Skip to content
DeutschBook an intro call

Explainer · AI search & GEO

Entity SEO: why AI search mixes up your company with others

Definition

Entity SEO is the part of search engine optimization that describes a company and the people behind it so unambiguously that search engines and AI systems identify them correctly and don’t confuse them with others of the same name. It involves consistent details everywhere, structured data on your website and well-maintained profiles such as a Google Business Profile.

Key takeaways

  • An entity is a uniquely identifiable thing, such as a company or a person. Google stores facts about entities in its Knowledge Graph.
  • AI systems mostly mix up companies that share a name, have little written about them online or have outdated or inconsistent details.
  • Google and Microsoft recommend consistent names and details, structured data on your website and up-to-date business listings.
  • There’s no evidence that sameAs links or a Wikidata item get you mentioned in AI answers. You can’t apply for a knowledge panel.
  • If AI gets your company wrong, correct the source of the information first. Then report the error to the AI provider.

What is an entity?

An entity is a uniquely identifiable thing, such as a company, a person, a place or a product. Google stores facts about entities and their relationships in its Knowledge Graph. Entity SEO aims to make sure that search engines and AI systems attribute every fact about your company to the right company.

In a patent, Google describes an entity as “a thing or concept that is singular, unique, well-defined and distinguishable,” such as a person, a place or an idea (Google patent US 10,339,190 B2 (external site)). A patent shows how Google describes a method. It doesn’t show whether or how Google uses it.

So that machines can recognize the same entity everywhere, it gets an identifier. In Google’s Knowledge Graph Search API (external site), every entity has its own identifier, such as kg:/m/0dl567 in the documentation’s example. In Wikidata, a free knowledge base, every item is identified by a Q followed by a number, such as Q42 for Douglas Adams (Wikidata’s introduction (external site)).

On your website, you can link your company to entries like these with structured data. The sameAs property from the Schema.org vocabulary points to a page that “unambiguously indicates the item’s identity,” such as a Wikidata entry or the official website (Schema.org on sameAs (external site)).

How do search engines and AI systems know who you are?

Google builds its picture of a company from several sources: the company’s own website, profiles, licensed data and input from users. AI systems also know a company from the text their language model was trained on, or they search the web when asked. Any of these routes can carry outdated or wrong details.

Our guide to generative engine optimization (GEO) compares how each AI system chooses its sources. For the facts about your company, four sources matter most:

Every route into an AI answer can carry wrong details
Every route into an AI answer can carry wrong details Facts about a company reach an AI answer mainly by four routes: profiles and the company’s own website, the Knowledge Graph, the search index and the language model’s training data. Any of these routes can carry outdated or wrong details, for example through an old address on the website or because training data stops at a cutoff date. The table below gives the same facts with sources. Facts about your company Profiles and your website Google compiles profiles from your website and third parties An old address on your website can support a wrong edit Knowledge Graph Google’s AI Overviews and AI Mode draw on it Google doesn’t rule out errors Search index Bing and Copilot; ChatGPT sometimes via search partners Search results can be outdated or incorrect Training data Answers without a web search; Claude’s cutoff: June 2026 A later move isn’t in it AI answer about your company
Four routes into an AI answer, as of Nov 7, 2026
RouteWhat’s knownHow errors get inSources
Profiles and your websiteGoogle compiles Business Profiles from your website and third parties and says they can help you become visible in AI answersAn old address on your website can support a wrong edit to your profile8, 10, 19
Knowledge GraphGoogle’s AI Overviews and AI Mode draw on itGoogle doesn’t rule out errors7, 9
Search indexBing and Copilot rely on it, and ChatGPT sometimes partners with other search providers for its web searchSearch results can be outdated or incorrect11, 12
Training dataWithout a web search, a language model answers from what it learned in training; cutoff of the current Claude models: June 2026Anything that happened after that, such as a change of address, isn’t in it13

Why does AI mix up companies?

Mix-ups happen mainly when several companies share a name, when little is written about a company online, when old details can still be found or when names and addresses differ from source to source. A system then attributes facts to the wrong company or blends several companies into one answer.

