Interpretability vs authority: which dimension matters most?

The AVO discipline names four dimensions: interpretability, structural coherence, authority encoding and intent alignment. A reasonable question, and one we are asked weekly, is which of the four moves the needle most.

We ran a controlled experiment across thirty test entities to find out.

The setup

Each test entity was a real, lightly-trafficked business in a non-saturated UK service category. We held three of the four dimensions constant and varied the fourth across three quality grades: weak, baseline, strong.

We then measured selection probability across ChatGPT, Gemini, Perplexity and Claude on a fixed set of ten buyer queries per category. Selection probability is the percentage of queries on which the entity was named.

The headline finding

Interpretability is the entry condition. Authority is the differentiator.

If interpretability is weak, no amount of authority compensates. The AI cannot reliably name what it cannot reliably classify. Below an interpretability threshold, selection is effectively zero regardless of every other dimension.

Above the interpretability threshold, authority encoding produces the largest marginal lift. Moving authority from baseline to strong increased selection probability by an average of 38 percentage points across the test set. No other dimension produced a comparable single-axis gain.

The four results in order

• Interpretability: a binary gate. Below threshold, selection is near zero. Above threshold, further gains are modest.
• Authority encoding: the largest continuous lever. Strongest single-axis effect once interpretability is sufficient.
• Structural coherence: a multiplier. It does not lift selection on its own, but it amplifies the gains from authority by reducing model uncertainty.
• Intent alignment: a precision lever. Lower headline impact on selection rate, but very high impact on selection in the highest-value queries.

What this means for sequencing

The experiment supports a clear order of operations:

1. Fix interpretability first. Make sure the AI can reliably classify what you do, for whom, where. 2. Resolve coherence second. Eliminate the conflicts across platforms that suppress confidence. 3. Construct authority third. This is where most of the headline lift comes from. 4. Tune intent alignment fourth. This is where you win the queries that actually convert.

Most engagements we see attempt all four in parallel and dilute the result. The data is clear: sequence matters.

The caveat

These results are drawn from the AVO research practice with our methodology and our scoring instruments. They should be read as directional rather than universal. The order of operations is the durable insight; the exact percentages will move as AI systems evolve.