Chapter 13: Brand and Entity

Last updated: November 1, 2026.

Imagine a customer calls and says, “ChatGPT told me you make this model,” when that model went out of production years ago. Or AI mixes up your brand with another company that has the same name. Most brands only start wondering how AI describes them after something like this happens.

I listed these risks in Chapter 3 and left the details for Chapters 13 and 14. This chapter is the first half of that work: does AI know your brand, and if it does, does it know it correctly?

At the center of this is the idea of an entity. Search engines and AI models try to store information not as strings of words but as connected “things,” and your brand is one of those things. The more clearly and consistently it’s stated what you are, what you do, where you are and who you’re connected to, the better the odds that AI describes you correctly.

This chapter matters most for small and mid-sized brands. AI already knows the big brands well. The real risk is with brands that little has been written about online.

1. From words to entities

Google announced its Knowledge Graph in May 2012 with a post titled “things, not strings” (Google). The example in that post explains the idea well. For years, when someone typed “taj mahal,” the search engine saw just two words. But those words could mean one of the world’s most famous monuments, a Grammy-winning musician or a casino in Atlantic City. The Knowledge Graph was meant to make that distinction and work out which “thing” the user meant.

At the time, Google said the Knowledge Graph held more than 500 million objects and more than 3.5 billion facts about and relationships between them. In a May 2020 post it said the number had grown to more than 500 billion facts about five billion entities (Google). According to the same post, this information comes from content on the web and from open-source and licensed databases; Wikipedia is a common source but not the only one.

Take a hypothetical filling machine manufacturer and call it Örnek Dolum. To a search engine, “Örnek Dolum” isn’t a string of words but a thing with properties like these:

• It’s a company.
• It makes filling machines.
• It’s headquartered in Bursa.
• It was founded in 2005.
• One of its products is a machine called the X-200.
• Its founder is a specific person.

Each of these facts is a relationship. When AI answers “filling machine manufacturers in Bursa,” it draws on these relationships. If one of them is missing or wrong, the brand either doesn’t make it into that answer at all or makes it in with the wrong information.

Entity work means making sure the basic facts about your brand are written clearly, and the same way, everywhere on the web.

2. Where does AI learn about your brand?

We saw AI’s two sources of information, the model’s memory and live search, in Chapter 3. When it comes to brands, that split creates two separate jobs.

In memory, the model forms an impression from what it read about your brand in its training data. A study by Kandpal and colleagues, “Large Language Models Struggle to Learn Long-Tail Knowledge,” showed that a model’s chance of answering a question correctly is strongly tied to how many documents related to that question it saw during training (arXiv). The study also says one way to reduce that dependence is for the model to pull in outside information when it answers.

A brand with thousands of news stories, reviews and forum threads holds a strong place in the model’s memory. A small business with a few pages about it online counts as “long-tail” knowledge. The model either knows little about a brand like that or mixes up the little it knows with other information. That’s where the risk of made-up answers is highest.

Live search works differently. When a user asks something current or specific about your brand, the model often searches and writes the answer from the pages it finds: your site, your Business Profile, directories, news, complaint sites. The problem here isn’t a gap in memory but what those pages say.

The two sides also run on different time scales. Memory is long-term work: the more the web writes about your brand, and the more consistently, the better the next model update will know you, and you can’t change that in a week. With live search, fixing the pages the model will find can pay off within weeks. For a brand that AI gets wrong, that’s always the first step.

There’s an upside here for small brands. If the model doesn’t know your brand well from memory, it will most likely search, which brings in the side you can control. But if it runs into outdated or inconsistent information when it searches, it passes that along with the same confidence.

3. Measure first: how does AI describe you?

Before you fix anything, you need to know where you stand. Put together a standard set of questions about your brand and ask them regularly on different platforms. The eight questions below are a good start.

The last row is different from the others. The first seven questions include your brand name and measure how accurately the model knows you. The last one measures whether you come to mind within your category. A model can know you correctly and still never mention you in a category question; those are separate problems.