Shared names

Search systems that retrieve documents for question-answering applications favor the better-known entity when two share a name. In a 2021 study, they retrieved wrong documents twice as often for questions about the less popular of two entities with the same name (Chen et al., 2021 (external site)).

Language models also struggle to keep entities with the same name apart. Researchers tested this with documents about different entities that share a name, such as several people named Michael Jordan. The leading models at the time often gave ambiguous answers or wrongly merged information about different entities (Lee et al., 2024 (external site)). Our assessment for small businesses: if a better-known company has the same name, its facts can easily end up in answers about you.

Little coverage

Language models know more about well-known entities than about little-known ones. A 2023 study with 14,000 questions about lesser-known entities found that models answer questions about less popular facts less accurately, and that larger models barely change this. By contrast, retrieving relevant documents and passing them to the model helped significantly (Mallen et al., 2023 (external site)). A second study found that the more documents relevant to a factual question were in a model’s training data, the better the model answered it (Kandpal et al., 2023 (external site)).

Researchers at OpenAI attribute hallucinations, meaning plausible but false statements, in part to individual facts with no pattern behind them in the data, such as birthdays (Kalai et al., 2025 (external site)). In our assessment, the owner, founding year or address of a small business is often exactly this kind of fact.

Outdated details

If you move or change your name, the old details can stay online for a long time, for example in directories and in articles on other sites. They can also remain in what language models learned in training. Bing’s guidelines advise notifying Bing quickly about new and changed content. According to Microsoft, this reduces outdated or incorrect links in Copilot’s answers.

Google may update a Business Profile itself if other sources report its details as incorrect or outdated. For certain edits that users suggest, Google notifies the owner before publishing them. If the owner doesn’t respond within four days, Google may publish the edit as long as other publicly available information, such as the business’s website, supports it (Google on updates to Business Profiles (external site)). So an old address on your website can support a wrong edit.

Inconsistent names

Bing’s guidelines call for “clear and consistent naming for people, organizations, products, and locations.” According to the same guidelines, a clearly defined entity improves citation accuracy. In a Business Profile, Google expects the name a business uses consistently on its storefront, website and stationery. Extras such as marketing taglines aren’t allowed in the name and can get the profile suspended (Google’s guidelines for Business Profiles (external site)).

What is a knowledge graph?

A knowledge graph is a database that stores entities and their relationships. According to Google, its Knowledge Graph held more than 500 billion facts about five billion entities in May 2020. It supplies the information in Google Search’s info boxes, called knowledge panels.

Google introduced its Knowledge Graph on May 16, 2012. It was meant to help Search understand “real-world entities and their relationships to one another: things, not strings” (Google’s Knowledge Graph announcement (external site)). Google’s example was a search for “Taj Mahal,” which could mean the monument, a musician or a casino. Now that Google can tell these entities apart, it can narrow the results to the one you mean.

According to Google, the facts come from content on the web and from open and licensed databases. Wikipedia is a commonly cited source, but just one of hundreds (Google on its Knowledge Graph and knowledge panels (external site)). We haven’t found a more recent total from Google than the one from May 2020.

Source: Google (external site),  · size of the Knowledge Graph

Google creates knowledge panels automatically when there’s enough information about an entity on the open web. Google doesn’t rule out errors: the scale of search is so large, it says, that “no system can be perfect.” That’s why anyone can report problems (Google’s help page on how the Knowledge Graph works (external site)).

The numbers in the Sources column refer to the source list at the end of this article.

