The Bearing Method

The five stages

Assess. We take the bearing first — a structured, repeated read of how your company currently appears, or fails to appear, across the AI systems your buyers actually use in the target market.

Understand. A missing or inaccurate result has a specific cause. We trace it — to the sources the platform draws from, the structure of the information available to it, the language and context it encounters.

Optimize. We work to correct the course — direct work on the sources, structure, and content that shape how AI systems describe and represent you. We control the inputs; we don't claim to control what a given model ultimately outputs.

Verify. We check the corrected position against the original baseline, under the same conditions, on the same platforms. Improvement is measured, not assumed.

Monitor & Refine. We take the bearing again, on a recurring basis, and correct again as needed — because platforms and sources don't stay still. This is the method's own fifth stage, not a follow-up service bolted on: a one-time correction to a moving target doesn't stay correct.

What we measure at the Assess stage

For a defined set of real buyer questions, we assess six things: whether your company appears at all; whether it is cited or referenced when it does; how prominently, relative to alternatives; whether the description is accurate; whether it is surfaced as a relevant recommendation; and whether these results hold up when the same question is asked again.

How we measure it

AI-generated answers are not stable from one query to the next. Independent research on AI search visibility has found that results vary across repeated runs of the same question[1] — which means a single test tells you very little. We run each query multiple times, across multiple platforms, over a defined period, rather than testing once and reporting a snapshot.

What the evidence looks like

A written report showing the actual outputs — not only a score, but the underlying evidence: what the AI system said, when, and in response to what question.

What we don't claim

A recent independent review of generative-engine-optimization research examined 45 studies published between 2023 and 2026 and found that most techniques marketed as improving AI visibility do not show a stable, reproducible effect across different AI platforms[2]. Two elements were consistent across the research: how directly relevant your content is to the question being asked, and how it is positioned within that content. We build on what the evidence actually supports, and we are explicit with clients about where the evidence is still thin.

Current independent research has not identified a reliable way to guarantee AI citation or ranking outcomes — so we don't offer one.

Why this matters

In a market where most AI-visibility claims are unverifiable, we think the more useful thing to sell is the evidence itself. For more on the research behind the repeated-run protocol, see Why AI Answers Change Between Runs in Insights & Research.