GEO security risk, by the numbers
Share of consumers using AI search
McKinsey finds nearly half of consumers deliberately use AI search today. That answer becomes your digital “first impression,” and SEO alone cannot govern how fair or accurate it is.
%
Share of consumers using AI-mediated search
Users who skip source checks
Industry reporting (via Exploding Topics / Inc.) cites that ~92% of users do not verify citations inside AI answers — so mistakes linger as perceived fact.
%
Respondents who trust AI output without verifying sources
Consumer spend influenced by AI search (2028 outlook)
McKinsey projects ~$750B of spend will flow through AI-mediated discovery by 2028 — proof that assistants are becoming a primary purchase channel, not a sidebar.
trillion JPY (McKinsey scale)
Order-of-magnitude spend touching AI search (McKinsey forecast)
How manipulable LLM outputs are under malicious data poisoning
Anthropic’s research shows that malicious content amounting to as little as 0.00016% of the full training corpus can still have a lasting impact on model outputs. Even very large generative models that appear robust can end up reflecting very small-scale information operations; as misinformation or slanted comparison pieces accumulate, perceptions formed through AI can quietly drift.
%
Share of attack surface where poisoned data can steer LLM outputs
