Entity Clarity: How AI Decides Who You Are

What "entity resolution" actually means

When you ask an AI engine who the best roofer in your city is, it doesn't start by ranking businesses. It starts by figuring out which businesses exist and which facts belong to which one.

That's entity resolution. The engine gathers everything it can find with your name attached — your website, your Google Business Profile, your Yelp and BBB listings, directory entries, third-party mentions — and tries to assemble one coherent picture of a single business. Where does it operate? What does it do? Is this the same company as that one, or two different companies with similar names?

If that picture comes together cleanly, you're a candidate for the answer. If it doesn't, you're a risk. And an engine that isn't sure who you are will quietly leave you out rather than say something wrong about you.

This is Signal 2 of the 8 Signals, worth 15 of 100 points. It's not the biggest category, but it's the one that caps the others: all the proof in the world from Signal 1 doesn't help if the engine can't confidently attach it to you.

The boring problem: your name, address, and phone

In the reviews I've run, the single most common entity problem isn't exotic. It's that the basic facts don't match.

A company has a Google Business Profile, but the business address isn't anywhere on the website. Or the address on the website isn't the one on the profile. Or the phone number on the contact page is the old one, and the tracking number on the ads landing page is a third. Each of those is defensible on its own. Together they tell an engine that the facts about this business are unreliable.

Marketers call this NAP consistency — name, address, phone — and it's been a local SEO fundamental for years. What's changed is the consequence. Google was tolerant of small inconsistencies because it had other ranking signals to fall back on. An AI engine assembling a single recommendation has far less room to be wrong, so it weights confidence heavily.

Why conflicting numbers cost you more than you'd think

This is the part that surprises people most.

It sounds like nitpicking. It isn't. An engine cross-checking you has no way to tell the difference between a stale number and a wrong one. It just registers a contradiction, and contradictions lower confidence in the source — which means the other facts on that page carry less weight too.

The usual suspects, in the order I find them:

• Review counts and star ratings quoted on a homepage that were true two years ago
• Years in business — "over 20 years" on one page, "since 2001" on another, and it's now a different number than when someone wrote it
• Job or client counts — "6,000 roofs" in one place, "8,000+ homeowners" in another
• Service area lists that don't match between the website, the profile, and the directories
• Team size or credentials that were updated in one place and not the others

The fix is unglamorous: pick the accurate version of every number, then make every page and profile agree. Where a number will keep changing, either commit to updating it or write it in a way that stays true.

Say what you do in the words people use

The second common failure is self-inflicted. A business names its services something distinctive — a branded process, a proprietary-sounding package — and then that name is the only label on the page.

A human visitor figures it out from context. An engine trying to match "who does roof replacement in Cincinnati" against a page selling the "Total Home Envelope Solution" has nothing to match on.

You don't have to give up the brand name. You have to make sure the plain-language version appears too, in the heading, the first sentence, and the page title. Name it plainly first, brand it second. This is also where a named framework earns its keep in the other direction — a proprietary name that's defined on a page becomes an asset rather than a barrier, which is its own signal.

Name your service area out loud

"Serving the greater metro area" tells an engine nothing. It can't resolve that into geography, and geography is most of what a local recommendation is about.

List the cities, towns, and counties you actually serve, in plain text, on a page a crawler can reach. Say it the way a customer would say it. If someone in a suburb twenty minutes out would describe themselves as being in that suburb rather than the metro, use the suburb's name.

This overlaps with local SEO, and if that's where you're starting, our local SEO work covers the same ground in more depth. The AI-specific difference is that an engine is looking to state a fact about where you work, not to rank you within a radius.

Make the machine version agree with the human version

Structured data is where you state your identity in a form a machine doesn't have to interpret — your legal name, address, phone, service area, and the services you offer, marked up as LocalBusiness or Organization data in the page's code.

The rule that matters more than the markup itself: it has to match what a human sees on the page. Schema saying one thing while the visible content says another is a contradiction in the place you least want one, since it's the exact source an engine leans on for identity. The mechanics of getting that markup in place are covered in Signal 6.

When you're more than one location

Multi-location businesses have a version of this problem that's harder to see.

I audited a health spa franchise with four locations, and the owner was convinced his homepage messaging was the problem. When I pulled the numbers, 8% of the site's traffic came through the homepage. The other 92% arrived at the individual location pages first — which meant that for most visitors, and for most engines, a location page was the business.

He wanted to rewrite the homepage. What actually needed fixing was four location pages, each of which had to independently establish who that location was, where it was, and what it offered.

If you operate in more than one place, the test is simple: could an engine landing on any single location page, with no other context, tell exactly which business and which location it's looking at? If the answer depends on the visitor having come through the homepage first, you have an entity problem.

The fifteen-minute entity audit

Run this yourself before hiring anyone, including us.

• Search your business name and open the first five results that aren't your own website. Note the name, address, and phone on each.
• Compare those against your website's contact page and footer. Any mismatch is a finding.
• Search your site for every number you publish — ratings, review counts, years, job counts — and check they agree with each other and with reality.
• Read your services page as if you'd never heard of your company. Would a stranger know what you sell?
• Find the sentence that names your service area. If there isn't one, that's the gap.
• Ask ChatGPT or Perplexity to describe your business, then check every fact it returns. What it gets wrong tells you where the confusion lives.

That last one is the fastest diagnostic in this entire framework. The engine will happily tell you what it thinks you are, and the errors are almost never random — they trace back to a specific stale profile or contradictory page.

What I don't know yet

I don't know how the engines weigh a contradiction between two of your own pages versus a contradiction between your page and a third-party listing you don't control. I'd expect the latter to matter less, but I haven't been able to test it cleanly.

I also don't know how quickly a correction propagates. When we fix a NAP inconsistency, I can't say whether an engine picks it up in days or in months, because the changes usually ship alongside other work and I can't isolate them.

What has held across every review: the businesses that get described accurately are the ones whose facts agree. I'll update this page as the timing questions get clearer.

Frequently asked questions

Related reading

• The 8 Signals — the full framework, all eight scored categories, and what your total means.
• The Proof AI Is Looking For — Signal 1, the 20-pointer. Entity clarity is what lets an engine attach that proof to you.
• llms.txt, Schema & the Files AI Looks For — Signal 6, and the mechanics of getting your identity into structured data.
• How AI Fact-Checks You Off-Site — Signal 8, and the listings you don't control.
• What Is a Total Online Presence? — why your website, your Google Business Profile, and AI search function as one system.

This is Signal 2 of 8. See the full framework →