Feeding the Machine
This is all a confidence model. AI is not ranking businesses based on one factor. It is weighing a collection of signals.
This isn't something I put together for a conference. It's what years of working in SEO and information architecture have distilled into, and I'm still iterating on it. The underlying question hasn't changed in all that time: how information gets organized decides what gets found and believed. AI didn't change that question. It changed who's doing the finding.
There isn't one thing to fix. There are five. I call the whole exercisefeeding the machine, because that's what you're actually doing. You're not petitioning anyone for a ranking. You're handing a system the raw material it needs to reach a confident answer about your business, in the same plain language it needs to reach that answer about anyone else's.
Before any of it, one precondition. If the assistants' crawlers can't reach your pages, none of the five signals matter. Not "you rank lower." Absent, the same as a business that doesn't exist. Check that gate first. It's broken more often than people expect and it takes minutes to find.
Start with the questions. The machine doesn't decide once that you're credible. It decides again every time someone asks something. Being the biggest name in your category won't carry you into every answer about it. So pick the questions you want to own. Then work the five signals for each one.
Below are the five. Each has the question the machine is really asking, and the line I use to make it stick. Read them together to see the shape of it. Each one links out to a longer piece on that question. That's where the real work happens.
A note on scope: I built this for businesses that serve a place, retailers and service providers with locations and delivery radiuses, because that is where I work and where it has been tested. Proximity in particular matters far less for a national ecommerce brand than for a store that has to get a couch to a house.
Relevance
Does this business clearly match the customer's need?
If you do not have this data you aren’t a consideration.
Relevance sounds simple until you watch it up close. A customer whose washer died on a Sunday isn't asking who the best retailer in town is. She's asking who has one in stock, who can deliver it this week, and what it will cost her. The machine is trying to match her exact problem to your exact data. Not your brand story. Your data.
Here's the part that catches people off guard. If you don't have that data on the record somewhere the machine can read it, you don't lose the match. You never enter it. Not "ranked lower." Not "second choice." Absent from the shortlist entirely, the same way a store that doesn't exist yet is absent.
So the fix isn't clever copy. It's inventory: what you rent, what it costs, how fast you deliver, who qualifies, stated plainly and consistently everywhere a machine looks.
Read the full post: The Lost Lead Never Looks Like a Lead →Proximity
Proximity is not just where the store is. It is whether the store looks like a practical local option.
Everyone assumes proximity is simple: closest store wins. It doesn't work that way anymore, and treating it as a distance problem alone will cost you customers you're actually equipped to serve.
A location twelve miles away that delivers same day, serves the right zip codes, and reads like a real local business can beat a location four miles away that looks generic, outdated, or unclear about its service area. The machine isn't measuring miles on a map. It's judging whether you read as a workable choice for someone standing exactly where the customer stands.
This is also the single biggest opening a smaller or regional business has. The best-known name wins the popularity contest. It does not automatically win the practical-local-option contest, and that gap is where a well-documented local business takes the customer instead.
Read the full post: Can a Small Business Beat a Bigger Competitor in AI Search? →Trust
Does the market prove this business is real and reliable?
Trust is what the market proves about you.
Notice the question isn't "does this business claim to be reliable." Every business claims that. The machine has learned, correctly, to discount a claim a business makes about itself. It's looking for proof from somewhere else: reviews, listings, mentions, the accumulated record other people have left about you.
Not your homepage copy. Not your mission statement. The stranger who wrote a review last month, the directory that lists your hours correctly, the customer who told a friend it worked the way you said it would. That's the evidence a cautious system reaches for, because it's the evidence a cautious person reaches for too.
One distinction worth making, because it gets muddled constantly. When an assistant answers a local question, who's near me, who delivers here, who's open now, it leans on local business data. In that world reviews are load-bearing. Count, rating, recency, and whether you bother replying all feed the prominence those surfaces rank on. I'm not hedging that one.
What's genuinely unsettled is a narrower and different question: whether review platforms themselves get cited as sources in general web answers. Often they don't, and sometimes for mundane reasons like blocking a crawler. So work your reviews hard for local, where the evidence is strong. Just don't let anyone use one question to sell you certainty about the other.
The obvious next question is whether stacking up more reviews is the trick to winning this signal. The honest answer is more complicated than a marketer wants to hear.
Read the full post: Do Reviews Affect Whether AI Recommends You? →Clarity
Can AI understand the offer?
Every buyer, human or machine, works through five questions before anything else matters: what can I get, what does it cost, how fast can I have it, what do I need to qualify, and what happens after I say yes. If your pages don't answer those five plainly, the machine can't hand you to the customer even if it wanted to. It isn't withholding the recommendation. It genuinely can't extract an offer that isn't there.
This is where most businesses waste their budget, because the instinct is to make the page look more official instead of more legible. I've watched businesses spend real money expecting polish to move this signal, and the evidence doesn't back that bet. What moves it is answering the five questions in plain sentences, the way you'd answer a customer standing in front of you.
Read the full post: Does Schema Markup Help You Get Cited by AI? →Authority
Can AI verify the recommendation elsewhere?
It’s not just what you say that’s important. It’s what everyone else says about you.
This is the signal that catches the most confident owners off guard, because they've spent years building a great story on their own website and assumed that was the whole job.
When a machine is deciding whether to name you, it doesn't stop at your homepage. It goes looking for corroboration: press mentions, industry listings, other pages that describe you the same way you describe yourself. If your own site is the only place your story exists, you don't have a verified record. You have an unconfirmed claim, and machines are built to discount those.
This is the same lesson I learned writing the industry's own book: the businesses that get defined fairly are the ones who built a record before someone else wrote it for them.
Read the full post: Accuracy Is a Form of Advocacy →The best-documented business wins by default.
That's the whole game in one line. Whoever gets named doesn't have to beat you on service, price, or care. They just have to be the name the machine can actually verify when your side of the record is thin. Feed the machine your five signals, and that default stops being automatic.
Read the blog