Google ranking: RESONEO maps 305 signal groups in its code

Abondance reports that RESONEO has published a count of 305 groups of ranking signals it says it found in Google's code, each group registered in a system called RankLab. The breakdown given is 167 groups tied to each result and 138 tied to each query, with 48 of the former concerning a web page and 31 used to score it. Generative AI is mentioned as a recent addition, though the report does not specify what that means in practice. Nothing in the announcement indicates whether these groups carry published weights, whether they are testable, or how RESONEO separated a signal from ordinary code. Those gaps matter more than the headline number.

What the count does and does not tell you

A taxonomy of 305 groups is an organisational artefact, not a ranking formula. Knowing that 31 groups contribute to scoring a page does not reveal how any of them interact, how they are thresholded, or how they behave across languages and verticals. Ranking systems compile from many features, and the same feature can matter enormously on one query type and barely at all on another. The figures also exclude the part SEOs actually observe: click patterns, freshness, intent classification and spam systems that are not cleanly attributable to a single group.

The useful reading is structural. The split between per-result and per-query signals mirrors something most practitioners already assume: some evaluation is document-level and reusable, some is computed fresh for the request. If that framing holds, it argues for investing in page-level fundamentals that survive across queries rather than chasing per-query tactics you cannot verify.

Three checks to run this week

• Take your five most commercially important URLs and test whether their core value proposition is legible without rendering JavaScript, since page-level signals plausibly draw on more than the rendered DOM.
• Compare titles and snippets across queries where you rank, to see whether the same page is being re-framed per request — evidence your per-query signals are being interpreted as intended.
• Log which pages depend on one query family for most of their clicks, and treat that concentration as a risk rather than a win.

The generative AI line to watch

The mention of generative AI is the part most worth following, because AI-surfaced results change the unit of competition. If answers are assembled rather than linked, being one of the retrieved sources matters more than position on a results page, and the signals that decide retrieval may not overlap neatly with classic ranking factors. The announcement does not say which of the 305 groups relate to AI surfaces, so treat that as unspecified. What you can do regardless is ensure your pages carry extractable, attributable statements rather than only narrative prose, since that is the format that survives being quoted. None of this replaces measurement on your own traffic; a list of signals is a hypothesis generator, not a playbook.