AI and Magazine Design: How Worried Should We Be?

Of course, it wouldn’t happen like this; the robot needs an artist to copy.

The recent release of Claude Design sent what might best be described as a silent shockwave through the design community. Inside, designers are mostly terrified by the prospect of an AI that specializes in hastening their professional redundancy. Outside, most seem to be whistling past the technological graveyard, as it were, leery of giving too much attention to yet another potential job-killer.

The advent of Claude Design and its like-minded peers was inevitable. This is a fight that writers, photographers and illustrators have been fighting — and losing — for years. As a designer, it would be churlish of me to expect big tech to somehow overlook our corner of the industry. It’s just our turn.

But as a magazine specialist, I have my reasons for believing that the threat, while very real, isn’t imminent.

As just about everyone knows by now, generative AI results are really only as good as their instructional prompts. “Make a lively and interesting magazine layout,” is pretty vague, as instructions go. What makes a layout interesting? Sidebars? Subheads? Infographics?

And yes, you could amend your instruction to “Make a lively and interesting magazine layout with two sidebars and a chart showing blah blah blah…” But can the AI find those opportunities within source material that doesn’t include them? How much more work will it take to prep content for a robot designer?

Prompting to Excess

Here’s a real world example of the challenge I’m describing. This is a layout I created some years ago for a real estate magazine. I don’t recall the original headline, but I’m pretty certain that I came up “States of Confusion” while brainstorming visual ideas. Why? Because the image and the headline really don’t make sense without one another.

Just to test my theory, I asked a generative AI to rearrange the state outlines as I did here, maintaining the general outline of the continental United States in the process. It couldn’t do it. It couldn’t even understand the idea. It swapped the names of the states around, it perpetuated generative AI’s famous track record for creative spelling, but it could not produce an incorrect map.

Or consider these two spreads from the same magazine. Let’s say you spotted the AI the concept, telling it you wanted to do something about “bad apples.” Could — or would — the AI do the added research to find the four most likely categories of review states for this profession (the “No Action Required,” “Educational Letter…” checklist in the opening artwork)? Would it find or manufacture supporting artwork that built upon the “bad apple” theme without repeating the same iteration of it? Could the AI have selected the pullquote and artwork that supported each other’s message (page 22)? Would it have broken the grid on page 23 to call attention to the sidebar or would it have mindlessly dropped it into the outside column?

With instruction, I’m sure the AI could do all this and more. But think about the volume of prompting these results would require. Very quickly, one gets to the point where it’s literally more work to explain the problem to the robot than it is to solve it oneself.

The Designer’s Mind

In episode 5 of Quarto Creative and LTD Creative’s shared podcast, “Covering the Spread,” magazine expert Bo Sacks rhetorically asked if an AI could be trained on George Lois’ Esquire layouts to produce new work as good as Lois’. Could it be trained? Could it render a new article in the style of George Lois? Well, yes, but so what? What is “style?” Here are two of Lois’ most famous covers.

If you instructed a machine to manufacture these images, I have no doubt that it could succeed. But that would constitute a gross misunderstanding of the designer’s role.

The problem, as with microstock and Canva and all the other low-rent shortcuts is that, invariably, for some people, the results of generative AI design are good enough. Can an AI take a completed manuscript with a few provided images and produce an attractive layout? I’m sure we’re already there.

But, as I’ve often said, that is truly the lowest bar for achievement that we can set for ourselves as magazine publishers. Of course, the layouts should be attractive. That’s the least of it. Ideally, the layouts should reveal aspects of the topic that would be impossible to glean from reading the text or viewing the images alone. The layout should continuously provoke the reader with new information, new types of information, and new styles of organizing that information. A well-designed magazine article is a teacher. A pretty magazine article can only be decorative.

What do you think? Do you see generative AI as an immediate threat to magazine designers? Send me a note at scott@quartocreative.com.