Ask the questions on each platform separately. ChatGPT, Gemini, AI Mode, Copilot and Perplexity look at different sources, and you can be described correctly on one and wrongly on another; we covered platform differences in Chapter 4. Ask each question several times, too. As we saw in Chapter 3, answers change every time, and an error that shows up in one answer doesn’t carry the same weight as one that shows up every time. Also note which pages are linked under the answer. That’s the fastest way to find where the wrong information came from.

I described the systematic version of this measurement in Chapter 7. The only thing to add here is to keep brand questions as a separate set from category questions.

4. The company information on your site

Your own site is one of the first places AI looks when it searches for something about your brand. Even so, many sites don’t state the brand’s basic facts plainly. The About page is full of lines like “Industry leader, innovative solutions,” while the founding year, address, product groups and service area are either missing or tucked into an image in a corner of the page.

Your site should have one clear page that says who the brand is. Usually that’s the About page. It should state the following in plain text:

• The brand’s full legal name, and its short name if it has one.
• What it does, in one sentence: “Örnek Dolum is a company that makes liquid filling machines for cosmetics and cleaning products.”
• Founding year and headquarters address.
• Product or service groups.
• Regions and countries it serves.
• Founders or executives.
• Certifications, memberships and notable references.
• Official social media and profile URLs.

The list may look ordinary, but it covers nearly everything an AI needs to answer “What is Örnek Dolum?” correctly. As we discussed in Chapter 3, the model reads passages. A sentence that makes sense on its own, like “Örnek Dolum has been making filling machines in Bursa since 2005,” makes its job much easier.

Adding structured data to this page is also a good idea. Google’s guide to Organization structured data says the markup can help Google better understand your organization’s administrative details and tell you apart from other organizations (Google Search Central). According to the guide, no properties are required; you’re encouraged to add the ones that apply to your organization. You don’t need it on every page. Putting it on the home page or on a single page that describes the organization, such as the About page, is enough.

Two properties are especially useful. The sameAs property holds the URLs of pages on other sites with more information about your organization; in Google’s example, these are profiles on social media or review sites. It’s how you say “this LinkedIn page, this YouTube channel and this website belong to the same brand.” Identity properties like the legal name (legalName) and tax ID (taxID) also help separate you from organizations with the same or similar names.

Structured data isn’t required, but it helps; I said the same in Chapter 2. Google’s AI optimization guide also says no special schema.org markup is needed for AI search. Still, giving the same information in plain text and in structured data together lowers the chance of being misread.

5. The same brand everywhere

Fixing the site is only part of the job. AI doesn’t learn about your brand from your site alone. Google Business Profile, Bing Places, Apple Business Connect, Yandex Maps, LinkedIn, social media accounts, B2B directories, marketplaces, industry associations, old press releases… There are dozens of places that give out information about your brand.

If the information in these places doesn’t match, AI has to choose, and it doesn’t always choose the right one.

A hypothetical scenario: two years ago Örnek Dolum moved its factory from an old building in central Bursa to an organized industrial zone. It updated its site and Business Profile. But three industry directories, an old trade show exhibitor list and a few news sites still show the old address. When a user asks “Where is Örnek Dolum’s factory?”, the model sees two addresses, and more sources list the old one. The odds of the old address showing up in the answer are far from low.

The basic facts to check for consistency:

• Name. Is the brand written the same way everywhere? Are “Örnek Dolum,” “Örnek Dolum Makineleri,” “OrnekDolum” and “Örnek Makine” one brand or several? That isn’t always clear to AI.
• Address and phone. This is the information that breaks most often, especially after a move or a branch closing.
• What you do. If one directory lists you under “packaging machinery,” another as a “cosmetics manufacturer” and another as a “machine importer,” the model may put you in the wrong category.
• Product list. Discontinued products live on in old catalogs, on marketplaces and on dealer sites.
• Founders and executives. A partner who left or a general manager who was replaced can stay the face of the brand in old news stories.

