Ask ChatGPT "what's the best tool for X" and you'll get three or four brand names. If yours isn't one of them, you have a problem Google Analytics will never show you: the user never reached your site, so there's no traffic to attribute.
I've been chasing this for my own tool. Here's the checklist I use, the data I got, and the fixes that actually moved the needle.
LLMs don't "browse" a category. They answer from:
Most indie products fail on #2 and #3, not #1. The model doesn't know you because nothing on the open web says the same, boring sentence about you ten times.
Pick 4 prompt families, and 3 variants each — 12 prompts total:
| Family | Example |
|---|---|
| Category | "best AI visibility tools", "tools to check if ChatGPT mentions my brand" |
| Problem | "how do I know if AI search engines recommend my product" |
| Comparison | "brand monitoring vs AI visibility trackers" |
| Brand | "what is bidvix" (baseline: does the model know you at all?) |
Run each prompt 3 times in ChatGPT, Gemini, Claude, Perplexity, and DeepSeek. Score:
Track it in a sheet. Repeat weekly. A single check is noise; a trend is a signal.
For bidvix, brand prompts were fine — the model knew the name. Category prompts were the problem: my tool was absent from "best AI visibility tools" answers, while the cited sources were almost entirely directory listings and listicles, not vendor sites.
Two conclusions that changed my plan:
X is a [category] that [does the thing] for [who]. Use it verbatim on your site, your profiles, and every directory submission. Don't get creative — get repetitive.
You can't A/B test your way into this. There's no dashboard that tells you "you were cited 4 times today" unless you build or buy one — which is, admittedly, why I built mine. If you want to skip the spreadsheet, bidvix runs the same 12-prompt check across five models and gives you a mention rate plus the cited URLs.
But the audit above works with a notebook and 20 minutes. Do that first — the numbers will tell you whether you have a content problem or an entity problem, and those have very different fixes.
If you run it, I'd genuinely like to see your results — especially the cited-URLs column. That list is usually a shock.