The inclusion threshold is a consequence of how AI models learn. Two mechanisms play a crucial role: training bias and frequency heuristics.
Training bias
An AI model learns from the data it is trained on. If that data reflects a particular market structure — in which large, established players are dominant — the model will reproduce that market structure. That is not a conscious choice by the model builder, but a property of how machine learning works.
Concretely, and as a hypothesis: a player who appears everywhere in the context of their category presumably has an edge over a player who is barely mentioned — a fixed ratio or automatism it is not, and nobody can inspect the training data to count it. What we do measure, across sectors: AI answers strikingly often reflect the established market structure. External research into brand preference in language models points the same way.
Frequency heuristics
Alongside training bias, frequency heuristics play a role: models reproduce patterns from their data, and what frequently occurred together returns more easily. That plausibly makes the most-mentioned brands also the most likely candidates for an answer — the precise workings per model are not published.
Implication: GEO is not just a content optimisation problem. It is also a presence problem. You simply have to be mentioned more often and more consistently in the right contexts — across the entire web.
Mental availability in AI models
The concept of “mental availability” from brand science — the degree to which a brand comes to mind spontaneously for a category — lends itself well as an analogy for what we measure here. In our measurements we see broadly present brands return more often in answers; whether that is also more extensive and more positive differs per brand and per channel — an automatism it is not.
The strategic lesson of the analogy stands: broad, consistent presence in the right categorical contexts, over a longer period. But it is a compass, not a cognitive equivalent.
What can you do?
- Invest in long-term, broad external presence — not just on your own website
- Ensure your brand is consistently mentioned in the right categorical context (PR, partnerships, directories)
- Build Wikipedia and Wikidata presence where attainable — a strong entity anchor; about its weight in training data only presumptions exist
- Monitor your inclusion rate per prompt category and use it as an indicator of your brand penetration