AI Can Find You. Will It Choose You?

Visibility is only the beginning. The more useful question is what happens when AI has to decide whether you belong on the shortlist.

I've spent the past year working inside a problem that most companies are only beginning to name.

I built my own AI startup without a large marketing machine behind it. I worked with companies trying to understand how they appeared inside AI systems. Then I started doing the same work with people: founders, experts, professionals whose public identity doesn't live neatly on one corporate domain.

I spent a lot of time looking at actual answers.

Not dashboards. Answers.

Who showed up. Who didn't. Who appeared when I asked by name but disappeared when I described the need instead. Which sources AI relied on. What it got right. What it confidently got wrong. What happened after something in the public information changed.

Eventually, the question I was asking changed.

It was no longer:

Can AI find you?

It became:

What would make AI choose you?

That sounds like a small distinction. I don't think it is.

A whole new category is being built around AI visibility right now. Companies want to know how they appear in ChatGPT, Gemini, Claude, Perplexity and the rest. Marketing teams want to see where competitors are winning. Agencies want to know which sources influence an answer and what content they should create next.

That market is real. Some of the products being built for it are excellent.

I also think the opportunity is much bigger than the marketing department.

Because AI doesn't only recommend brands.

It recommends companies to hire, products to buy, founders to know, experts to call, people to work with.

And the need to understand those decisions doesn't suddenly begin when a company has a large marketing budget.

A founder running a small team has the same basic question.

So does a professional whose expertise is the thing being sold.

So does someone who has deliberately built a public name around their work.

These aren't the same entities, and treating them as if they are creates problems of its own.

A founder isn't her company. A company isn't one of its customers. A professional isn't automatically a "personal brand." A person's current work isn't necessarily the thing they were best known for three years ago.

AI has to work all of that out before the recommendation even begins.

And I've seen how easily it doesn't.

A system can read every page on a website and still misunderstand what the company actually is.

It can know a company perfectly well when you ask about it by name, then leave it out when a buyer describes the exact problem that company exists to solve.

It can find years of information about a person and assemble those perfectly readable pieces into the wrong identity.

Readable isn't legible. And legible doesn't automatically mean chosen.

That's why I've become less interested in visibility as an endpoint.

I want to know what happens after visibility.

When someone describes a real need, are you considered?

Who is considered instead?

What does AI seem to believe about you?

What evidence is it relying on?

Where is that evidence missing, contradictory or simply pointing to the wrong conclusion?

And most importantly: what, if anything, should you actually change?

Because this is where the current conversation about "optimizing for AI" starts to bother me.

There's already a familiar checklist.

Publish more content. Add FAQs. Get mentioned on Reddit. Add schema. Create comparison pages. Get more citations.

Maybe.

But if AI has confused your company with your customer, another blog post isn't necessarily the answer.

If it understands what you do but doesn't have enough independent evidence to trust a particular claim, rewriting your own About page may do nothing.

If your founder, company and product are being understood as three disconnected things, publishing more of the same information can simply create more of the same problem.

The fix depends on the failure.

That sounds obvious. In practice, it changes the product completely.

I don't think most people want another dashboard full of prompts, scores and charts they have to learn how to interpret.

They want to know:

Where am I being chosen?

Where am I losing?

Why?

What should I fix first?

Can you help me fix it?

Did it work?

That's what I'm building Legibility to answer.

Underneath those questions, the system I'm building is deliberately rigorous. It separates what was actually observed from what we infer. It preserves the original AI responses and the evidence behind them. It treats different models, languages and markets as different observations rather than blending everything into one reassuring score.

And when there isn't enough evidence to recommend a change, the answer should be exactly that.

No action yet.

But none of that complexity should become the customer's job.

That part matters to me a lot.

The enterprise version of this problem can justify an enormous amount of infrastructure. Large marketing organizations will want deep analytics, workflows, integrations and teams working across them.

But the underlying question exists long before any of that.

A founder shouldn't need a marketing department to understand why AI keeps choosing three competitors.

A professional shouldn't need to become an expert in AI search to understand how their work is being interpreted.

And a company that does have a sophisticated marketing team should still be able to go deeper, inspect the evidence and understand exactly what sits underneath a recommendation.

Same problem. Different depth.

This is why I think about Legibility less as a visibility tool and more as a layer for machine-mediated choice.

Visibility tells me I appeared.

Legibility should tell me whether the system understood enough about me to put me in the right decision, what happened when that decision was made, and what I can reasonably do about it.

Then we measure again.

Not because a second measurement magically proves causality. AI systems are variable, and pretending otherwise would make the product prettier but less useful.

We measure again because without looking at what happened after an intervention, "optimization" is mostly advice.

I'm much more interested in the loop.

Understand. Measure. Diagnose. Fix. Measure again.

And eventually, keep watching.

Because public reality changes. Companies change positioning. Founders move. New evidence appears. Old information survives much longer than anyone expects. Different languages tell different versions of the same company. AI systems change too.

Legibility has to live inside that movement.

There's also something slightly ridiculous about the name I chose.

Legibility is an ordinary English word.

So while I'm building a product designed to help companies and people become more legible to AI systems, I also have to make Legibility legible.

The product has to become associated with the problem it solves.

My name has to be correctly connected to the product I'm building.

The public evidence has to make sense.

And when an AI system gets any of that wrong, I want to see where the understanding broke.

So I'm not standing outside this problem explaining it.

I'm building inside it.

And I think that's where the next part of AI marketing gets much more interesting.

The question is no longer only whether AI can find you. It's what happens when AI has to decide whether you belong on the shortlist at all.