Build an inventory for this. Search your brand name on Google, on Bing and on AI platforms. Put every profile, directory listing and significant news story you find into a spreadsheet. For each one, mark whether the address, phone, description and product information are correct. Then fix them, starting with the most visible.

This work is tedious. But I think it’s one of the lowest-risk, most concrete things a brand can do for its AI visibility. You can fix the profiles you control in a day. Directories and marketplaces take a bit longer. You often can’t change the information in old news stories; you reduce their influence by adding to the amount of current, correct information.

6. The Knowledge Graph and the knowledge panel

When you search for a brand on Google, the box on the right side (at the top on mobile) is called the knowledge panel. It’s the visible face of what Google’s Knowledge Graph holds about that brand.

If you have a knowledge panel, Google recognizes you as an entity. But the panel can contain wrong information, and Google offers two ways to deal with that (Google). Anyone can report wrong information with the “Feedback” link at the bottom of the panel; Google says it reviews these reports to understand how errors get through its systems.

The second way is to claim the panel. A person or company can claim its own panel through the “Get verified on Google” process. According to Google’s help page, you click “Claim this knowledge panel” at the bottom of the panel and sign in to one of the official sites or profiles listed in it, such as a YouTube, Search Console, X (Twitter) or Facebook account (Google Help). Once verified, you can suggest changes to the panel and tell Google which photo you prefer.

The help page also draws a distinction. For local businesses that serve customers at a specific address or within a service area, the right tool isn’t the knowledge panel but Google Business Profile. The two look alike but are separate systems. And not every knowledge panel can be claimed.

If you don’t have a knowledge panel, don’t worry. Not having one doesn’t mean Google doesn’t know you at all. But for a panel to appear, Google needs to find enough consistent information about your brand. The road to a panel runs through the rest of this chapter: a clear company page, consistent profiles and trustworthy sources that mention your brand.

7. Wikipedia and Wikidata

According to Google’s 2020 post, Wikipedia is a common source for the Knowledge Graph. Wikipedia and its structured-data sibling, Wikidata, are sources AI models know as well. That’s why many brands get the idea of opening a Wikipedia page for themselves. Be careful here.

Writing about your own brand on Wikipedia counts as a conflict of interest. Under Wikipedia’s policy, if you were paid or expect to be paid for a contribution, you have to disclose it; this is a requirement of the Wikimedia Foundation’s terms of use (Wikipedia). Paying an agency to quietly write an article about your brand breaks the rules and can get the article deleted.

Wikipedia doesn’t accept articles about every company, either. The subject has to be covered well enough in independent, reliable sources. Independent sources need to write about you first; a Wikipedia article can follow from that.

Wikidata’s rules are a little looser, but they exist. Under Wikidata’s notability policy, an item is accepted if it has a valid page on a Wikimedia project, if it refers to a clearly identifiable entity that can be described using serious, publicly available sources, or if it fulfills a structural need for other items (Wikidata).

If a Wikipedia article or Wikidata item about your brand already exists, check whether its information is correct. If something is wrong, follow the rules: disclose your conflict of interest and suggest a correction on the talk page, citing an independent source. If there’s no article, don’t force one. Focus first on getting independent, reliable sources to write about your brand; we’ll get to that in Chapter 14.

8. When your name gets mixed up with something else

Some brand names are a common word, a place name or the same as another company’s name. That makes it harder for AI to match you to the right entity. Google’s 2012 “taj mahal” example was about this same problem.

If your brand name is easy to confuse:

• Use your name with the same qualifier everywhere. Instead of just “Örnek,” use a name that carries the category, like “Örnek Dolum Makineleri.”
• Put the facts that set you apart up front. City, industry, founding year and main product should appear in the first paragraph and in profile descriptions.
• Link your profiles together. Link from your site to your official profiles and from your profiles back to your site. The sameAs property in structured data is the machine-readable version of this.
• Spell out how you differ from the brand you’re confused with. If it happens often, even one sentence on your FAQ page can help: “Örnek Dolum has no connection with the company X of the same name.”

