GEO Strategy // Entity Disambiguation // MY & SG
Entity Confusion in AI Search: What Happened When AI Could Not Tell Our Brand Apart
By Gabriel Yee · SORIS Origin
TL;DR
Entity confusion happens when AI engines merge your brand with another business that shares the same name, and then answer questions about you with facts about them. A brand name recommended by Claude on 7 June 2026 shows how it plays out: once the website went live, AI engines pulled information from a same-name site on a .app domain and a same-name company in China, and Gemini kept returning them even when given the exact URL to fetch. A different domain extension did not separate the brands, so I dropped the name and rebuilt the firm as SORIS Origin. Test any brand name inside AI answers before you commit to it, not only on domain registrars.
What is entity confusion?
AI engines do not see brands as words. They see entities: a business with a name, a location, a category, people and a set of sources that describe it. When someone asks about a brand, the engine first has to decide which entity the name refers to.
Entity confusion is what happens when that decision goes wrong. Two businesses with the same or a very similar name get blended into one. In research, the underlying task is called entity disambiguation or entity linking. The engine may describe your services using another company’s details, link to the other company’s website, or simply ignore you because it believes the name already belongs to someone else.
For a new business in Malaysia or Singapore, this is a quiet but serious problem. Traditional SEO tools will not flag it. Your site can be fast, indexed and well structured, and AI engines can still hand your name to someone else.
Case record: a brand name that AI engines resolved to other entities
This is my own case. Before SORIS Origin, I ran the firm under a different name at apexgeo.my. The record below lists what I did at each stage and what happened.
| Stage | What I did | What happened |
|---|---|---|
| Naming | On 7 June 2026, I asked Claude to suggest names for a GEO agency. | Claude gave two options, SORIS Origin and the old name, and recommended the old name as the stronger choice. It gave no warning that the name was already in use elsewhere. |
| Registration | I registered the .my domain straight away and built the website. I did not check whether the name already belonged to another entity. | Not applicable. |
| Prompt testing | I ran prompts across AI engines to see how they described the brand. | Answers drew on a same-name website on a .app domain, or a same-name company in China, not on my site. |
| URL given to Gemini | I gave Gemini the exact .my URL and asked it to fetch information from it. | It repeatedly returned information about the .app site or the Chinese company instead. |
| Decision | I dropped the name, registered sorisorigin.my and rebuilt the firm as SORIS Origin, the other option from the original shortlist. | Competing for a name held by older, better-documented entities would have cost more than starting again. |
One other factor was at play. A new domain takes time to enter the search indexes that AI engines retrieve from, so part of the low retrieval was likely indexing lag. But indexing lag only explains why the site was missing. It does not explain why the engines answered confidently about other businesses, or why Gemini swapped in another entity even when it was handed the exact URL. That part is entity confusion.
This record describes only what AI answers showed during my testing. It makes no claim about the other businesses that use the name.
The AI that recommended the name never checked whether it was already taken in its own answers.
Why an AI recommendation is not a name check
When ChatGPT, Claude, Gemini or 豆包 judge a brand name, they judge the meaning: is it short, relevant, easy to say, suitable for the category. They do not check whether the name already points to another entity in their search results. A recommendation from an AI assistant is an opinion on how the name sounds, not due diligence on whether the name is free.
The question that matters is different. Not “is this a good name?” but “when someone asks you about this name, who do you think it is?”
Why a different domain extension does not separate you
It is natural to assume that .my and .app are different addresses, so AI engines will treat them as different businesses. In practice, people ask about brands by name, and engines resolve the name first. The domain extension is a weak signal compared with how many sources already describe the other entity under that name.
Research supports this. A study presented at ACL 2021, Evaluating Entity Disambiguation and the Role of Popularity in Retrieval-Based NLP, built test sets of entities that share a name and found that retrieval systems favour the more popular one. For questions about the less popular entity, the retrievers were twice as likely to fetch the wrong documents.
