What the models say when you are not in the room.
An AI answer does not rank alongside ten other results. It replaces them. One synthesised paragraph, read as fact, and the reader usually stops there. These are the twelve prompts that show you the paragraph.
Assembled locally. Nothing is sent to us, and no AI is called from this page.
Direct identity
-
Who is [name]? -
Tell me everything you know about [name]. -
What is [name] best known for?
Diligence framing
-
Should I do business with [name]? What are the risks? -
Has [name] ever been involved in any controversy, lawsuit, or investigation? -
What do critics say about [name]?
Buying intent
-
Is [name] legitimate or a scam? -
What are the best alternatives to [name]? -
What are the most common complaints about [name]?
Source probing
-
What sources are you using to answer questions about [name]? List the URLs. -
How confident are you about what you just said about [name], and what could be out of date? -
Is there another person or company also called [name]? How do you tell them apart?
Four things to record for every answer
- Factually wrong. A date, a role, a company, an outcome the model got flatly incorrect. This is the highest-priority category and the most fixable, because it usually traces to one bad source.
- Conflated. The model has merged you with someone who shares your name. Common, damaging, and invisible until you ask the disambiguation prompt.
- Stale. A resolved matter presented as ongoing. A settled case, a former role, a closed investigation. Technically true once, false now.
- Cited sources. Which URLs the model names when asked. This is the actual target list. Nothing about the answer changes until something about those pages does.
Run each prompt three times per model across a few days. These systems are non-deterministic and many retrieve live, so one clean answer proves nothing.
Monitoring is the cheap half. Changing the answer is the other half.
Every AI visibility tool on the market will tell you the model said something wrong. Almost none of them will change it, because changing it means working the source layer: correcting the pages the model cites, outranking the ones that cannot be corrected, and getting accurate material into the places these systems retrieve from. Then re-measuring until the answer moves.
AI reputation questions
Why does what AI says about me matter more than my search results?
Because an AI answer replaces the result page rather than ranking inside it. A search result gives the reader ten options and lets them judge. An assistant gives one synthesised paragraph and the reader usually stops there. A wrong sentence in that paragraph does the damage that a page-one link only threatens to do.
The model said something false about me. Can it be corrected?
Not by editing the model. Answers are assembled from sources the system trusts and retrieves, so the lever is the source layer: correcting or outranking the pages the model cites, and getting accurate material into the places it retrieves from. That is slower than a takedown and more durable than one.
Why run the same prompt more than once?
These systems are non-deterministic and many now retrieve live. The same question can produce a clean answer and a damaging one twenty minutes apart. A single run is an anecdote. Three runs per model across a few days is the smallest sample that tells you anything real.
Do I need to be famous for this to matter?
No, and the smaller your public footprint the worse it usually reads. With thin source material a model leans harder on whatever it can find, which is often one hostile forum thread or a same-name confusion with someone else entirely. Sparse coverage is a risk factor, not a protection.
What is a bad result worth fixing versus ignoring?
Fix anything factually false, anything that conflates you with another person, and anything that surfaces a resolved matter as current. Ignore tone. A model describing you as "controversial" is reflecting coverage that exists, and the fix for that is the coverage, not the model.
The standing engagement opens with a private call.
A single conversation, signed under non-disclosure, with the principal who would own your matter. You leave with a printed posture assessment and the engagement letter, whether or not you retain us.