Ask an answer engine about your company and sometimes it answers a different question: which of several things you might mean. That hesitation is entity fog. The engine has more than one candidate for your name and no clear reason to pick yours.
It is easy to miss, because your own branded searches look fine. It shows up in the places you do not check: the opening line of an AI answer, the comparison a buyer asks for, the summary an assistant writes when someone pastes your name.
What does entity fog look like in a real answer?
Here is ours. On 2026-10-03 we read the question "what is Attensira" on Perplexity three times, as part of the brand questions our own workspace tracks every day.
Two of the three answers opened with: "Attensira can refer to two different things, depending on context." One was our software. The other was an ADHD medication sold under a similar name in some countries. The answers cited Cleveland Clinic, Mayo Clinic and drugs.com pages about the medicine, ahead of our own About page.
ChatGPT handled the same question better: all three of its answers described the software and cited our pages. Two of them still closed by asking whether the reader meant a medication instead.
None of this is false, exactly. It is fog. A buyer who heard about us from a colleague and asks an assistant gets half an answer about a pill.
Entity fog comes in three shapes:
Collision
- What the answer does
- Describes another entity with a similar name
- Typical cause
- A product, drug, person or place shares the name
Split
- What the answer does
- Treats one company as two or three
- Typical cause
- "Acme Flow", "AcmeFlow" and "Flow by Acme" all in use
Merge
- What the answer does
- Attaches another company's facts to you
- Typical cause
- Your profiles link to the wrong page, or a listing mixes two vendors
Our case is a collision. Splits and merges are more common for companies that renamed, were acquired, or sell several products under one brand.
Why do answer engines get confused about names?
An answer engine has to decide which entity a word refers to before it can say anything about it. It decides from context: the words around the name, the pages that mention it, the profiles that link to each other.
When the context is thin, the most documented entity wins. A medicine with clinical reference pages has more documentation than a young software company. A company with three spellings has its mentions spread across three weak candidates instead of one strong one.
Google has a formal version of this: Organization structured data, which it says can help it "disambiguate your organization in search results", with the sameAs property pointing to your official profiles elsewhere. Answer engines read more loosely, from text, but the principle carries over. The more places agree on one name, one description and one set of profiles, the less room there is for the wrong candidate.
There is a measurement trap here too. If your brand name is an ordinary word, a tool that counts mentions will count every unrelated use of the word. Our own share of voice documentation says it directly: brands whose names are common words read high, never low. Read the answers, not just the count.
How do you find out if you have entity fog?
Run the identity questions, not the category ones:
- "What is [brand]?"
- "Who makes [product]?" and "Is [product] made by [company]?"
- "[brand] vs [main competitor]"
- "[brand] pricing"
- "[brand] reviews"
Ask each on every engine you care about, at least three times. For every answer, note three things: did it describe the right entity, did it hedge between two, and did it attach anyone else's facts to you.
Then read the citations. A collision usually shows up as citations to pages about the other entity. A split shows up as different spellings in different answers, or in the same answer. A merge shows up as a fact you recognise from someone else's product.
Write down the exact spellings you find. That list is your work queue.