Learning With AI Images: Leave Room for Your Own First Sketch

An AI assistant turns a difficult paragraph into an elegant illustration. You understand the picture, save it and move on. Later, when you try to explain the idea without looking, the explanation is harder to produce than the image was to recognise.

One useful question is what part of the learning activity you delegated. Asking a system to draw something and drawing a representation yourself can produce similar-looking pages while involving different work by the learner. The finished image does not tell the whole story.

Research on the drawing effect offers a reason to examine that distinction. It does not prove that generating images harms learning, or that everyone should replace AI with a pencil. It does suggest that making a representation deserves attention as an activity, alongside the quality of the representation you receive.

What the drawing experiments found

In a 2016 paper, Jeffrey Wammes, Melissa Meade and Myra Fernandes reported seven free-recall experiments comparing drawing with other ways of studying information. In the initial experiments, participants drew or wrote presented words; drawn words were recalled better. Other experiments tested alternative explanations and different study conditions.

The authors proposed that drawing integrates meaning, visual information and motor activity. That is an explanatory proposal, rather than a direct measurement of every process involved. Crucially, participants created drawings: the work does not establish that viewing an AI-generated picture produces the same benefit.

For a reader using AI, the practical inference is limited but useful. If you delegate the drawing, you have changed the activity. Judge that new workflow on its own merits instead of borrowing evidence from experiments in which people made the drawings themselves.

Task-relevant drawing is different from decorative doodling

A 2019 study by Meade and colleagues compared drawing the material with writing and doodling unrelated to it. Across its experiments, task-relevant drawing supported recall, while free-form doodling produced poorer recall than drawing or writing. Shading prepared shapes did not improve recall over writing.

The implication is to keep the sketch connected to what you are trying to learn. Filling a notebook with attractive patterns is a different exercise from deciding how to represent a concept. Neither the presence of a pencil nor the beauty of the page identifies which activity took place.

These are established memory studies, not experiments on contemporary AI image tools. They do not give a universal ranking of every visual learning method, and they do not establish that an elaborate diagram always beats a clear written explanation.

Sketch one relationship before requesting a polished image

Consider a hypothetical business owner learning the difference between a customer placing an order and the business fulfilling it. An AI-generated illustration might show a shop, a parcel and a smiling customer. It could be attractive without making the distinction the owner needs to explain.

A first sketch could instead contain an order slip, a separate packing table and a line between them. The useful question is what that line means. Does it represent a request, a confirmed handoff or completed delivery? Deciding that forces the owner to specify the relationship they are representing.

That is an editorial example of how to use sketching, not a tested learning intervention. The drawing need not be publication material. Keep it small enough to finish and specific enough that you can say what each element stands for.

Start with the source material and select one idea. Sketch your interpretation, then add a short written explanation. If the idea resists a picture, record that difficulty rather than forcing an elaborate metaphor. Some information may be clearer as a sequence, a comparison or a sentence.

Give AI a role after your first interpretation

You can then ask an assistant to question the representation: “Which relationship in this sketch is ambiguous? What might a reader infer that the source does not support? Ask me to explain before suggesting a replacement.” This is a proposed prompt, not a claim that an assistant can reliably grade understanding.

Compare its observations with the original source. A fluent criticism can be mistaken, just as a fluent explanation can. If the assistant proposes an extra step or a causal arrow, check whether that connection is actually established before adding it.

If you want a polished visual afterwards, keep the learning sketch and the presentation illustration as separate versions. The first records your interpretation; the second serves an audience. You may approve different amounts of detail for those two purposes.

This also gives an independent creator a practical record of decisions. A collaborator can see which elements communicate the essential idea and which are decorative choices. Do not let visual refinement silently change the meaning you originally intended.

Check what remains when the picture is hidden

After working through the idea, put the drawing and source aside and try explaining it in words. Then return to the source to identify omissions and mistakes. MaryChuks.com’s guide to retrieval practice and rereading discusses the value of examining what you can produce without the material in view.

For the hypothetical order example, ask whether you can explain what confirms the order, what initiates fulfilment and what evidence shows completion. Being able to remember the parcel in the picture does not settle those distinctions.

Treat this as a check on your current explanation, not a diagnosis of your memory or proof that sketching caused an improvement. If you compare different methods informally, differences in difficulty, familiarity and time spent can affect the result. A useful personal routine is not automatically a controlled experiment.

Choose an activity you can actually use

Drawing should remain an option. If making marks is uncomfortable, inaccessible or simply unsuitable for the task, use another way to express the relationship and check it against the source. Do not turn a research finding into a requirement that excludes somebody from learning.

Likewise, there are sensible reasons to request a finished AI illustration immediately: preparing a presentation, exploring visual treatments or communicating an idea you already understand. The question is whether your immediate goal is to make an asset or practise representing unfamiliar material.

When learning is the goal, reserve a little room for your own first representation before outsourcing the visual work. Then use AI, sources and a later explanation to examine it. A polished picture can be useful; your own decisions about what belongs in the picture are worth preserving too.

Sources

Primary studies: Wammes, Meade and Fernandes, The drawing effect (2016); Meade, Wammes and Fernandes, Comparing doodling, drawing and writing at encoding (2019). The suggested AI workflow and business example are editorial applications, not reported results from those experiments.