whether you are making a chart, a diagram or a poster — three different jobs that get confused constantly.

Choosing the form that fits the comparison

The most common craft failure is picking the visual first and forcing the data into it. The form should follow from what you are comparing.

• Comparing quantities across categories — a bar chart. It is boring and it is correct, and length is the thing people judge most accurately.
• Change over time — a line. Time runs along the horizontal axis because that is the convention every reader has.
• Parts of a whole — a bar or a stacked bar. Pie charts work for two or three slices and become unreadable beyond that, because angle is hard to judge.
• A process with decisions — a flowchart, and keep the branches to a page.
• A sequence of events — a timeline, with proportional spacing. Evenly spaced years that were not evenly spaced is a quiet lie.
• One striking number — just set it large with a sentence. Not everything needs a chart, and a big honest number often beats a diagram of it.

Avoid the decorative forms that circulate: 3D bars, donuts with a dozen segments, area charts stacked so that no individual series can be read. They come from template libraries because they look designed, and every one of them makes the comparison harder.

What AI does well here, and the one thing to never let it do

Genuinely useful: distilling a long document down to the claim it supports, proposing a structure, drafting concise headings and captions, suggesting which comparison a dataset actually supports, and producing the illustrative and layout elements around the data.

The hazard is specific and serious: do not let a generator produce the chart itself. Ask for one and you may get a finished-looking graphic with plausible values that came from nowhere — not a placeholder that announces itself, but a polished chart with numbers on it. An infographic is a format built to be screenshotted and shared, so it travels without your caveat, without the article, and without you.

Build charts from your own data in a tool that is plotting real values — a spreadsheet, a charting library, whatever holds the numbers — and bring the result in as an image. Where the design tool draws the chart for you, type the values in yourself and check each one against the source.

The ways a chart misleads without anyone lying

Most misleading infographics are made by people with no intention to mislead. These are the four that do the damage.

• A truncated axis. Starting a bar chart's value axis above zero turns a two per cent difference into a visual doubling. Bars encode quantity by length, so their baseline must be zero. Line charts showing change over time are the accepted exception, and even then say so.
• Percentages without the base. "Up 200%" from three to nine is a different story from three thousand to nine thousand. If the base is small, print it.
• A chosen window. Picking the start date where the line looks best is the most common form of accidental dishonesty. Show the longest range you have, or say why you cut it.
• Area used for a single number. Making one circle "twice as big" by doubling its width quadruples its area, and readers judge by area. Scale by area or use a bar.

A useful self-check before publishing: would the person whose data this is recognise their own result in your picture? If you are shading that answer, fix the chart rather than the caption.

Put the source on the image

This is the single highest-value habit in the whole format and it takes thirty seconds.

An infographic is designed to be lifted. It gets screenshotted, reposted, dropped into someone's deck and shared onward — detached from your article, your byline and your citations. Whatever is not on the image does not travel with it.

So put on the image itself: the data source, the period it covers, and your own name or domain. The first two make it verifiable, which is what separates a chart from a poster. The third is the only attribution that survives the journey, and it is why infographics can bring people back to you at all.

Data storytelling

A good infographic tells one clear story. AI helps with:

• Distilling the message: "what is the one thing this data says?"
• A narrative structure: opening, key points, conclusion — an order that leads the eye.
• Choosing what to include: what to keep, what to drop. Less is clearer.
• Concise text: short, sharp headings and captions.

Be careful with the first one. Ask a model what the data says and it will tell you something — confidently, whether or not the data supports it. Read its answer as a hypothesis to check against the numbers, not as a finding. It is very good at proposing a story and has no ability to know whether that story is true.

Design & readability

• Visual hierarchy: the most important thing is the biggest. Tell the eye where to start.
• Colour with meaning: colour groups and emphasises; it is not decoration. Keep the palette small.
• Breathing room: do not fill every pixel. White space is what makes the rest readable.
• Reading direction: lay out along the direction your audience reads. For a right-to-left audience that means the entry point sits on the right, and it is not something a template will do for you.
• Label directly where you can. A label next to its line beats a legend the reader has to keep referring back to.

The accessibility problem built into the format

An infographic is an image, which means that by default its entire content is invisible to anyone using a screen reader, and unreadable to anyone who cannot make out the colours or the type size. That is a large audience to design out of your best-performing content.

Four things fix most of it:

• Write real alt text — the finding, not "infographic about sales". If the chart shows that returns tripled after a policy change, the alt text says that.
• Put the key numbers in the surrounding article too. The page carries the substance; the image carries the impact. This helps search engines as well, which read text and not pictures.
• Never use colour as the only signal. Add a label, a pattern or a shape, and check the thing in greyscale.
• Check contrast and type size at the size it will be seen — which is a phone, not your monitor. Grey-on-white captions at ten pixels are decoration, not information.