Entity, Knowledge Graph, profiles and markup compared
TermWhat it isWho maintains itWhat you can doSources
Entitya uniquely identifiable thing, such as a company, a person or a place not specifieduse the same name and details everywhere1, 11, 20
Knowledge GraphGoogle’s database of facts about entities and their relationshipsGoogle, automatically, from content on the web and from open and licensed databasespublish clear details on your website and profiles5, 6
Knowledge panelinfo box in Google Search with facts from the Knowledge GraphGoogle; it’s created automatically when the web offers enough informationclaim it where possible and suggest changes, otherwise send feedback6, 7, 24
Google Business Profilea business’s listing for Google Search and Google Mapsyou, plus Google and users who suggest editscreate or claim it, keep it current, review Google’s updates7, 8, 19
Bing Places for Businessfree listing for Bing search and Bing Mapsyoucreate it or import your data from Google26
Organization structured datamachine-readable company details in your website’s source codeyouadd it to your home page or the page about your company21
sameAsreference in the markup to pages that describe the same entityyoulist only your own, maintained profiles4, 21
Wikidata itementry with a fixed identifier in a free knowledge basethe Wikidata communitywait until serious, publicly available sources describe you3, 34

What helps prevent mix-ups?

Google and Microsoft recommend using the same name and details everywhere, describing your company and its people with structured data on your website, and keeping your Google Business Profile and Bing Places listing up to date. You can often claim an existing knowledge panel. How much each step does for AI answers has hardly been studied.

  1. One name, the same everywhere. Settle on one spelling and use it on your website, your signage, your invoices and every profile. Step 3 of our how-to shows how to write a standard description and check your profiles against it: Are your company facts clear and consistent everywhere?
  2. Structured data for your company. According to Google’s documentation, structured data of the type Organization on your home page helps Google understand a company’s administrative details and tell it apart from others in search results. Some properties are only used behind the scenes for this, such as identifiers under ISO 6523 (Google on Organization structured data (external site)). For your entity, the key properties are name, legalName for a registered legal name that differs from name, alternateName for another common name, and url. Google says the URL helps it identify a company uniquely. Our article Structured data and schema markup: what helps, what doesn’t explains how to build and test this markup.
  3. Structured data for people. If one person is the face of your company, such as the owner, give them their own profile page and describe them with the type Person. For authors, Google recommends a link to a page that uniquely identifies the person. Google says it can understand both url and sameAs for this (Google on author markup (external site)). Our article What is E-E-A-T? explains what Google’s quality raters look at.
  4. Use sameAs only for real profiles. With sameAs, you list pages on other websites that describe your company. Google gives profiles on social media and review sites as examples. Schema.org defines the purpose more narrowly: the page should unambiguously indicate the item’s identity. So only list profiles that you own, that you maintain and whose details match your website. Otherwise, an abandoned profile with an old address would point to exactly the details you want to change. For a short description that sets an item apart from similar ones, Schema.org provides the property disambiguatingDescription (Schema.org on disambiguatingDescription (external site)). Google’s Organization documentation doesn’t mention it.
  5. Business listings on Google and Bing. Local businesses and businesses with a service area manage their presence in Google Search and Google Maps through a Business Profile, not a knowledge panel (Google on claiming knowledge panels (external site)). For its AI features, Google advises keeping your Business Profile information up to date (Google on AI features and your website (external site)). There should be only one profile per business. Otherwise, your information may not display properly in Maps and Search (Google Business Profile Help). Microsoft’s counterpart is Bing Places for Business. Listings there are free, appear in Bing search and Bing Maps, and can be imported from Google (Microsoft on the new Bing Places (external site)). Sign-up is now at bing.com/forbusiness.
  6. Claim your knowledge panel. If Google already shows a knowledge panel for your company or for you, you can claim it as its subject or an official representative (Google Knowledge Panel Help). You verify your identity by signing in to an official account, such as Search Console or YouTube. After that, you can suggest changes and back them up with links. Not every panel can be claimed. If yours can’t, you can still use the Feedback option.

What isn’t proven about entity SEO?

There’s no evidence that sameAs links, a Wikidata item or special markup get a company into AI answers. Google doesn’t require special markup for its AI features, and a study of 1,885 pages found that their citations barely changed after structured data was added. You also can’t apply for a knowledge panel.

