The biggest brand in the category was the one machines couldn’t read

There is a check I run on franchise brands where I fetch their location pages the way a machine does. No browser, nothing rendered, just the file the server hands over. Then I read what is actually in there: does each location have a name, an address, a phone number, hours, some way to act. The stuff a person sees on the page without thinking about it, and that a machine only sees if somebody put it in the markup on purpose.

That is the first of three layers in what we run as the AI Storefront Audit, and it is the one that decides whether an assistant knows your locations exist at all. A location people can see and machines cannot is what I have been calling an Invisible Storefront.

The results do not sort by size.

I expected them to. When I started doing this I assumed the answer was going to be boring, that big brands would have it and small brands would not, because big brands have people whose job this is. That is not what comes back. I have pulled brands with several hundred locations that publish nothing at the location level at all. Every store invisible, the whole network, no markup anywhere.

And then one comes back inverted, which is the one worth writing about.

A brand with fewer than two dozen units. Every location page had it. Real per-location structured data, not brand-level boilerplate copied across. Street address. Phone in the right format. The parent brand named as the parent. And on every single page, an explicit link tying that location to its listing on the map platform, which is a thing I almost never see. Somebody built that on purpose and knew what they were doing.

Then I checked the category’s largest player, many times their size, the brand everybody in that category is measured against.

Nothing. Not thin, not partial. No location markup at all.

Why size stops helping

I sat with that for a minute, because it does not match how anything else in franchising works.

The big brand wins on media budget. It wins on real estate, because it gets the corner. It wins on franchise development, because candidates have heard of it. It wins on vendor pricing. In almost every field the advantage compounds and the small brand is playing defense.

The machine layer does not work like that, and the reason is unglamorous. It is a template decision.

You write the fields into the location page template once, and every location in the system gets it. The tenth and the four hundredth cost the same as the first, which is roughly nothing. So the budget that wins every other fight buys you no advantage here, and having twenty locations instead of four hundred costs you nothing either.

Which means this is one of the very few places where a small brand can simply be ahead. Not scrappier. Ahead.

The layer where you find out if it worked

Markup is an input. It is not evidence.

So the second layer is the one that actually answers the question, and it means going and asking. We put the same real customer questions to five assistants, ChatGPT, Perplexity, Gemini, Claude and Google’s AI Overviews, in three of the brand’s real markets, and read what comes back word for word.

That is where the surprises live. A brand can be perfectly legible and still never get named, and you would never learn that from your own site. It does not show up in traffic. It shows up as a customer who asked, got three names, and none of them were you.

The layer that decides who gets picked

Which brings me to the distinction I keep coming back to, so I will say it plainly.

Being readable makes you eligible. It does not make you chosen.

When an assistant answers “who does this near me” or “where should I go for this,” it has to first know you exist as a business at an address, which is the markup, and then decide you are worth naming, which is reviews, photos, and what other people have written about you somewhere that is not your own website. Legibility gets you into the pool. Selection decides who comes out of it.

The brand I just described had the first one and almost nothing on the third, on its own site anyway.

So “ahead on the machine layer” is a real advantage and it is not the same sentence as “winning.”

The part that bothered me more

The data was there and one field was wrong everywhere.

The name on each location was the town it sits in. Not the business name, not the business name with the town after it. Just the town.

Which means a machine reading that network sees a set of businesses named after towns, connected to the brand only by an inherited parent field. Every automated check scores that green, because the field is populated and the file parses. Presence and correctness are two different tests, and almost everybody only runs the first one.

One template field. That is the whole fix. It was also the single most valuable line in the audit, which tells you something about where the leverage actually sits.

It’s a window, not a moat

Here is the part I would want to hear if I were the small brand.

The advantage is real right now and it has an expiration date, because it costs the giant exactly as little to close as it cost you to open. One template change on their side and the gap is gone in an afternoon. They are not choosing not to do it. Nobody over there has looked.

So the move is to finish the layer while nobody is looking, and then go win the part that does not close in an afternoon, which is the selection side. Reviews and third-party presence take years and cannot be templated. That is where a lead becomes durable.

And if you run the big brand in your category: somebody smaller than you is already the readable one, and you will not find that out from any dashboard you currently look at.

Go look at yours

The cheap version is free and you can do it today. Pull one of your own location pages without a browser and read what comes back. Ten minutes, and you will either be fine or you will be glad you looked.

If you want the whole picture rather than one page, that is what the audit is. Two business days, $2,500. We score all three layers zero to one hundred, run five assistants across three of your real markets, and hand back the verbatim answers with screenshots so you can see exactly what a customer sees, plus a fix list written as tickets your web vendor can actually execute. Then thirty minutes on a call to walk you through it.

Either way, go find out what the machines say about you. It is a strange thing to have never checked.