Discovery Infrastructure: The AI Strategy That Was Right for Ten Years Before Anyone Called It That

TL;DR: When a buyer asks ChatGPT, Gemini or Google's AI Overviews for a recommendation, the answer is assembled from content those systems can find, read and trust. Businesses that own deep, structured content on their own domain get named. Businesses that rent their visibility mostly do not. We have been building that kind of content system for dealerships since long before AI search existed, and it turned out to be exactly what AI search rewards. It deserves a name: Discovery Infrastructure.

For most of the last twenty years, getting found online meant ranking. You wanted to be one of the ten blue links on the first page, and an entire industry grew up around moving you a few positions higher.

That is no longer the whole game. A growing share of buyers now ask a question and get an answer. Not ten links. An answer, usually naming two or three businesses, written in plain language by an assistant that read the web on the buyer's behalf.

If you are one of the names, you are in the conversation before the buyer has visited a single website. If you are not, you are not in second place. You are simply absent.

What an answer engine actually needs

Strip away the hype and an AI assistant needs four things from a business before it will confidently name it.

• Content it can find. Pages that exist, are crawlable, and stay put.
• Content it can read. Plain, specific language that answers real questions, not brochure copy and not a wall of inventory widgets.
• Content it can trust. Depth on a subject, kept current, from a source that has been saying consistent things for a long time.
• Content it can attribute. A clear entity behind it: this business, this person, this domain.

None of that can be bought as an ad. You cannot purchase a citation. The only way to earn one is to have built the material in advance, on a domain you control, about the things buyers actually ask.

We built this before it had a name

I started WebStager in 2005. We were building dealer websites by 2008 and working with auto dealers in earnest by around 2010. OneBigBroadcast came out of that work as a way to describe what we were really doing: publishing a business's knowledge from one place and broadcasting it everywhere buyers were looking. The OneBigBroadcast newsroom itself has archived posts going back to July 2010.

At the time, the industry standard was a dealer website with an inventory feed, a staff page, and maybe one blog that published whenever someone had a spare afternoon. We made a different set of decisions, and over the next few issues of this publication I am going to walk through them one at a time. Four of them matter most.

1. Many engines, not one blog. Every site we built could run multiple independent publications, each with its own address, archive and subject, instead of one blog competing for one publishing slot.

2. The content lives on the client's own domain. Not on a vendor's platform, not on a listing site, not in a social feed that resets every day. On the business's own address, where every piece adds to an asset they keep.

3. The salesperson is a publisher. Each person on the floor can have a profile, their own picks, their own content and their own reviews, all on the store's domain, so buyers can choose a person before they ever walk in.

4. Inventory is content. Every unit gets a crawlable page that earns its own entry in search, instead of disappearing into an application that search engines struggle to read.

Every one of those decisions was unfashionable when we made it. Every one of them maps almost exactly onto what generative search now rewards: depth, ownership, attribution and specificity.

The receipts

A thesis like this is cheap without evidence, so here is some, with the sources stated.

• Growth. One franchised dealership went from roughly $39 million a year in sales to close to $100 million in under four years while we handled their digital. We did not sell a single car. But the system we built was a large part of how buyers found them.
• Discovery you do not pay for. One dealer content library we measure pulled more than 100,000 non-branded search impressions in a ninety day window. Non-branded matters: those are people who were not searching for the dealership by name and found it anyway. (Source: Search Console, one dealership client, 90 days.)
• Inventory as a search asset. On one dealer site across sixteen months, the inventory tool alone produced about 14% of organic clicks. (Source: Search Console, one dealership client.)
• What the lead report misses. In one month on one store, 79% of sold units had tracked online activity before the sale, and only 14% produced a form. (Source: Signal, one dealership client, sold units matched against site activity.)

That last one is the reason most businesses underestimate all of this. If your reporting only counts form fills, most of the work your website does is invisible to you. Buyers research, compare, decide, and then walk in or phone. And when they arrive from an AI assistant, the visit often carries no referrer at all, so standard analytics files it as direct traffic. The channel is working and the dashboard says nothing happened.

So why give it a name?

Because categories get named by whoever explains them first, and right now this one is still forming. Vendors are starting to sell an "AI discovery layer" as a bolt-on. I think that framing gets it backwards. Discovery is not a layer you add on top of a website. It is the structure underneath it.

Here is the definition I am working from:

Discovery Infrastructure is the set of owned content systems, structured data and distribution that make a business findable, readable and citable by people, search engines and AI assistants, measured against what actually sells.

Every word in that sentence is doing work. Owned rules out rented reach. Systems rules out one-off campaigns. Citable is the AI part. Measured against what actually sells is the part almost everyone skips.

What it is not

It is not an ad product. Paid placement can rent you attention for as long as you keep paying, and that has its place, but it builds nothing that persists.

It is not a plugin. You cannot install authority.

And it is not "AI-written content." Machines can help produce material faster, and we use them for exactly that, but volume without substance is noise, and answer engines are getting better at ignoring noise every month. The architecture is the point. The writing has to be worth citing.

How to tell whether you have it

Five questions. Answer them honestly.

1. If your business stopped all paid advertising tomorrow, would buyers still find you next month?
2. Does new writing appear on your own domain at least weekly, about the specific things your buyers ask?
3. Could a buyer find, judge and choose a specific person on your team before contacting you?
4. Does every product or unit you sell have its own readable page that search engines index?
5. Can you connect what happens on your website to what actually sold, including the sales that never involved a form?

If most of your answers are no, you are renting your visibility. That was survivable when search meant ten blue links. It is much less survivable when the answer only has room for three names.

If you want a quick read on where you stand, our free Digital Scope scores your business across these areas in a few minutes, with your result on screen the moment you finish.

What comes next

This publication, Built Early, is where I will make the argument properly, one decision at a time. The next issues cover each of the four decisions above in detail. Alongside it, The Build Log will document what we are shipping as we ship it, and Field Notes will report what the data shows on live dealer sites, with client identities withheld and every figure sourced.

We were right early. Now it is time to be clear about why.