Common assumptions about entities and what’s known, as of Nov 7, 2026
AssumptionWhat’s knownSources
Organization markup and sameAs get a company into AI answers.Google requires neither structured data nor special markup for its AI features. Ahrefs found that citations in AI answers barely changed after 1,885 pages added JSON-LD. The study analyzed all types together, including Organization. We haven’t found a study of sameAs on its own.10, 27
You can apply for a knowledge panel.Google creates knowledge panels automatically when the web offers enough information. You can only claim an existing panel, and not always.7, 24
A Wikidata item creates a knowledge panel.An essay by Wikidata editors states that Wikidata doesn’t control Google’s knowledge panels. It strongly discourages creating an item about yourself or your own company.28
Your own website and press releases are enough evidence for Wikidata.According to the same essay, neither your own websites nor social media pages, press releases or marketing materials are enough. What’s needed is independent coverage in reliable sources.28
A Wikipedia article about your own company is a good shortcut.Under its own rules, the German-language Wikipedia isn’t a business directory. Anyone writing there about their own company has a conflict of interest.29, 30

The Ahrefs study only covered pages that AI answers were already citing often. It leaves open whether structured data helps pages that don’t appear at all yet.

Source: Ahrefs (external site),  · vendor study · 1,885 pages, 4,000 control pages

We also haven’t found any research on sentences that companies put on their own websites to set themselves apart from businesses with the same name. Whether such a sentence makes sense is also a question of trademark law. Check it with a lawyer before you publish it.

What can you do if AI gets your company wrong?

First find out where the wrong information comes from, and correct it there: on your website, in your profiles and in directories. Then report the error through the system’s feedback option and check the answer again a few weeks later. The providers’ help pages don’t say how quickly a correction takes effect.

  1. Find the source. Open the sources an answer cites. OpenAI recommends this itself and warns: “Search results and citations can be incomplete, outdated, or incorrect.” If an answer cites no source, the information may come from what the model learned in training.
  2. Correct your own details. Start with your website, your legal notice (Impressum) and your structured data. Google also looks at your website when it decides on suggested edits to a Business Profile (Google Business Profile Help).
  3. Update profiles and directories. Update your Business Profile and review any changes Google has made itself. The profile editor shows them in blue (Google Business Profile Help). Change the same details in Bing Places, in industry directories and on review sites. If a Wikidata item about you already exists, an essay by Wikidata editors (external site) says you may correct clear factual errors. Request other changes on the item’s talk page, backed by reliable sources.
  4. Submit your changes. Notify Bing of new and changed pages so its index is updated sooner (Microsoft Bing). Our how-to on Bing and IndexNow shows how.
  5. Report the error to the provider. In an AI Overview, click the thumbs-down icon and then “Report a problem.” Google itself says that AI Overviews “can and will make mistakes” (Google’s help page on AI Overviews (external site)). Below a knowledge panel, you’ll find a “Feedback” link for this (Google Knowledge Panel Help). In ChatGPT, you can give feedback on a wrong answer with the thumbs-down button (OpenAI’s help article on ChatGPT (external site)).
  6. Check again later. Ask the same questions again after a few weeks, ideally on a fixed schedule. Our how-to on measuring AI visibility explains how to choose the questions and evaluate the runs, in the section Which questions belong in your question panel?

What’s specific to Germany and the EU?

A small business is usually too small for an article in the German-language Wikipedia. Its relevance criteria for companies include at least 1,000 full-time employees or more than €100 million in annual revenue. Identifiers that anyone can check are more useful, such as a VAT identification number. Google calls it an important trust signal.

Wikipedia: For companies, the German-language Wikipedia’s relevance criteria include at least 1,000 full-time employees, annual revenue of more than €100 million or shares traded on a regulated market. A company that meets none of the criteria needs other compelling arguments, because Wikipedia is explicitly not a business directory (Wikipedia’s relevance criteria, in German (external site)). If you write about your own company there, you have a conflict of interest. According to Wikipedia, intended promotion can backfire, because anyone can watch, criticize and edit its entries (Wikipedia on conflicts of interest, in German (external site)).

Source: Wikipedia (in German) (external site),  · relevance criteria for companies

The English-language Wikipedia sets no such thresholds. Under its notability guideline for companies (external site), a company is presumed notable if multiple reliable, independent secondary sources have covered it in depth.

