We built a brand from scratch to test AI visibility. 7 months later: 42,000 AI citations, 100 website users
Our team wanted to see what AI visibility looks like for a brand with no history, so they built one.
In February we created Civoryx com, a Global Fraud Index based on search interest in fraud keywords. Real site, zero prior footprint - no reviews, no mentions, nobody searching the name.
We published 93 pages (the brand site plus content on 11 external sites), then tracked 165 prompts daily across ChatGPT, Perplexity, Gemini, AI Overviews and AI Mode. 75 named the brand, 90 didn't. Publishing stopped March 31 and we left it all untouched until September 23.
The part we didn't expect to be writing about
Citations of the brand site went from 17,023 in April to 42,435 in August. 149% growth, after we'd stopped publishing.
Over the same period GA4 recorded 100 users total. Google Search showed 11 clicks. Server logs found exactly one visit referred by an AI assistant.
Citations in tracked answers are not people reached. Worth knowing before anyone reports citation counts to a client as if they're traffic.
What else came out of it
- Branded visibility doesn't transfer. 84.8% of branded answers mentioned us. 1.8% of non-branded ones did.
- Third-party pages did the work. External sites got ~19,000 non-branded citations, the brand site under 2,000. With an external article in the sources, 16% of answers named the brand. Without, 0.29%.
- Being cited isn't being mentioned. Our listicles were cited in 4,399 non-branded answers but only 14.2% actually named the brand. The one in-depth guide was cited far less, yet 69.2% mentioned it.
- Engines aren't one channel. September non-branded mentions: Perplexity 30.6%, ChatGPT 8.3%, AI Overviews and Gemini 0.6%, AI Mode 0%.
- Time to first citation varied wildly. Short articles 21 days median, longer how-to guides 126 days. A quiet month doesn't mean a page failed.
The 11 external sites were ours, so that's a controlled test rather than earned coverage. And ChatGPT changed its citation behaviour twice mid-experiment - Aug 9 much longer source lists, Aug 28 cut from ~17 per answer to ~5 - so part of our August movement was the engine, not us.
Full write-up with the tables is on the SE Visible blog. Happy to get into the method here!