What You Need to Measure

Transcript

If you want to win the agentic shelf, you have to measure it. And almost everyone measures the wrong thing. They check whether you appear at the first prompt, scrape that answer, and call it a score. But being seen at the first prompt was never what decides whether you get chosen.

Start with presence, but measure it properly. Not just, am I mentioned. Breadth: how many of the questions you show up in. Depth: how much of the answer is actually about you. Resilience: would you still be there if they asked again tomorrow. And sentiment: is the mention helping you, or quietly hurting you.

But all of that is still just the first prompt. And the decision doesn't happen there. It happens turns later, deep in a conversation you can't see. So you can't measure it with a single question. You have to measure the whole journey.

Which means volume, and variety. Multi-turn conversations, not single questions. Anchored, when they name you, and generic, when they don't. Short four-turn journeys, and long ones, eight to twelve turns deep. The way a human shops, and the way an agent shops. Run again and again, over time, because the answer drifts. And all of it live. Never scraped.

Then, inside every journey, measure the reasoning chain itself. First, where you fall out. Not whether you started strong, but the exact turn you drop out of contention. For most brands, that's the comparison turn, deep in, when the field narrows. Then why. Pushed out on price? Beaten in a comparison? Swapped for a near-equivalent? Quietly deflected? Each reason is a different problem, and you can't fix what you haven't named.

And now the measurement almost no one takes. First, everything the model already knows about your brand. The full picture. It's usually a great deal. Then, how much of that it actually uses at the moment it decides. Which is usually a fraction.

So ask the question that turns measurement into action. If the model had used everything it already knew, would the recommendation have changed? Where the answer is yes, that's the Linkage Gap, and it's winnable. Where the answer is no, where the model holds a real judgement about fit that evidence won't move, that's the Reasoning Gap. You measure that too, honestly, because it tells you what you can't win. Which is exactly what makes the Linkage Gap so valuable. It tells you what you can.

That's what winning the agentic shelf actually asks you to measure. Not a number on a scraped dashboard, but the whole reasoning chain, live. Where you fall. Why. What the model knows. What it uses. And how much of the gap between them you can close. Measure that, and for the first time, you can see the decision. And change it.