Why do AI engines put your company in the wrong category?

Why does an AI engine describe your company as the wrong kind of business?

Because it assembled that description from every page that mentions you, not from the one page where you explain yourself properly. If your own site, your profiles and your press coverage describe you three different ways, something has to pick, and it will not always pick yours.

This is the complaint I hear most often about answer engines, and it is almost never a mysterious model failure. It is a consistency failure that was already sitting on the site, invisible, because humans are good at ignoring contradictions and retrieval systems are not.

The useful news is that the fix is mostly editorial rather than technical. The unhelpful news is that it involves auditing pages you wrote years ago and had forgotten were still live.

What does a search or AI engine use to decide what you are?

Everything it can read, weighted by how confident it is. Your homepage, your about page, your headings, the anchor text pointing at you, and third-party pages describing you. Google also uses specific structured data properties for this, and it says so plainly.

Google's Organization structured data documentation states that some properties are used behind the scenes to disambiguate your organization from other organizations, naming iso6523 and naics as examples. That is Google telling you outright that disambiguation is a job it performs and that you can feed it. Most sites feed it nothing.

The same page is clear about what it recommends. There are no required properties; instead, Google recommends adding as many properties that are relevant to your organization. It suggests focusing on properties useful to your users, such as name or alternateName for your business name, plus an indication of real-world presence like address or telephone and online presence like url or logo.

Does adding schema fix a category problem?

Not by itself, and not in the way vendors imply. Google's own guide to generative AI features says you do not need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search including its generative AI capabilities, because Search itself does not use them.

Read those two documents together and the picture gets clearer rather than contradictory. Structured data helps Google work out which entity you are. It is not a magic input that changes what a generative answer says about you. Anyone selling schema markup as an AI visibility product is selling you the wrong causal story.

So I treat schema.org markup as hygiene, not strategy. Get the Organization block right once, on the page Google recommends, and move on. Google advises placing this information on your home page, or a single page that describes your organization, for example the about us page. One place, done properly, beats markup scattered across forty templates.

Why does your own site disagree with itself about what you do?

Because nobody ever went back. Your homepage says one thing after the last repositioning, your about page still says what you did two pivots ago, an old service page describes an offer you retired, and a blog post from three years ago introduces you as something else entirely. All of it is still indexed.

I find this on almost every site I audit, and the giveaway is the about page. It is the page founders write once, at the beginning, with the most feeling and the least review. It ages worse than anything else on the site and it happens to be exactly the page Google points at for organizational information.

The exercise that fixes it is boring and effective. Search your own domain for the phrases that describe your old positioning, then read every page that comes back and either update it, consolidate it or retire it. You are not optimising here. You are removing the contradictions that force a machine to guess.

What does Google say it actually needs to show you in AI answers?

Something simpler than most people expect. Its guide states that to be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. Eligibility, not a special technique.

That reframes the whole problem. If a page describing what you do is not indexed, or is blocked from showing a snippet, it cannot participate in the description of you, no matter how well written it is. Before you rewrite anything, confirm the pages that define you are actually eligible to be read.

Google's guidance on the writing itself is equally unglamorous: write content for your human audience, make sure it is well written and easy to follow, and organise pages by paragraphs and sections with headings that provide a clear structure to navigate content. That is the same advice as a decade ago, which is either reassuring or annoying depending on what you were hoping to hear. If you want to know whether any of it moved the needle, you need a baseline for AI visibility before you change anything.

How does the rest of the web override your homepage?

By outnumbering it. You control one site. Everyone else controls the directories, review sites, profiles, partner pages and articles that also describe you, and collectively they say more words about your company than you do. Weight of consistent description matters.

You can tell this is expected behaviour from the structured data itself. Google defines sameAs as the URL of a page on another website with additional information about your organization, giving the example of a profile page on a social media or review site. The property exists because the model of your identity is assumed to span multiple sites, not just yours.

That is why the highest-value work is often off your own site. Go update the profiles you abandoned. Fix the one-line description on the directory that describes a service you no longer sell. It feels like admin rather than marketing, and it does more for how you are described than another blog post will. It is closely related to why answer engines cite competitors instead of you.

Should you describe yourself differently in different places?

No, with one narrow exception. Use the same sentence about what you do everywhere, in the same words, with the same category noun. Variation feels like good writing and reads to a retrieval system like uncertainty about what you are.

The exception is a genuinely different name rather than a different description. Google's documentation defines alternateName as another common name that your organization goes by, which is the correct home for a trading name, an old brand or an abbreviation people actually use. That is not inconsistency, it is telling the system that two labels point at one entity.

So pick one canonical category noun and repeat it without embarrassment. If you are an AEO and GEO consultancy, be that on every page, in your profiles and in your markup. Writers hate this because repetition feels lazy. One canonical term per concept is how you stop being described as three different companies, and it is the same reason third-party descriptions of you can outrank your own.

How do you check whether the fix worked?

Ask the engines the plain question a buyer would ask, and write down the answer before you change anything. "What does this company do" and "who are the alternatives to this company", across several engines, recorded with the date. Without that record you will be guessing about progress.

Then give it time and re-ask the identical questions. What you are watching for is not a ranking but a description: does the category noun match yours yet, are the competitors listed the ones you actually compete with, has the outdated service stopped appearing. Those three shifts are the real signal that your identity work landed.

On the Google side, Google announced Search Generative AI performance reports in Search Console on its Search Central Blog in June 2026, which is a better place to look than inferring generative traffic from wobbles in your overall numbers. Across 350+ articles I have written about answer engines, the consistent lesson is that people change five things at once and then cannot tell which one worked. Change the identity layer on its own.

What should you do next?

Ask an answer engine what your company does, then open your about page and read it honestly. If those two descriptions do not match, you have found your project, and it starts with rewriting one page rather than adding any markup at all.

Then work outward in this order: the about page, the homepage, the old pages that contradict them, the third-party profiles, and only then the Organization markup. That sequence matters, because markup describing a story your pages do not tell is just a second contradiction in a machine-readable format.

I am a Certified Webflow Partner and most of what I do is this unglamorous alignment work between what a company says it is and what the web has concluded it is. If an engine is describing your business as something you retired two years ago, reach out and let's go find the pages still saying so.