I built a tool that checks whether AI cites your site. It had never cited mine

I maintain GEO Optimizer, an open-source CLI that audits whether ChatGPT, Perplexity, Claude and Google AI Overviews can crawl, understand and cite a website. The hosted version runs at geoready.dev.

I pointed it at itself, the way a customer would: ask the engines the questions a buyer asks, and count who gets cited.

108 calls to Claude with web search on, across Haiku and Sonnet, three runs per prompt. Prompts like "best GEO optimization tool" and "What are the best GEO and AI visibility optimization tools?". The same prompts had already gone through Perplexity.

geoready.dev: 0 citations, 0 mentions. On both engines.

Otterly, Peec and Profound came up again and again. Fair enough, they're good products. But zero for a tool whose whole job is this number deserved a postmortem.

The site itself passed. robots.txt allows every Anthropic, OpenAI and Perplexity crawler. llms.txt and llms-full.txt both return 200. The JSON-LD declares a Brand with alternateName and sameAs pointing to the GitHub repo and PyPI.

None of it mattered, because the engines never got to the site.

1. Retrieval. In 108 answers, geoready.dev never appeared among the search results Claude read, let alone the ones it cited. On Google it sits around positions 25–50 for "geo tools". The engines read roughly the first page. A page that isn't in the retrieved set can't be cited, however good its schema is.

2. The answer is someone else's listicle. The citations went to third-party "best GEO tools" articles. In my own tally of the 18 most cited pages, Otterly appears in 14 and my tool in none.

3. The awesome lists. On 26 October 2026 I checked eight GitHub lists on GEO, llms.txt and AI SEO. Otterly is in three of them. My repo, at 1.1k stars, is in none. I had submitted it to two of them and left it there: one pull request has been open since March, the other since last week, neither merged.

4. The name is taken, four times. At least four other GEO products are called GeoReady, on four other domains. Even when an engine saw "GeoReady", it had no way to know which one.

5. The two halves of the brand never met. This one surprised me most. "GEO Optimizer" was being named, in 10 answers out of 108, and the repo was cited 4 times. But the repo description, the PyPI summary and the README never said "GeoReady". And in this series I kept writing "the CLI I maintain" without naming it: one link to geoready.dev, in one post. The engines met two unrelated entities and credited neither.

6. Bing. I had never set up Bing Webmaster Tools for the site. Copilot retrieves from Bing, and so do other AI search products, so a site Bing doesn't index is missing from their results.

On-site GEO is the floor. Everything I audit for (crawler access, schema, llms.txt, passage structure) decides whether you can be cited once you're retrieved. Whether you get retrieved is decided mostly by other people's pages: listicles, lists, forums, posts that name you next to the category.

The audit I'd run on a client now starts with one question: which pages do the engines read for your category, and are you on them?

I'll rerun the same 108 calls in a few weeks and post the numbers, zero or not.

GEO Optimizer is MIT-licensed: github.com/Auriti-Labs/geo-optimizer-skill. The hosted version, GeoReady (geoready.dev), is at geoready.dev.

Method note: the 108 calls, the 10 mentions of GEO Optimizer, the 4 repo citations, the 18-page tally and the Google positions are my own measurements, written up on 26 October 2026. The count of eight lists was re-checked on that date.