AI assistants do not read your brand the way a customer does. They reconcile it. When ChatGPT, Gemini or Perplexity answers a question about your company, the model is matching strings, categories and facts across many sources at once. If those sources disagree, the model has to guess which version is true. Sometimes it guesses wrong. Sometimes it hedges. Sometimes it recommends a competitor whose signals are cleaner.
That reconciliation problem is what people mean by a brand entity. An entity is not a keyword and not a page. It is the thing the model believes exists: a company with a name, a category, a location, a set of people and a set of claims attached to it. Entity SEO is the work of making that belief accurate and stable.
Why AI gets brands wrong
Most inconsistency is accidental. A founder writes "we're a customer feedback platform" on the homepage, "voice-of-customer software" on LinkedIn, "survey tool" in a directory profile, and "CX analytics suite" in a press release written two years ago. Each phrase is defensible. Together they describe four different products.
A language model resolving your brand sees four candidate categories and no instruction about which one wins. The safest answer is often the vaguest one, or the one repeated by the most sources. If a directory you barely maintain says you are a survey tool, and that phrase appears in three other places, the model may adopt it.
The same thing happens with names. If your legal name, product name, domain and social handles do not line up, the model may treat them as separate entities or merge you with a similarly named company. Short, generic names are the worst case. So are rebrands, where old and new names coexist across the web for years.
The four facts to standardize first
You do not need perfect consistency everywhere. You need consistency on the attributes a model uses to identify you. Four matter most:
- Name. One canonical form, plus any legal or former names you want associated with it.
- Description. One sentence, roughly the same everywhere, describing what you do and for whom.
- Category. One primary category, with secondary categories only where they genuinely apply.
- Core facts. Founding year, headquarters, key people, and what you sell.
Write these down before you touch anything else. A one-page entity sheet that lives in your docs is enough. Every profile, bio, directory listing and schema block should be checkable against it.
Pick one description and stop rewriting it
Marketers get bored of their own description long before anyone else notices it. Changing the homepage tagline every quarter is a common cause of entity drift, because the old version stays live on partner pages, review sites and archived press coverage.
Aim for a description that is specific enough to be checkable. "We help B2B SaaS teams find and fix onboarding drop-off" is better than "we transform the customer journey." The first gives a model something to match. The second gives it nothing.
Where consistency actually breaks
The usual failure points, roughly in order of how much damage they do:
| Surface | Common problem |
|---|---|
| Homepage and about page | Different category language on each |
| LinkedIn and other social profiles | Outdated tagline, wrong industry field |
| Directory and review listings | Auto-generated descriptions nobody reviewed |
| Press and funding coverage | Old positioning frozen at the time of writing |
| Crunchbase and similar databases | Stale employee counts and categories |
| Your own subdomains and docs | Product described differently than the company |
You cannot fix all of these at once. Start with the surfaces that already rank or get cited, because those are the ones models are most likely to have seen.
Organization schema as a consistency anchor
Structured data does not make an AI assistant recommend you. What it does is give a machine-readable version of the same facts you state in prose, in one place, in a format that is unambiguous about which string is the name and which is the description.
For an entity problem specifically, Organization markup is the useful one. It lets you declare the canonical name, URL, logo, description, sameAs links to your official profiles, and founding details. The sameAs property is the part people skip and later regret: it is how you tell a machine that your LinkedIn page, your Crunchbase entry and your GitHub org are the same entity as your website.
If you are starting from nothing, our free schema markup generator builds JSON-LD for Organization, LocalBusiness, Product or FAQPage from a short form, which is faster than hand-writing the markup and less error-prone than copying a template from a blog post.
Two rules for using it well. First, the values you enter should match your entity sheet exactly, not approximately. Second, schema describes what is on the page. Do not use it to assert facts the page does not support.
Handling rebrands
A rebrand is the hardest entity problem, because you are asking models to merge two entities that were previously distinct. The goal is not to erase the old name. It is to make the relationship explicit.
- Keep the old name visible somewhere on the site, in a sentence that states the change. "Formerly Acme Analytics" is enough.
- Update
sameAsto include profiles under both names while they exist. - Update directory listings rather than deleting them. A deleted listing often leaves a cached copy that still describes the old company.
- Do not run two websites with two descriptions for long. If you must keep the old domain, redirect it and make the destination clear.
- Expect lag. Models are trained and refreshed on their own schedules, and you do not control them.
The same logic applies in reverse when you are the smaller company with a name similar to a larger one. You cannot force a model to separate you, but you can make the distinction easy: a distinctive description, a clear category, a location, and consistent profiles all give it something to attach to.
Similar names and disambiguation
If your name collides with another company, add disambiguating detail everywhere, not just on your homepage. Location, category and audience are the cheapest discriminators. A model that sees "Northwind, a logistics visibility platform for freight forwarders in Rotterdam" in six consistent places has a much easier job than one seeing "Northwind" alone.
This is also where your own content helps. Pages that answer specific questions in your category give models more distinct material to associate with you. Writing FAQ sections that double as AI answers covers how to phrase those so they are quotable, and comparison pages that AI assistants actually quote covers the pages that tend to get pulled into category questions.
A note on crawler access
Consistency only matters if assistants can read the sources. The documented controls differ by vendor, and they are not interchangeable:
- OpenAI's OAI-SearchBot surfaces sites in ChatGPT search; sites that opt out are not shown in search answers. Blocking GPTBot affects training collection, not search.
- Anthropic's ClaudeBot collects content for training, Claude-User fetches pages for user questions, and Claude-SearchBot indexes for search. Anthropic says its bots honor robots.txt.
- PerplexityBot surfaces and links sites in search results; Perplexity-User fetches pages for user questions and generally ignores robots.txt because the user asked for it.
- Google-Extended controls Gemini training and grounding, not Search inclusion. AI Overviews follow normal Googlebot crawling.
robots.txt expresses intent, not access control, and changes can take about 24 hours to propagate. If your entity facts live on a page that search bots cannot reach, the rest of this work does not matter.
Check your own entity in one sitting
Open your homepage, your LinkedIn page, your two most important directory listings and your schema markup side by side. Write down the name, description and category each one uses. If they do not match, you have found the problem, and the fix is editing, not strategy. Start with the two that disagree most, align them to your entity sheet, and update the sameAs list so the profiles are explicitly linked. Then re-check in a month, because the old versions will still be out there for a while.