Zero to 62,000+ AI citations
Reverse-engineered how AI engines pick sources, then rebuilt the content roadmap around citation gaps rather than search volume.
Fintech ยท Fundraising platform
The situation
Generative answers were starting to absorb the top of the funnel. Users were asking ChatGPT and Perplexity the questions that used to arrive as search queries, reading the synthesised answer, and never clicking. The site had solid classic-SEO foundations and effectively no presence inside those answers.
The problem
Nobody could say why an AI engine cited one source and ignored another. Without that, any AEO work would be guesswork dressed up as strategy, and the obvious instinct, to keep publishing against high-volume keywords, was optimising for a surface that was shrinking.
What I
actually did.
In order, with the reasoning behind each step.
Built a prompt corpus, not a keyword list
I assembled the questions a real buyer would actually type into an assistant: not keywords, but full natural-language prompts across the whole consideration journey. That corpus became the measurement baseline.
Ran the corpus across every engine and logged the sources
Each prompt went through ChatGPT, Perplexity, Gemini and Google AI Overviews. I recorded which domains were cited, in what position, and what kind of page it was. Patterns showed up fast: listicles, comparison pages and review platforms carried far more citation weight than brand-owned thought leadership.
Prioritised by citation gap instead of volume
The roadmap was re-sorted by where competitors were being cited and we were not: a gap analysis on answers rather than rankings. Low-volume prompts with high citation frequency beat high-volume keywords that never produced a citation.
Made the content machine-extractable
Entity and schema optimisation so the engines could resolve who we were and what we covered; content structured so a specific claim could be lifted cleanly out of a page; markdown twins and llms.txt so crawlers reading for synthesis got a clean version rather than a rendered one.
Earned placement in the sources AI already trusted
Rather than only publishing on-domain, I went after inclusion in the third-party listicles, comparison pages and review platforms the engines were demonstrably pulling from. Citation-earning link acquisition, judged on whether a source appeared in answers, not on domain rating.
What I took
from it.
AEO is not a new content format, it is a different unit of measurement. Once the team stopped asking "where do we rank" and started asking "who gets cited and why", the roadmap reordered itself, and the third-party placements turned out to matter more than anything we could publish on our own domain.
Want the version with names and dashboards?
Employer names, client names and the underlying reporting go out with applications, or on request.