How to check what ChatGPT says about you.
Five minutes, no tools, and you will know three things: whether you are named at all, whether the description is right, and who gets named instead of you.
A wrong description is repeated with confidence, every time.
Buyers ask an assistant before they email anyone. What it says about you is now part of your first impression, and nobody tells you when it is wrong.
- Are you in the answer at allAbsent is different from wrong. Absent means a sourcing problem; wrong means a facts problem.
- Is the description accurateName, what you sell, who you serve, where you are. One stale detail is enough to lose a qualified buyer.
- Who is named insteadThe answer is also a competitor list you did not have to pay for. Write the names down.
- Does it moveOne reading is an anecdote. The same reading next month is evidence.
Four steps, in this order.
1. Write five questions a real buyer would ask
Not "what is my company", which nobody asks. Real ones sound like: "Who does generative engine optimisation in Singapore?", "Which agency helps a B2B exporter get cited by ChatGPT?", "How much does GEO cost in Singapore?", "Is there a Singapore agency that measures AI visibility monthly?", "Who should I hire to fix wrong information about my company in AI answers?" Adjust to your trade, but keep them unflattering: the answer only matters if the question is one a stranger would ask.
2. Ask them on the engine your buyers use, one at a time
Start a fresh chat for each question. Do not add context, do not correct it, do not mention your own company. Ask in the language your buyers write in, and note the country setting if the tool exposes one.
3. Record the answer, word for word
Paste the whole reply into a document, with the date, the engine, the language and the exact question. Not a summary, not a screenshot of the good part. The wording is the measurement: if it called you a "web design studio" last month and a "GEO practice" this month, that change is the result.
4. Sort the result into one of four buckets
Absent, wrong, right but vague, or right and specific. Each has a different fix, and confusing them is why budgets get spent on the wrong thing. Absent needs third-party mentions. Wrong needs the source facts corrected. Vague needs a sharper one-sentence description. Specific needs nothing but repetition.
Four buckets, four different jobs.
- AbsentNobody outside your site writes about you. Directories, associations, partner pages. This is the slowest fix and the most decisive.
- WrongA source says something false and gets repeated. Find the source, correct it, then re-ask. Writing a new page will not fix it.
- Right but vagueYou are described in words too general to be chosen. Rewrite the one sentence you want repeated, identically, everywhere.
- Right and specificNow it is about frequency and freshness: keep publishing, keep re-asking, watch for drift.
Where the free version stops being enough.
Doing this yourself once is genuinely useful, and we would rather you did it than not. The free version stops working at the point where you need a record: the same questions, asked on four engines, in two languages, every month, kept in one place where you can see what changed. That is not difficult, it is just repetitive, and repetitive work is what gets dropped in a busy month.
Our free check is that first reading, done for you, with the questions and the raw answers sent to you whether or not you hire us. If you would rather run it yourself, the four steps above are the whole method.
Four straight answers.
Does asking the same question twice give the same answer?
Often not. Runs differ, accounts differ, and the tool may use different context. That is exactly why one screenshot proves nothing and why the record matters. Ask each question at least twice before drawing a conclusion.
Should I ask about my own company by name?
Yes, separately. "What does your company do?" tells you whether the facts are right. The buyer questions tell you whether you exist in the shortlist. They are two different measurements and both are worth having.
Does a wrong answer mean someone is sabotaging me?
Almost never. It usually means an old directory entry, a subsidiary name, or a press release that used different wording. The model repeated the most consistent version it found. Fix the source, not the model.
How often should I re-check?
Monthly is the useful rhythm: frequent enough to catch a change, slow enough that the change means something. Weekly is noise unless you are actively publishing into the same question set.
Find out what AI says about you.
Free, with no obligation. You keep the report either way.