VAT identification number: If you have one, you can add it to your structured data as vatID. Google calls it an important trust signal, because users can look up your business in public VAT registries (Google Search Central).

LEI: The Legal Entity Identifier (LEI) is a globally unique 20-character code for legal entities (GLEIF on the LEI (external site)). If your company has one, Google recommends the iso6523Code field with the prefix 0199 (Google Search Central).

Brand and legal name: Many small businesses operate under a brand that isn’t their legal name. In that case, state both together: in your legal notice, in your website’s footer and in your structured data as name and legalName.

From our work: how we describe Datafalke online and in our markup

Our brand is Datafalke. Legally, the business behind it is Mohammad Alboush – Alboush IT & Webdesign. Both names appear in fixed places on datafalke.de, and side by side in the legal notice, on the About page and in the structured data. The table shows what the pages that were online on November 8, 2026, said.

Where each name appears on datafalke.de, as of Nov 8, 2026
LocationWhat it says
Header of every pageDatafalke wordmark, linked to the home page
Footer of every pageMohammad Alboush – Alboush IT & Webdesign, with postal address and email address
Legal notice (Impressum)a sentence stating that Datafalke is a trading name of Mohammad Alboush
About page, company detailstrading name Datafalke, operated by Mohammad Alboush – Alboush IT & Webdesign
Structured data, type Organizationname Datafalke, legalName Mohammad Alboush – Alboush IT & Webdesign, plus a short description and our standard description
Owner’s profilea sentence stating that Mohammad Alboush is a software engineer and the owner of Datafalke in Bottrop
  • One source: Name, legal name, address, short description and standard description are stored in a single configuration file. Our pages and structured data read them from there. Where they appear as plain text, we check them against that file before publishing. You’ll find the full details in our legal notice.
  • Fixed identifiers: The company and its owner each have a fixed @id, the same on our German and English pages. The company points to the owner with founder, and his profile points back with worksFor. Every article names him as author.
  • Standard description in the markup: description contains our short description of about 25 words, and disambiguatingDescription contains our standard description: “Datafalke – websites, SaaS and business software from Bottrop, Germany. Operated by Mohammad Alboush – Alboush IT & Webdesign.” On our German pages, both are in German.
  • No sameAs yet: On November 9, 2026, none of the 98 pages that were online contained a sameAs reference in their structured data. We’ll only add profiles once we maintain them and link to them visibly on the website. We won’t add directories or review sites, even though Google names review sites as an example.
  • No Wikidata item: A Wikidata search for “Datafalke” returned no results on November 8, 2026. Under Wikidata’s rules, whether an item can be created depends on whether serious, publicly available sources describe us (Wikidata’s notability policy (external site)).
English home page of datafalke.de, Organization node, as of Nov 8, 2026 (excerpt, formatted)
{
  "@type": "Organization",
  "@id": "https://datafalke.de/#organization",
  "name": "Datafalke",
  "legalName": "Mohammad Alboush – Alboush IT & Webdesign",
  "url": "https://datafalke.de/",
  "disambiguatingDescription": "Datafalke – websites, SaaS and business software from Bottrop, Germany. Operated by Mohammad Alboush – Alboush IT & Webdesign.",
  "founder": { "@id": "https://datafalke.de/team/mohammad-alboush/#person" }
}
Owner’s English profile, Person node, as of Nov 8, 2026 (excerpt, formatted)
{
  "@type": "Person",
  "@id": "https://datafalke.de/team/mohammad-alboush/#person",
  "name": "Mohammad Alboush",
  "jobTitle": "Software Engineer",
  "worksFor": { "@id": "https://datafalke.de/#organization" },
  "url": "https://datafalke.de/en/team/mohammad-alboush/"
}

We don’t know yet whether search engines and AI assistants identify Datafalke correctly as a result. The website has been online since November 7, 2026, and there’s no measurement log with data yet. So we don’t claim any effect.

To check your own site, open the page source of your home page and search for Organization. Compare name, legalName, address and the entries in sameAs with your legal notice and your Business Profile. Fix every mismatch, starting with your website.