The last item may seem odd. But when AI is looking for an answer to “Are A and B the same company?” and finds a clear sentence like that, it can use it. If it doesn’t find one, it guesses.

9. The people behind the brand

The company isn’t the only entity. Your founder, your engineers and your content writers are entities too. Google’s 2020 post explains that knowledge panels can be claimed by people as well as companies.

In Chapter 12 we covered expert opinion as one of the kinds of content that can’t be copied. If it’s unclear who the expert is, that contribution only goes halfway. Google’s content self-assessment questions also list background information about the author and the site, such as author pages and the site’s About page, among the signals that build trust (Google Search Central).

In practice:

• Create profile pages for authors and experts. The person’s name, title, experience, area of expertise and the content they’ve written should live on that page.
• Connect the person to the company. The profile page should link to the person’s official profiles, such as LinkedIn, and those profiles should link back to the company site.
• Use the same name and title for the person everywhere. “Mech. Eng. Ayşe Y.” in one place and “Ayşe Yılmaz, Technical Director” in another can read as two different people.
• Update information about people who have left. Pages that still show a departed expert as “our chief engineer” are another piece of outdated brand information.

I think the same logic applies to this book. When it’s clear who an idea belongs to and which organization that person works with, the idea keeps its owner even when it’s carried into someone else’s answer.

10. Are other people talking about you?

Everything so far has been about how the brand describes itself. When AI evaluates a brand, it also looks at what others say about it.

A study Ahrefs published in May 2025, covering 75,000 brands, paints an interesting picture (Ahrefs). The three factors most strongly correlated with a brand being mentioned in AI Overviews were all off-site factors:

• Brand mentions on the web: 0.664 correlation.
• Branded anchor text: 0.527.
• Branded search volume: 0.392.

The correlation for the classic backlink count was only 0.218. Branded ad traffic and ad spend also showed a weak relationship. Ahrefs also notes that brands in the bottom 50% for web mentions barely appeared in AI Overviews at all.

This is a correlation study. Heavily mentioned brands are already big, well-known brands, and the data can’t tell whether mentions lead to visibility or whether both come from the size of the brand. Google also says plainly in its AI guide that chasing artificial “mentions” isn’t as useful as people think (Google Search Central). According to the guide, its core ranking systems focus on high-quality content, other systems keep spam out, and AI features rely on both.

Being mentioned is valuable, but there’s no shortcut to getting mentioned. Fake forum posts, paid content placements and the same promotional text spread across dozens of sites are exactly what Google and the AI platforms try to filter out. Real mentions come when there’s a real reason: new data, a news story, a customer experience, an expert view. That’s why Chapter 12 on content that can’t be copied and Chapter 14 on third-party sources should be read alongside this chapter.

11. Fixing a brand that AI gets wrong

If AI is saying something wrong about your brand, set your expectations right from the start. There’s no official brand dashboard where you can tell ChatGPT or Gemini to correct a fact about your brand. Corrections happen indirectly, by fixing the sources the model reads. You can work in this order:

