The same question to five AI engines, five different answers
If you measure visibility on one assistant, you are measuring one assistant. The spread between them is the real finding.
Take a single buying question, the kind a real customer would type, and put it to five assistants. Same wording, same day, signed out. You will not get the same answer five times.
You will get different companies named, different sources cited, and occasionally outright disagreement about factual details. I have run this enough times that the disagreement is the predictable part.
Why they diverge
Each assistant reads a different, overlapping set of sources, weights them differently, and assembles at a different moment. Some search live. Some lean on training data that is months old. Some have a browsing tool and use it inconsistently.
So the answers are not competing versions of one truth. They are five different processes producing five plausible outputs.
What the spread tells you
- If you appear in all five, your position is settled. You are in the shared part of every model's source set.
- If you appear in one or two, you are a fringe source. Visible in one lens, absent in the others, and most people are looking through a different lens.
- If a competitor is in all five and you are in none, that is a settled position you have not noticed, and it will not correct itself.
- If an assistant states something wrong about you, check which sources it cites. The error usually traces to one of them.
How to run this yourself
- Write ten questions a buyer would ask. Their problem, their words. Not your product name.
- Use at least four assistants. Signed out, so your history does not skew it.
- Log three things per answer: whether you are named, who is named instead, and which sources are cited.
- Repeat a week later. One reading is an anecdote. Two tells you whether it is stable.
- Build the citation list. Every domain that appears across the answers is your actual work list, and it is usually not your own site.
The part that surprises people
It is not that the answers differ. It is how confidently each one is stated. Every assistant gives a fluent, finished-sounding answer, with no signal that four other systems disagree with it.
The person asking has no way to know they are getting one possible answer rather than the answer. That is worth sitting with, and it is the reason measurement has to cover more than one engine.
What it does not tell you
These answers will not hold still. Models update, sources change, and a result from today is a snapshot rather than a standing. The useful output is the pattern across engines and across weeks, not any single reading, and a single reading is what most people are actually looking at.
ShowUp Labs checks your category across multiple assistants, not one.
ShowUp Labs →