Chapter 5 — Be Someone It Can Name

A mid-market software company I’ll leave unnamed spent eighteen months repositioning. New category, new messaging, new website. Beautiful work.

Eight months after launch, they asked an AI assistant what their company did. It described the business they had been three years earlier.

Not because the model was stale. Because the model was right. Their LinkedIn page still carried the old description. So did Crunchbase. So did four industry directories, two conference speaker bios, and the boilerplate at the bottom of every press release they’d issued since 2019 — which their PR agency had copied forward, faithfully, for six years.

The website said one thing. Eleven other places said another. The system did what it is built to do: it went with the consensus.

Why this is lever one

Chapter 3 established the fact that reorganizes everything: most questions never trigger a search. For those, the answer comes from what the model already holds — which is a function of how consistently and how widely your company has been described across the web, over years.

The correlational evidence points the same direction. Ahrefs studied roughly 75,000 brands and measured how strongly various signals track AI visibility:

Two things jump out. Being talked about beats being linked to. And publishing more is almost irrelevant — the weakest signal on the list is content volume, which is where most budgets go.

The distribution is brutal, too. Brands in the top quartile for web mentions averaged 169 AI Overview mentions. The next quartile down averaged fourteen. The bottom half registered between zero and three.

One honest caveat, which the researchers state themselves: correlation is not causation, and there’s an obvious confound. Big companies have more mentions, more YouTube presence, more branded search, and more AI visibility — all downstream of being big. Nobody has run the controlled experiment. What the numbers establish is the shape of the thing, not the mechanism.

The consistency audit

Find out what the ecosystem currently believes about you. This takes about a week and it’s the highest-yield week in the whole program, because it nearly always surfaces contradictions nobody knew existed.

First, write down the canonical facts. One page, signed off by whoever owns positioning:

Legal name, trading name, every former name

A one-sentence category statement: X is a [category] that [does what] for [whom]

Founded date, headquarters, employee band, ownership status

Product names and their categories

Founder and executive names with titles

Official domain, and any other domains you own

Then check every place your company is described against it. Homepage and About page. Your Organization schema. LinkedIn. Crunchbase. Wikidata. Wikipedia, if you have an article. Google Business Profile. Industry directories — G2, Capterra, Clutch, trade associations. Review platforms. App store listings. Your executives’ own LinkedIn profiles. And the press boilerplate on your last ten releases.

That last one catches more errors than any other check on the list.

Score it by counting contradictions, not properties. A category described four different ways across nine sites is four contradictions. That’s a project with an end, which makes it fundable.

Building the reference point

Somewhere has to be the definitive statement of what your company is. Usually the homepage or the About page. It needs four things.

Plain text identity. Somewhere in the HTML, in a sentence a machine can lift: “Acme Logistics is an enterprise supply chain software company serving mid-market manufacturers in North America.” Not a tagline. Not a video. Words.

Brands resist this because it reads flat next to the copy the site was designed around. Put it in the first paragraph of the About page if the homepage can’t carry it. But it has to exist, in text, in the initial HTML.

Organization schema with a complete sameAs. This is the one piece of structured data worth building regardless of anything else in this book, because its job isn’t citation — it’s disambiguation. The sameAs array is your machine-readable assertion that all these scattered profiles are the same company, which is exactly the puzzle the knowledge graph is trying to solve on its own.

{ "@context": "https://schema.org", "@type": "Organization", "name": "Acme Logistics", "legalName": "Acme Logistics Holdings, Inc.", "url": "https://example.com/", "description": "Enterprise supply chain software for mid-market manufacturers.", "foundingDate": "2016-04-12", "sameAs": [ "https://www.wikidata.org/wiki/Q00000000", "https://www.linkedin.com/company/acme-logistics", "https://www.crunchbase.com/organization/acme-logistics", "https://www.youtube.com/@acmelogistics" ] }

Maintenance. A stale reference page teaches the system your facts are unreliable. Quarterly review, named owner.

Identities for your people. If your expertise argument rests on named humans, those humans need resolvable identities: consistent bylines, author pages with real credentials, author markup pointing at them, matching LinkedIn and speaker profiles. An expert the system can’t identify contributes nothing.

Wikidata: an hour well spent

Wikidata’s bar is much lower than Wikipedia’s. Wikipedia wants significant coverage in multiple independent secondary sources. Wikidata wants a clearly identifiable entity describable with serious public references. Most real businesses qualify for Wikidata. Most do not qualify for Wikipedia.

The process takes about an hour. Search first, using your exact legal name, because duplicate items are common and annoying to merge. Create an account under a real name or a clearly branded handle, and disclose any paid relationship. Add your label, a short disambiguating description, and aliases including former names. Save it and record your Q-identifier — that’s the durable machine-readable handle for your company, and it belongs in your sameAs.

Then add referenced statements: what kind of organization, country, headquarters, inception date, founder, official website, industry, and external identifiers.

The failure mode is simple. Statements without references get reverted. Your own blog and your own press releases don’t count as references. Thin entries get deleted, and a deleted item is harder to rebuild than a good one is to create.

Wikipedia: read this before your agency pitches you

Most companies don’t qualify, and pursuing an article anyway is an active risk.

Paid advocacy is forbidden. Paid editing must be disclosed — employer, client, affiliation — and failing to disclose violates the Wikimedia Terms of Use. You are not supposed to edit the article directly; the sanctioned route is a request on the talk page with full disclosure, which may simply be declined. Getting it wrong can mean account blocks, exposure under FTC guidelines and European fair-trading law, a press cycle about the attempt, and permanent loss of control over what the article says.

Notability for companies requires significant coverage in reliable secondary sources independent of you. Funding announcements, press releases, routine trade coverage, and interviews with your own executives generally don’t count.

An agency promising to get you a Wikipedia page is selling you either a terms-of-use violation or a deletion debate. Fund Wikidata, which you can legitimately build, and let Wikipedia follow real notability if it ever arrives.

There’s an irony here worth noticing. Wikipedia is the most-cited domain in AI answers, somewhere between 5 and 13 percent of ChatGPT citations depending on the study. Its own human pageviews fell about 8 percent year over year, which the Wikimedia Foundation attributes partly to generative AI. The most valuable source in the answer economy is being drained by it.

The entity checklist

Write the canonical facts page and get positioning sign-off — 1 day

Audit every external property against it; count contradictions — 1 week

Fix them, starting with LinkedIn, Crunchbase and press boilerplate — 2–4 weeks

Put a plain-text identity statement on the homepage or About page — 1 day

Ship Organization schema with a complete sameAs — 1 day

Create a referenced Wikidata item; record the Q-ID — 1 day

Build author pages for your named experts — 1 week

Brief the PR agency; update the boilerplate everywhere — 1 day

Put a quarterly consistency review on someone’s calendar — ongoing

None of this is expensive. Most of it has been sitting undone in every organization I’ve looked at, because it belongs to nobody in particular.

Assign it.