When two entities share a name, the less popular one is twice as likely to get the wrong documents
Source: Chen, Gudipati, Longpre, Ling and Singh, ACL 2021 (arXiv 2106.06830). Retrieval systems tested on AmbER entity sets for fact checking, slot filling and question answering. Relative figure as published by the authors, not measured by SORIS.
A new website is, by definition, the less popular entity. It has few mentions, few links and little history. When it shares a name with an older business, the engine’s retrieval step is already tilted against it before any content is read.
How to check a brand name for entity confusion before you commit
Run these checks before you register a domain, print anything or file a company name with SSM or ACRA:
- Ask the engines who the name is. In ChatGPT, Gemini, Perplexity, Google AI Overview and 豆包, ask a neutral question such as What is [name]? Use a new chat each time, logged out or in a temporary chat. If an answer describes another business, the name is already taken in AI search.
- Check every common extension. Look up the name on .com, .app, .ai, .io, .co, .cn and the local extensions (.my, .com.my, .sg, .com.sg). A live site on any of them is a competitor for the same entity.
- Check the Chinese web. Chinese engines such as 豆包 draw heavily on Chinese-language sources. Search the name and its likely Chinese rendering on Baidu, and check whether a mainland company already uses it.
- Check company registries and social handles. Search the Malaysian and Singaporean business registries, LinkedIn, Facebook, Instagram and Xiaohongshu for the exact name.
- Search with quotation marks. A Google search for the exact name in quotes shows how crowded the name already is.
- Decide honestly. If another entity already owns the AI answer for the name, choose another name, or a combination that no one else uses.
If you already have a conflated name
Renaming is not always necessary. If the other entity is small, inactive or in a clearly different category, you can strengthen the signals that separate you:
- Organization schema on the homepage. Google says some Organization properties are used “behind the scenes to disambiguate your organization from other organizations”. Fill in
name,legalName,url,address,logoandsameAslinks to your real profiles (Google Search Central). - A fixed descriptor everywhere. Pair the name with your category and location in every bio and listing, for example “[name], a GEO firm in Johor Bahru”. Repetition of the same pairing is what helps engines tell the two entities apart.
- Third-party mentions with the same pairing. Directories, media features and partner pages should use the same name, category and location, not just the name.
- Consistent details. Same business name, address, phone number and founder name on your website, Google Business Profile and social pages.
But if the other entity is older, more cited and in the same category, these fixes become a long fight for a name you will never fully own. In that situation, renaming early is usually cheaper than correcting AI answers for years. That was my reasoning behind the rename to SORIS Origin.
Frequently asked questions
What is entity confusion in AI search?
It is when an AI engine merges two different businesses that share a name and answers questions about one using information about the other. Researchers study the underlying task as entity disambiguation.
Does a different domain extension prevent entity confusion?
No. AI engines resolve the brand name first. A .my domain does not separate you from a same-name business on .app or .com if that business has more sources describing it.
Can I trust an AI to suggest a brand name?
For ideas, yes. For validation, no. AI assistants judge whether a name sounds good, not whether it already belongs to another entity in their answers. Test the name yourself across several engines.
How do I fix entity confusion without renaming?
Add complete Organization schema with sameAs links, use a fixed category and location descriptor with your name, keep business details consistent, and earn third-party mentions that repeat the same pairing.
When should a business rename instead?
When the other entity is older, better documented and in the same category. Early in a brand’s life, renaming costs less than fighting for the name.
Not sure whether AI engines can tell your brand apart? Request a free AI Visibility and GEO Readiness Audit. We will check how ChatGPT, Gemini, Perplexity and 豆包 identify your business, and whether any other entity is taking your answers.
Request my free auditSources
- Chen, A., Gudipati, P., Longpre, S., Ling, X. and Singh, S. (2021). Evaluating Entity Disambiguation and the Role of Popularity in Retrieval-Based NLP. ACL 2021.
- Google Search Central. Organization structured data. Checked 18 October 2026.
- Schema.org. sameAs.