How Do B2B Companies Monitor Their Presence in ChatGPT and Perplexity?
September 10, 2026
The method is the same everywhere it is done well: a fixed set of buyer questions, run repeatedly against each assistant, with the results and cited sources recorded over time. Everything a tool adds is a convenience on top of that. Understanding the method is what lets you judge whether a tool is doing it properly.
Table of Contents
- Step 1: build the prompt set
- Step 2: run each prompt more than once
- Step 3: record sources, not just mentions
- Step 4: rescan after every change
- ChatGPT and Perplexity behave differently
- FAQ
Step 1: build the prompt set
Twenty to forty questions is enough for most categories. Split them across three stages:
- Discovery: how do I solve this problem, what are the options.
- Evaluation: best tools for X, which vendors should a company like mine shortlist.
- Decision: which provider should we pick, does vendor Y do Z.
Then freeze the list. A prompt set that changes every month cannot produce a trend.
Step 2: run each prompt more than once
Assistant output varies run to run. Three runs per assistant per prompt is the practical minimum to get a rate instead of a coin flip. Report "named in seven of nine responses", never "yes".
Step 3: record sources, not just mentions
The cited domains are the most actionable output of the whole exercise. They tell you exactly which pages are shaping the answer in your category, which converts a vague content plan into a specific target list.
Step 4: rescan after every change
Publish something, then rescan the same prompts. Without the second reading, you cannot separate your work from ordinary answer drift. This is the step that turns AI visibility from a vibe into evidence.
ChatGPT and Perplexity behave differently
Perplexity is retrieval-heavy and cites nearly everything, which makes it the best assistant for learning which sources matter in your category. ChatGPT mixes retrieval with what it already absorbed, so it can name companies without citing a page, and it carries the most buyer traffic in most B2B categories. Track both, and expect them to disagree; the disagreement is information.
Claude and Gemini round out coverage and are worth including, since a shortlist you miss in one assistant you may still make in another.
Doing it without a spreadsheet
Forty prompts, three runs, four assistants is 480 responses per cycle. That is where teams either automate or quietly stop. If you want the automated version, see how Monroya works, compare the field in the AI visibility tools comparison, or just get a baseline with the free check.
FAQ
How often should we scan? Weekly while you are actively publishing, monthly when you are not. Consistency matters more than frequency.
Should we track competitors on the same prompts? Yes. Absence is only half the picture; knowing who took the slot tells you what the model treats as a credible answer.
Is this the same as tracking Google rankings? No. It is a separate signal with separate mechanics, which is why it needs separate measurement.