We tested our own GEO tool on ourselves: 0 of 81 AI answers named us
We build LazyGEO, a GEO (generative engine optimization) tool: it checks whether AI assistants name your product and works to make them do it. So the first product we pointed it at was our own. This post shows the real result, how we got it, and what we changed. We will add later weeks to this page as they come in.
What did we find?
In week 3 of LazyGEO, we asked ChatGPT, Perplexity and Claude 27 questions our customers ask before they choose a tool. That made 81 answers. None of the 81 named LazyGEO. The answers named other tools, or no tool at all.
The full numbers from that week:
| What we measured | Result |
|---|---|
| Customer questions asked | 27 |
| AI assistants asked | 3: ChatGPT, Perplexity, Claude |
| Answers collected | 81 (27 questions × 3 assistants) |
| Answers that named LazyGEO | 0 |
| Times AI crawlers read our pages, last 7 days | 32 |
| Clicks from Google, last 28 days | 11 |
Gemini was not part of this first check. These are the only numbers in this post; we have not rounded or estimated anything.
Why did no AI answer name us?
Because almost nothing outside our own site talked about us yet. AI assistants answer from the pages they find when they search: “best X” lists, reviews, articles and comparisons. In week 3, LazyGEO was on almost none of them, and our own pages were new and barely indexed outside Google.
That matches what published research says. In an Ahrefs study of ChatGPT recommendation answers, “best X” blog lists were 43.8% of the cited page types (Ahrefs), and Muck Rack found about 84% of AI citations came from earned media (Muck Rack). A three-week-old product is on neither.
The crawler number tells the same story from the other side. AI crawlers did read our site, 32 times in 7 days. Being read is not the same as being recommended. Our earlier ChatGPT test saw the same gap on another product.
What did we change?
We made six changes, all of them things LazyGEO now does for its customers: moved our comparison pages into a subfolder, wrote fair “best X” list pages, started writing to list authors, checked Bing and Brave indexing, asked key questions three times, and moved the work into one team chat.
- A subfolder instead of a subdomain. Our comparison pages moved from a separate subdomain to lazygeo.com/answers/, so they sit on our main domain and the links count for it.
- “Best X” list pages. Since lists are what AI quotes most, we write fair list and comparison pages, with facts read from each tool’s own site.
- Writing to list authors. We write to the people who run the lists AI already quotes and ask to be considered. Nothing goes out in our name, or yours, without a 24-hour window and a Stop button.
- Bing and Brave indexing. ChatGPT search leans on Bing and Claude on Brave (Profound), so we check our pages there, not only in Google.
- Asking key questions three times. AI answers change from run to run; SparkToro found the same brand list came back less than 1 in 100 times (Search Engine Journal). The questions closest to a purchase are now asked 3 times each.
- One team chat. Results, pages and emails now arrive in one chat, where every page and email waits for an OK before it goes out.
How did we run the check?
We wrote down 27 questions a founder asks before choosing a tool like ours. We sent each one to ChatGPT, Perplexity and Claude through their official APIs with web search turned on, and kept every answer word for word. An answer counted only if it named LazyGEO.
Details that matter when you read the numbers:
- APIs, not apps. We used each assistant’s official API with web search on. Answers in the ChatGPT, Perplexity and Claude apps can differ from what the APIs return.
- One run per question in week 3. Each question went to each assistant once, which gives 27 × 3 = 81 answers. Since then, the questions closest to a purchase are asked 3 times each.
- Crawler visits are AI crawler requests to lazygeo.com, counted by the site’s own crawler tracking over 7 days. Google clicks are clicks from Google Search over 28 days.
- No sampling or rounding. Every number above is a direct count.
This is the same method LazyGEO uses for customers, and it has the same limits: it is a sample of answers at one point in time, not a guarantee of what any one person will see.
What happens next?
We keep running the same check every week and will add the results here, with the date of each update at the top of the page. If the count stays at zero, we will say so. Nobody can promise when an AI will name a product, including for our own.
Updates
- 19 October 2026: first version, with the week 3 numbers.
If you want to run the same check on your own product, the free check shows what AI can read on your site today. For the background, read What is GEO (generative engine optimization)? and how to get recommended by Perplexity, Claude and Gemini.