If you've read anything about "getting cited by AI" in the last year, the advice was probably some combination of: add structured data, write clearer copy, fix your headings. All reasonable. We build a tool that does exactly that kind of analysis, so I'm biased toward it being useful.
Then we started logging where the AI answers were actually pulling from — the source types cited when an assistant answers a buying question in a given category — and the data made me rethink the priority order.
Company homepages were rarely the primary source.
What the citation mix actually looks like
When you ask ChatGPT, Perplexity, Gemini, or Claude something like "best project management tool for a small agency," the model (usually with web search on) assembles an answer from retrieved documents. Log what those documents are across a category and a pattern shows up fast. Roughly:
