It can, and a single-question test will not show it. Buyers ask a broad question, then narrow it: by size, by region, by budget, by a feature they need. Each follow-up changes what the assistant searches for and which companies fit. A company absent from the first answer can be named once the conversation reaches its niche, and a company named first can drop out.
Most AI visibility checks, including the simple ones we describe in other notes, ask one question and record one answer. That is a good start. It misses the moment a lot of shortlists are actually made, which is two or three turns in, when the buyer has said what they really need.
This note shows how to script a short buyer conversation, run it the same way each time, and read where you enter and where you drop out.
How do buyers actually use an assistant?
In a conversation that narrows.
A buyer rarely knows the exact question at the start. They begin broad, read the answer, and push back. “Which of those work for a team our size?” “Which have customers in Europe?” “Which ones integrate with what we already use?” Each turn adds a constraint.
Each constraint changes the answer. With search on, the assistant may search again with the new words and read different pages. From memory, it filters what it knows against the new detail. Either way the list moves.
That movement is where specialists win. A company that is not the obvious name in a broad category can be the obvious name for a narrow need, if its pages say so plainly.
A company that is not the obvious name in a broad category can be the obvious name for a narrow need, if its pages say so plainly.
How do you script a test conversation?
Write the turns before you start, and never improvise.
Build three to five conversations of three to five turns each.
- Turn one is a broad buyer question with no company names.
- Turn two narrows by the buyer’s situation: size, stage, industry or region.
- Turn three narrows by a need you meet well: a feature, an integration, a standard, a language.
- Turn four asks for a comparison: “Which of these would you pick, and why?”
- An optional turn five asks the follow-up a careful buyer asks: “What do customers say about them?”
Write every turn down word for word. Do not react to the answer with a different follow-up than the script says. The test only works if every run asks the same things in the same order.
Base the turns on what real buyers ask. Sales calls are the best source. The questions prospects ask your team on a first call are a good guide to the follow-ups they ask an assistant.
How do you run it and read the results?
Fresh chat, same script, one row per turn.
Run each conversation in a fresh chat with personalisation off, so earlier chats do not shape the answers. OpenAI says memory and custom instructions can personalise ChatGPT’s answers, so turn them off for the test. Do the same in Claude and Gemini.
| Turn | Named | Are we named? | Position | Cited pages |
|---|---|---|---|---|
| 1 broad | Firm A, Firm B, Firm C | No | None | Two roundups, one review site |
| 2 our size | Firm B, Firm C, Firm D | No | None | One roundup, one forum thread |
| 3 our need | Firm B, us | Yes | 2nd | Our integration page, a partner listing |
| 4 compare | Firm B, us | Yes | 2nd | Same as turn 3 |
The row where you first appear tells you which constraint brings you into the answer. The cited pages at that row tell you which pages carried you in. If you appear at turn three and drop out at turn four, the comparison found something about you, or did not find enough. Read the answer’s reasons.
Run each conversation on two different days before drawing conclusions. Answers vary between runs, and a turn where you enter on both days is a pattern. A turn where you enter once is a sample.
How do you get named earlier in the conversation?
Answer the narrow questions on your own pages, in the buyer’s words.
If you enter at turn three on a specific need, that need is your strongest card. Give it its own page that says plainly who it is for and what it does, so the assistant can find it when the buyer asks the narrower question.
To enter earlier, you need to be on the pages the assistant reads for the broad question. That is roundups, review sites and community threads in your category. Our note How to get your business recommended by ChatGPT covers how to earn a place on each.
To avoid dropping out at the comparison, make sure the facts a comparison draws on are right and easy to find: who you serve, how pricing works, which customers and integrations you have, what reviewers say. A comparison turn rewards the company whose facts are clearest.
Buyers do not ask one question. They have a conversation. Test the conversation.
See the check page for ChatGPT: Does ChatGPT name you when buyers ask?
Questions people ask next
How do I test if ChatGPT mentions my brand in a longer conversation?
Script three to five buyer conversations of three to five turns each, from a broad question to a narrow need and a comparison. Run each in a fresh chat with personalisation off and record, turn by turn, who is named and which pages are cited.
Why is my company named only after a follow-up question?
The follow-up adds a constraint, such as size, region or a specific need, that matches your pages better than the broad question did. That constraint is a strength worth its own page.
Does ChatGPT remember earlier questions in a conversation?
Within the same chat, it uses the earlier turns. With memory or custom instructions turned on, its answers can also be personalised by what it knows from you, which is why tests should use a fresh chat with personalisation off.
How many turns should a test conversation have?
Three to five is enough. It covers a broad question, one or two narrowing follow-ups and a comparison.
Can Throughline test buyer conversations for us?
Request an AI visibility check. A growth strategist reviews how AI assistants see your company and replies by email within two business days.