Sources

  1. Google LLC, Google Patents: Question answering using entity references in unstructured data (US 10,339,190 B2). . Accessed .
  2. Google for Developers: Method: entities.search (Knowledge Graph Search API). . Accessed .
  3. Wikidata: Wikidata:Introduction. . Accessed .
  4. Schema.org: sameAs. . Accessed .
  5. Google: Introducing the Knowledge Graph: things, not strings. . Accessed .
  6. Google: A reintroduction to our Knowledge Graph and knowledge panels. . Accessed .
  7. Google Knowledge Panel Help: How Google’s Knowledge Graph works. Accessed .
  8. Google Business Profile Help: Understand how Google sources & uses info in Business Profiles & local search results. Accessed .
  9. Google: AI Overviews and AI Mode in Search. . Accessed .
  10. Google Search Central: Optimizing your website for generative AI features on Google Search. . Accessed .
  11. Microsoft Bing: Bing Webmaster Guidelines. Accessed .
  12. OpenAI Help Center: Searching the web with ChatGPT. . Accessed .
  13. Anthropic, Claude Platform Docs: Models overview. Accessed .
  14. Chen, Gudipati, Longpre, Ling, Singh (ACL-IJCNLP 2021): Evaluating Entity Disambiguation and the Role of Popularity in Retrieval-Based NLP. . Accessed .
  15. Lee, Ye, Choi (preprint): AmbigDocs: Reasoning across Documents on Different Entities under the Same Name. , revised . Accessed .
  16. Mallen, Asai, Zhong, Das, Khashabi, Hajishirzi (ACL 2023): When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories. . Accessed .
  17. Kandpal, Deng, Roberts, Wallace, Raffel (ICML 2023): Large Language Models Struggle to Learn Long-Tail Knowledge. 2023. Accessed .
  18. Kalai, Nachum, Vempala, Zhang (OpenAI, Georgia Tech; preprint): Why Language Models Hallucinate. . Accessed .
  19. Google Business Profile Help: Understand Google updates on your Business Profile. Accessed .
  20. Google Business Profile Help: Guidelines for representing your business on Google. Accessed .
  21. Google Search Central: Organization (Organization) structured data. . Accessed .
  22. Google Search Central: Article (Article, NewsArticle, BlogPosting) structured data. . Accessed .
  23. Schema.org: disambiguatingDescription. . Accessed .
  24. Google Knowledge Panel Help: Get verified on Google. Accessed .
  25. Google Search Central: AI features and your website. . Accessed .
  26. Microsoft Bing: Introducing the New Bing Places for Business: Built for Business Owners, Powered by Research. . Accessed .
  27. Ahrefs: We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved. , updated . Accessed .
  28. Wikidata, essay: Wikidata:Self-promotion. . Accessed .
  29. Wikipedia: Wikipedia:Relevanzkriterien (in German). . Accessed .
  30. Wikipedia: Wikipedia:Interessenkonflikt (in German). . Accessed .
  31. Google Search Help: Find information in faster & easier ways with AI Overviews in Google Search. Accessed .
  32. OpenAI Help Center: What is ChatGPT? . Accessed .
  33. GLEIF: The Legal Entity Identifier (LEI). Accessed .
  34. Wikidata: Wikidata:Notability. . Accessed .
  35. Wikipedia: Wikipedia:Notability (organizations and companies). . Accessed .

Editorial note

Researched and drafted with AI assistance. Mohammad Alboush has not yet checked the sources, figures and claims.

Content not yet reviewed

Changes to this article

DateTypeChange
First publicationArticle published
CorrectionCopyedited references, punctuation and terms. The statements remain the same.
CorrectionStandardized the dates in the source list. No statements have changed.
UpdateUpdated the count to November 9, 2026 (98 pages)

Spotted a mistake? Email us at kontakt@datafalke.de. We’ll fix it and log the change.

Your next step

How visible are you in AI answers?

In the intro call, we’ll go through the questions your customers ask AI and look at where you stand today.

Menu

Book an intro callDeutsch