IA Prompting: Reframe and Relate

Good AI needs good IA. Information architecture is the old school skill designers and developers need to reclaim. Start with two things: reframe and relate.

Job titles and design. It’s an interesting, never-ending quest for clarity.

I started my design career as a graphic designer. My boss didn’t even know what to call it, honestly. For him, it was: I need someone who makes graphic work. And I started without a degree, so I didn’t know either. Graphic designer. Whatever. I relabeled it later when cleaning up my resume to visual designer—because in essence I was focused on the visual side of things. Wrong icons, PowerPoints, booth designs, brochure work, DTP, stuff like that.

But I was getting more and more involved in interface design. So I started studying communication and multimedia design. At that point, it wasn’t necessarily about chasing a job title. I did have a subject called interaction design—one of the best subjects I had, taught by one of my greatest teachers, Ferry den Dopper. He really showed the deeper layers of designing software. And that’s pretty much where it started. From visual designer, I became an interaction designer.

At the time, that was the common way to look at UX design. UX design wasn’t even the label yet. Interaction design was common, and so was user researcher. Then UX design became a thing. So I relabeled again. I became a UX designer, and later a UX/UI designer—because most hiring managers had no clue. Companies had no clue. Our whole design community was just trying to figure it out. UI designers were moving into UX work. User researchers were suddenly responsible for visual design too. Interaction design became old school, and information architect became an even more dying trade. Eventually we became experience designers, design systems engineers or product designers.

The titles kept shifting.

And more and more, the focus moved to output: great-looking user experiences. I noticed—working in very complex environments—that UX researchers were still fighting in boardrooms for budget just to talk to users. But in the end, the systems took over.

What has happened now is that those systems are powered by AI to generate the screens as well. Once the tokens are in place, once the rules and guidelines are set, generation becomes easier. The output we drifted toward is no longer what we need humans for.

The designers doing well in this shift are the ones going back to the roots. Back to the essence of design.

Information architecture.

AI Needs Good IA

Everyone is prompting now. Developers, designers, business owners and product managers. You describe what you want and the system generates it. Fast and impressive.

But most people just throw raw information at AI and hope for the best. No structure, no curation, no defined relationships. Just input and whatever comes out.

Think about how your brain filters sound. Walking down the street, your ears pick up everything—traffic, a dog barking, voices around you. But your brain decides what to hear. Not because it has better microphones. Because it has context. It knows the car passing is a risk for the kid holding your hand. It knows the dog might be your neighbor’s and you want to say hi. Your brain knows what matters because it understands relationships.

That’s information architecture. Built into human intelligence by default.

AI doesn’t have that by default. It has to be given it. When you dump raw, unstructured information into a chat window, the model does its best—but its best is a statistical average. A plausible answer built on assumptions. Without structure, without relationships, without context, there’s no way to know what actually matters.

Great intelligence—artificial or human—runs on well-structured information.

How to Architect Information

Information architecture is a serious discipline. There are entire books, frameworks and careers built fully around it. I won’t be covering all of that here.

What I want to cover is where to start. Two things you can do today that will immediately improve the quality of your AI output, your designs and your systems.

The essence of design. Drawing boxes and drawing lines.

I call them reframe and relate.

Reframe

Every piece of information has a name. Most of the time, you don’t even realize it. You chose a label that felt logical or felt obvious, and it stuck. Now the whole team uses it, it’s in the system, it’s in the class naming, it’s everywhere. And now the AI is getting trained on it too.

Bad names become a bad foundation.

Reframing means stopping and asking: what is this, actually? What do we call it, and does that name carry the right meaning for everyone using it? This is what reframing is about—consistent vocabulary, shared definitions and metadata that makes information findable, searchable and more important than ever: machine-readable.

Take a simple example from an engineering dashboard. A field shows a value from a sensor on a piece of equipment. Someone called it “reading” when they built it. It made sense at the time. But across the system, the same concept gets called “value” in one place, “measurement” in another and “output” in a third. The AI getting trained on this data has no way of knowing these are the same thing. Neither does the new engineer onboarding. Neither does the search index.

Reframing means stopping and agreeing: this is a measurement. That’s what we call it, everywhere, always. You update the label, the metadata, the tags. Now the system can find it. Now the AI can reason about it correctly.

But a measurement on its own still means nothing.

Relate

Once the boxes are named and defined correctly, it’s time to draw the lines.

That measurement only makes sense in relation to the equipment it came from, the threshold it’s being compared against and the operational phase it belongs to. A vibration reading of 4.2 during drilling startup means something completely different than the same reading during steady-state operation. The number is the same. The meaning is not.

When you relate that measurement to its context, you give the AI what it needs to interpret correctly. Not guess. Interpret.

That’s the difference between structured information and raw data.

How does this connect to that? Why? In what order and with what priority? Those relationships turn a collection of content into a system that actually makes sense. This is where most people stop. The taxonomy is solid, the data is curated and they call it done. But without the logic that connects one thing to another, the structure is static. It doesn’t move when applied.

Information that relates is information that works—for users navigating a product, for a team maintaining a system and for an AI that needs to understand not just what things are but how they belong together.

Draw the boxes. Connect them. That’s your architecture.

Closure

Information architecture never went away. It got buried under the pressure of shipping more screens. We knew this once. Then we let it go. Now it matters again.

Now the screens are being generated. We finally have the time back to think, to provide structure and to build real foundations.

This is old school. Designers and developers were doing this long before AI made it a necessity. If you have that background, dust it off. This is your differentiator.

And if you’re new to it, start simple. Two questions before you prompt, before you build anything:

• What are the boxes? Reframe them.
• What are the lines? Relate them.

That’s your foundation. That’s how you make AI work better. And that’s how your work gets better too.