Entity SEO: making sure Google knows who you are

Search for a business called "Apex Plumbing" and Google faces a question your customers never think about: which one? There are dozens across the world, plus pages that mention apex in the mountaineering sense, plus a defunct company by the same name. Before Google can rank anything, it has to resolve what things the words refer to. The machinery it uses for that is the entity system, and whether your business is a crisp, well-documented entry in that system or an ambiguous string of characters quietly shapes everything downstream: your rankings, your knowledge panel, your map listing, and now whether AI assistants can recommend you at all.

"Entity SEO" is the work of making machines certain about who you are. It's less glamorous than chasing keywords and more durable than most tactics, because you only have to establish an identity once, and everything you publish afterward accrues to it.

Things, not strings

The concept has an official birthday. In May 2012, Google announced the Knowledge Graph with a phrase that still defines the whole field: an intelligent model "that understands real-world entities and their relationships to one another: things, not strings". Instead of treating "taj mahal" as ten characters that might match pages, Google began treating it as possibly-the-monument, possibly-the-musician, possibly-your-local-restaurant, each a distinct *thing* with facts attached.

The database started at over 500 million entities and 3.5 billion facts. By 2020, Google described it as holding 500 billion facts about 5 billion entities, drawn from sources like Wikipedia, licensed databases, and the structured information sites publish about themselves. Every knowledge panel you've ever seen, the fact box beside search results, is a rendered entity entry.

Here's why this concerns a small business: your company either is one of those entities, with a name, category, location, and relationships the machine is confident about, or it isn't. When it is, Google can show a knowledge panel, connect your reviews to your website to your map pin, and rank you for searches about what you do in your area with confidence it's you. When it isn't, every mention of your name is just text, and the machine hedges.

AI search turned this from nice-to-have into infrastructure

For a decade, entity work was a mild optimization. Then search started answering in sentences.

An AI assistant composing "who should I hire?" answers can only recommend things it can identify. And the citation research shows exactly where assistants anchor identity: Wikipedia alone is ChatGPT's most-cited source, and company homepages make up nearly a quarter of its top citations, while Yext's study of 17.2 million AI citations found that verified, structured business data accounted for more than half of distinct citation sources. In other words, when an AI decides which brands to name, it leans precisely on the entity layer: reference databases, structured facts, and the canonical pages where a business says who it is.

That reframes the work. Your homepage isn't just a landing page anymore, it's the document machines read to learn what you are. Your consistency across the web isn't just tidiness, it's the corroboration that lets a cautious system state facts about you. Getting cited by assistants starts with being identifiable, because nobody, human or machine, recommends an ambiguity.

How Google learns who you are

Google assembles your entity from evidence, and the evidence has a hierarchy.

What you declare. Your website's plain-text statements ("Apex Plumbing is a family-run plumbing company serving Tauranga since 2009") and your structured data, which is machine-readable markup restating those facts in a standard vocabulary. That vocabulary, schema.org, is used across more than 45 million domains marking up hundreds of billions of items, and it's how you say "this is an Organization, this is its name, logo, address, phone" in a format that removes guesswork. Google's own documentation says organization markup helps it "understand your organization's administrative details and disambiguate your organization in search results". Disambiguate is the load-bearing word.

What managed databases say. Your Google Business Profile, the industry directories, review platforms, and, for some businesses, Wikidata, the free structured knowledge base of "items" that feeds many systems, including, historically, Google's own graph via Wikipedia and its sibling projects.

What the open web says. Mentions, links, citations in local news, supplier pages, chamber-of-commerce listings. No single one matters much. Their agreement is what matters.

Confidence comes from consistency across all three layers. Ten sources agreeing you're "Apex Plumbing Ltd, plumber, 14 Grey Street, Tauranga" build an entity. The same ten sources with three name spellings, two old addresses, and a defunct phone number build a puzzle, and machines respond to puzzles by hedging, which in practice means showing you less.

The entity checklist

The practical work, in priority order. Most items are one-time fixes plus light maintenance.

• Pick one canonical identity and enforce it. One exact business name, one address format, one phone number, one description of what you do, written down in a reference doc, and used identically everywhere: site footer, Google Business Profile, directories, social profiles, invoices. Every variation you allow is doubt you're injecting into the machine's model of you.
• Make your homepage state the facts in prose. Who you are, what you do, whom you serve, where you operate, since when. Machines read your homepage to learn your identity, and plain sentences remain the most reliably parsed format there is. A homepage that leads with a slogan and never plainly says what the business is fails its most important reader.
• Add Organization or LocalBusiness structured data. LocalBusiness markup (for businesses with a physical presence or service area) requires name and address and should carry hours, phone, and coordinates; Organization markup covers the rest. Include the `sameAs` property, which schema.org defines as a link that "unambiguously indicates the item's identity", pointing to your official profiles: Google Business Profile, Facebook, LinkedIn, Wikidata if you have an entry. One honest caveat, straight from Google's docs: structured data enables understanding and rich results, and Google claims no direct ranking boost from it. You're buying certainty, not rank, and certainty is what pays in the entity era. (The schema part of the small-business checklist covers implementation.)
• Claim and complete your Google Business Profile. For any business serving a local market, this is the single most direct write-access you have to Google's understanding of you: category, hours, service area, photos, reviews. Keep it complete, and keep it agreeing with your site.
• Sweep your citations. Once, painfully: find every directory and profile mentioning you, and fix names, addresses, and dead links to match the canonical identity. Then quarterly, lightly.
• Consider Wikidata, be honest about Wikipedia. Anyone can propose a Wikidata item, and for an established business with press coverage it's a legitimate, free anchor other systems can point at. Wikipedia, by contrast, requires genuine notability, and a self-promotional article will be deleted, burning goodwill. If you're not written about independently, skip it without regret: consistency everywhere else carries most small businesses fine.
• Give your people entities too. Author pages with real bios, the same name and headshot across your site and LinkedIn, credentials stated. Google's quality systems care about who stands behind content, and "who" is an entity question. This is the plumbing under the experience signals that make content credible.

What an entity cleanup actually looks like

The checklist compresses a lot of small actions, so here's the shape of a typical cleanup for a made-up but representative case: a fifteen-year-old family firm that's traded as "Harrison & Sons Electrical", "Harrison Electrical", and "Harrison and Sons" across its history.

Day one is the decision nobody had made: the canonical identity. The owners pick "Harrison Electrical" (the name on the van and the invoices), write the one-paragraph description, who, what, where, since when, and put both in a shared doc. The website gets the visible fixes the same week: the homepage's slogan-first hero section gains a plain second line stating the business, the footer gets the exact name, address, and phone, and LocalBusiness structured data goes on with the canonical details and sameAs links to the Google Business Profile, Facebook page, and the industry association's member listing.

Week two is the archaeology: searching every variant of the name plus the town finds eleven listings across directories, two with the old address from before the 2019 move, four with name variants, one with a dead phone number. Each gets claimed and corrected to match the doc, tedious, unskilled, and exactly the work that was never anyone's job before. The Google Business Profile, half-completed years ago, gets finished: categories, service area, hours, photos, and a habit of actually answering reviews.

What changes, over the following weeks and months: searches for the business name resolve cleanly to one confident result with a correct knowledge panel instead of a scatter of half-matches; "electrician [town]" searches connect to a profile whose details agree with the site they land on; and when someone asks an AI assistant about the firm, the answer draws on eleven agreeing sources instead of eleven arguing ones. None of it is dramatic, and all of it is durable, which is the entity work trade in miniature: no single fix matters much, and the accumulated consistency is the asset.

How to tell it's working

Entity health is checkable in an afternoon. Search your exact business name: does Google show a knowledge panel or your Business Profile, with correct facts, and does it confidently surface your site rather than a soup of similarly-named things? Search your name plus your city, and your category plus your city, and see whether Google connects you. Then ask the assistants: ChatGPT, Perplexity, and Google's AI mode, "what is [your business name]?" and "who does [your service] in [your city]?". Wrong facts in those answers trace back, almost always, to an inconsistency in the evidence layers above, and fixing the source fixes the answer on the machines' next pass. Expect lag, since nothing in search updates instantly.

The one-line version

Machines can only rank, cite, and recommend what they can identify, so make identification effortless: one canonical identity everywhere, a homepage and structured data that state it plainly, a complete Business Profile, and corroborating profiles that all agree. It's unglamorous work you do mostly once, and it's the foundation the whole AI-search era quietly runs on: things get recommended, strings don't.