1. Record it. Save the question, the platform, the date and a screenshot of the answer where the wrong information appeared. Ask the same question several times and note how often the error repeats.
2. Find the source. Check the source links under the answer. If there are none, search Google and Bing for the wrong information word for word. An old directory listing, the Business Profile of a closed branch, an old catalog PDF or a forum post is often where the problem starts.
3. Fix your own site first. If the wrong information comes from an old page on your site, update it, redirect it or remove it. Cleaning up old pages, which we discussed in Chapter 3, is the most effective step here.
4. Then fix outside sources. Update the profiles you control right away. Send correction requests to directories, marketplaces and dealers. If a news site has a serious error, write to the editor.
5. State the correct information clearly. If the model has nothing newer to compare old information with, it may keep using the old. For a discontinued product, a page or paragraph that says “Production of the X-200 stopped in 2024; the X-300 replaced it” works better than a product page that was quietly deleted.
6. Give the platforms feedback. Google asks you to report wrong, inaccurate or problematic answers with the thumbs icons under AI Overviews and the “Report a problem” option (Google Help). For knowledge panel errors there’s the panel’s feedback link. In ChatGPT you can use the feedback buttons under an answer and OpenAI’s model behavior feedback form (OpenAI). These don’t fix individual answers, but they put the problem in front of the platforms.
7. Wait and measure again. Answers based on live search can change once the source pages are updated and recrawled. An error in the model’s memory can last until the next model update. So ask the same question set again a few weeks after the fix.

While you’re doing this, don’t publish new content that repeats the wrong information over and over. Pages like “Common misconceptions about the X-200” that repeat the false claim in headings and paragraphs are, in my view, an unnecessary risk; they add to the number of copies of that claim online. Writing the correct information briefly, clearly and with a date is enough.

12. Tracking brand strength

The most direct way to measure where a brand stands in AI is to ask the question set from the start of this chapter on a regular basis. Beyond that, there are two more sources of data on a brand’s overall strength, and both are free.

The first is branded search. In November 2025 Google added a branded queries filter to the Search Console performance report; since March 2026 it has been available for all eligible sites (Google Search Central Blog). The filter automatically separates out queries that include your brand name, common misspellings and products associated with the brand. According to Google, this split comes from its own classification, not a simple keyword rule.

In Chapter 1 we discussed how a user who sees your brand in an AI answer and doesn’t click may later come to Google and type your brand name. If your AI visibility is growing, one trace of that may be a rise in branded searches. It isn’t proof on its own, but it’s a signal worth tracking.

The second is what customers tell you themselves. Add ChatGPT, Gemini and other AI tools as options to the “How did you hear about us?” question on quote forms and in sales calls. Customers can tell you what Analytics can’t see.

We covered both in detail in Chapter 8, on measuring AI traffic and business results. Since brand work pays off slowly, it makes more sense to look at these numbers quarterly and yearly rather than monthly.

Write the description yourself

AI tries to understand your brand not as a word but as an entity with properties and relationships. If you don’t write the description of that entity clearly and consistently, AI builds it from what others have written and from its own guesses.

The order of work is straightforward. First use the question set to see where you stand, then fix your own site, then the profiles and directories. Since there’s no correction dashboard for brands, every fix goes through the sources the model reads.

Some of those sources aren’t under your control. What gets written about your brand on Reddit, YouTube, complaint sites and industry publications is the subject of the next chapter.

Checklist

Sources

• Google: Introducing the Knowledge Graph: things, not strings (May 16, 2012)
• Google: Google’s Knowledge Graph and knowledge panels (May 20, 2020)
• Google Knowledge Panel Help: Get verified on Google (accessed November 1, 2026)
• Google Search Central: Organization structured data (last updated September 8, 2026)
• Google Search Central: Optimizing your website for generative AI features on Google Search (last updated July 10, 2026)
• Google Search Central: Creating helpful, reliable, people-first content (last updated October 5, 2026)
• Google Search Help: AI Overviews (accessed November 1, 2026)
• Google Search Central Blog: Introducing the branded queries filter in Search Console (November 20, 2025, updated March 11, 2026)
• arXiv: Large Language Models Struggle to Learn Long-Tail Knowledge (Kandpal et al.) (November 2022, updated July 2023)
• Ahrefs: An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied) (May 26, 2025)
• Wikipedia: Paid-contribution disclosure (accessed November 1, 2026)
• Wikidata: Notability (accessed November 1, 2026)
• OpenAI: Model behavior feedback (accessed November 1, 2026)

Previous: Chapter 12, Content That Can’t Be Copied · Table of contents · Next: Chapter 14, Third-Party Sources