Write when the model's answer turns foggy
In a frequent scene, a model names you correctly, then pulls in an old address, confuses your work with a neighbor's, or places your firm in a category that is too broad. For a first message, send a few queries a client might ask, a link to the site and, if possible, an answer that seemed off to you. The rest can wait for the diagnostic.
Frequently asked questions
How do you usually work?
I begin with a short description of the problem and a few queries that a client might genuinely ask. Then I read the model answers, the company's site, and the neighboring categories. The result takes the form of a map: solid points, zones that fray, sources to revise.
What subjects do you work on?
I work with service companies, especially in B2B, in technical, educational, consulting, and regulated professional fields. A subject fits this work when the name, city, specialization, and evidence need to remain precise. If the niche is regulated, I begin with the acceptable limits of communication.
How quickly do you reply?
I reply after reading the context, not with an automatic acknowledgement pretending to understand. When the message is dense, I answer first with what is clear and the few questions needed to avoid a diagnosis that goes off course.
What format does the consultation take?
Most often, it is a written review, with sample answers, page-by-page notes, and short recommendations. A working session can follow when formulations need to be revised together. I prefer a clear document to a theatrical presentation.
Roughly how much does it cost?
I give a range after the first message, because the volume depends on the queries, languages, pages, and neighboring categories. A small entity reading and a full answer map are not the same amount of work. An honest estimate is better than a price hidden behind a pretty sentence.
What requests do you leave aside?
I do not guarantee top positions in AI-system answers. I leave aside fake reviews, invented biographies, decorative client logos, and texts written to trick the model. If the problem is not source readability, I say so.
Start with an answer that made you look up from the screen.
A small model error often shows which source is old, weak, or too close to a neighboring category.
Describe the case