Google AI Overviews and AI Mode do not pick pages the way the ten blue links did. They fan a question out into sub-questions, fetch candidates for each, and assemble an answer from the pages that cover the most ground credibly. The implication for a writer is simple and uncomfortable: a thin page cannot win, however well it is optimised.
What “depth” means in practice
Take the question your page answers and write down the five questions a careful reader would ask next. If your page answers three of them in passing and ignores two, you have found your edit. This is the “query fan-out” idea, and it is the single most useful lens for GEO work.
Signals that read as trust
- Outbound citations to primary sources, with real links.
- Original numbers: your own measurements, survey, or worked example.
- A table where a comparison is being made.
- Semantic HTML: the article in an
<article>, navigation in<nav>, so the model can discard the chrome. - A clear “Updated on” date that matches
dateModified.
What the GEO score checks
Rankavia’s generative-engine score looks for exactly these things on the rendered page: sub-topic coverage against the heading structure, the presence of citations, tables and original data, semantic wrappers, and freshness. It then lists the specific gaps for the page in front of you. It will not write the missing section for you unless you ask it to draft one with your own AI key, and even then you approve every word.