Entity SEO is making every source a machine reads about you agree: one name, one description, one set of numbers, on your site, in your schema, in your llms.txt and on your profiles. Models assemble an entity from all of those, and contradictions make them quote someone else instead.

Why does brand name consistency matter for AI?

Because a model holding two versions of you takes the safe route and skips you. Models forgive a small site but not a contradictory one. Identical spelling and one repeated description remove every excuse a machine has to doubt that the sources describe the same thing.

What should I check in an entity audit?

Six things: identical name spelling, one description reused word for word, current numbers everywhere, a permanent schema id plus sameAs profile links, an audit tied to every release, and a monthly read of what the models actually say about you.

Entity Hygiene: Same Name, Same Story, Everywhere

¶ CONTEXTTHE CROSS-CHECK

01A model never reads just one page about you

When an engine decides whether to repeat you, it does not read a page. It assembles a picture: what your site says, what your schema declares, what your llms.txt claims, what your profiles state, and what other people write about you. Then it checks whether those agree.

An entity is not what you call yourself, it is what every source about you agrees on. That is the whole job hiding behind the phrase entity hygiene.

What a contradiction actually costs you

People assume a contradiction produces a wrong answer. It rarely does. It produces no answer at all, because the safest move for a model holding two versions of you is to skip you and quote a source it does not have to reconcile.

That is why small brands with clean identities beat larger ones with messy ones. Models forgive a small site. They do not forgive a contradictory one.

¶ WAR STORYMY OWN MESS

02Three contradictions I shipped myself

I audit other people's entities for a living, which makes the following more embarrassing and more useful. Every one of these was live on my own properties.

The file that undercounted my own work

My llms.txt listed 15 products while 21 were live. I had shipped six things and never updated the one file whose entire job is introducing me to machines. For weeks the number a model was most likely to repeat about me was wrong, and nothing anywhere warned me. A stale file does not look stale to a machine. It looks like a claim.

The tool that never said its own name

I built a product whose job is checking whether AI mentions your brand, and its homepage was titled AI Visibility Tool. Accurate, descriptive, and it never stated the product name in a place a machine could lift. The site and the schema disagreed about what the thing was even called, so models described it instead of naming it.

The article with two different authors

The sharpest one was invisible to readers. On a site I run the visible byline named one author while the page metadata, inherited from a site wide default, named another. Every human saw one writer. Every parser saw a disagreement about who wrote it, on the exact pages I wanted attributed to me.

Every contradiction you publish is a reason for a model to quote someone else.

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

Say the same true thing everywhere and you become a fact.

¶ PLAYBOOKTHE CLEANUP

03The entity hygiene checklist

This takes an afternoon once, then ten minutes per shipped change. Open your site, your llms.txt, your schema, and every profile you own, side by side, and make them agree.

The six things that must match

[1]

One name, spelled identically. Not a variant on the profile and a shorter one in the footer. If a legal name and a known name are both true, do not choose between them. Declare the known one as the name and the other as alternateName in schema, which is exactly what my own Person block does. If two names for you are both true, say so in your schema instead of picking one and hoping.

[2]

One description, reused word for word. Write the single sentence you want repeated, then paste that same sentence into your title area, your llms.txt, your schema description and every bio. Rewriting it each time is how three versions of you get born.

[3]

One set of numbers, current everywhere. Product counts, founding year, languages supported. Numbers are the most quotable thing you publish, which makes a stale one the most expensive mistake you can leave lying around.

[4]

One identity graph. Give yourself a permanent id in schema, reference that same id as the author of everything you publish, and list your real profiles in sameAs. That is what tells a machine the accounts are one person rather than several.

[5]

An audit trigger tied to shipping. The llms.txt failure happened because updating it was not part of launching anything. Attach the check to the release, not to a calendar you will ignore.

[6]

A monthly read of the answer. Ask three engines who you are and read the replies as a stranger would. Contradictions surface there first, in plain language, before any dashboard shows them.

None of this is glamorous and none of it takes talent. It is filing work. But it is the cheapest trust you will ever buy, because you are not asking a machine to believe something impressive. You are only asking it to stop finding reasons to doubt you.

Tomorrow: the traffic chart on your dashboard, and the share of it that was never human.