The workflow

1. Message: decide the one claim before touching design. Write it as a sentence.
2. Data and structure: confirm the numbers against the source, then have AI propose a structure — comparison, timeline, process.
3. Charts from real data, built where the data lives.
4. Design around them in a layout tool or with AI design for the surrounding elements.
5. Check: every number against the source, readability at phone size, source line present, alt text written.

Size, crop and where it will actually be seen

An infographic is usually designed on a wide monitor and consumed on a phone held in one hand. That mismatch kills more of these than bad data does.

The traditional tall infographic — the one that scrolls for several screens — was built for a desktop web page. In a social feed it gets shown as a cropped thumbnail, and what people see is a slice of the middle with no context. If the piece is going to social, design for the feed: a single square or portrait image with one comparison on it, at a size where the smallest label is still legible on a phone.

Practical rules that follow:

• Check it at thumbnail size before publishing. Shrink it on screen until it is the size of a feed preview. If the point is not visible, it will not stop anyone scrolling.
• Put the headline claim at the top left (or top right for a right-to-left audience) so it survives a centre crop.
• Split a long one into a series. Several single-idea images outperform one scroll-and-pinch monster, and each can be shared on its own.
• Export at twice the display size so it stays sharp on high-density screens, and keep the file small enough not to slow the page it lives on.

Before you design: check the numbers survive scrutiny

Ten minutes here prevents the failure that matters, because a wrong figure in a shareable image is the one mistake you cannot quietly edit later.

1. Go to the original source, not the article that cited it. Second-hand numbers acquire errors, and the chain is usually shorter than you fear.
2. Check what was actually measured. "Users" and "accounts" and "sessions" are different things that get reported interchangeably.
3. Note the period and the population. A figure from one country in one quarter is not a general fact, and presenting it as one is how honest people mislead.
4. Redo one calculation yourself. If you are showing a percentage change, compute it from the raw figures. This catches both your arithmetic and the source's.
5. Ask what would make this wrong — a definition change, a survey method, a small sample. If you cannot answer, you do not know the number well enough to draw it.

Use cases, honestly rated

• Social media — the strongest fit. A single clear chart travels well and is easy to reshare.
• Inside a blog post — good, as the visual anchor for a point the text makes. The old advice that infographics reliably attract backlinks comes from an era when far fewer people made them; treat links as a bonus rather than the plan.
• Presentations and reports — see presentations with AI, remembering that a dense infographic is unreadable on a projected slide.
• Repurposing — turning an article or a study into a visual, and the reverse — repurposing.

Getting useful structure out of a model

Asking an assistant to "make an infographic about X" produces a description of a generic poster. A few changes turn it into a genuinely useful collaborator on the part it is good at — structure and wording — while keeping it away from the part it is bad at.

• Give it the data, not the topic. Paste the actual figures with their labels and units. Now it is reasoning about your numbers instead of recalling what is usually said about your subject.
• Ask what the data supports, not what it says. "Which of these comparisons is strong enough to build a graphic on, and which would be overstating it?" A model will happily find a story in noise if you ask for a story.
• Ask for three structures, not one, and make it say what each emphasises and what each hides. Every chart choice foregrounds something at another thing's expense, and making that explicit is a better use of the tool than accepting its first idea.
• Ask it to argue against the headline. "What would a sceptical reader say about this chart?" tends to surface the missing base rate, the cherry-picked window or the confounder before a commenter does.
• Then ask for the words: a headline of six words, labels under three, one caption of a sentence. Tight constraints produce usable copy; open requests produce paragraphs that will not fit.

What you are doing is using it for editorial judgement about framing and language, with the arithmetic and the drawing kept elsewhere. That division holds for most of this category, and it is the difference between a tool that speeds you up and one that quietly introduces errors into a picture thousands of people will screenshot.

When not to make one

• You have three numbers. Write the sentence. A chart of three numbers is a slower way to read three numbers.
• The data is weak. Visual polish makes a shaky finding look authoritative, which is the wrong direction to move it.
• The information changes often. Anything baked into an image is expensive to correct and impossible to correct in the copies people have already shared.
• You have not decided the claim. An infographic without one argument becomes a poster of assorted facts, which is what most of them are.

Common mistakes

• Information overload. One infographic, one message. Three messages means three images.
• Wrong data. Verify every number against the source. A beautiful chart with a wrong figure does real damage, and it spreads faster than the correction.
• Letting a generator draw the chart. Plausible invented values in a finished-looking graphic, travelling without you.
• Design over clarity. Attractive is good; unclear is a failure whatever it looks like.
• Misleading scales. Truncated axes and area-scaled circles. Visual honesty is part of the craft, not an afterthought.
• No source on the image. Once it leaves your page, an unsourced chart is an unverifiable assertion.