Research

How AI picks who to name.

How we test answer engines, where measurement gets tricky, and the studies we're running.

How we test

Method before findings.

Every number we publish will come from the same process. If a result can't be reproduced, we won't publish it.

Step 01

Fixed prompt sets

We write buyer-intent questions up front and never change them mid-study.

Step 02

Repeated runs

Each prompt runs several times per engine, because answers vary from run to run.

Step 03

Structured logging

We record who is named, how they are described, which sources are cited, and the date.

Step 04

Scheduled re-tests

We re-run on a schedule, so changes over time are measured, not remembered.

Limits

Why measuring AI answers is hard.

Honest measurement starts with what can distort it.

Answers vary

The same question can produce different answers on different runs, so one screenshot proves little.

Context matters

Location, settings, account history and how a question is phrased can all change what an engine says.

Engines change

Models and search features are updated often, so a result is a snapshot of one date, not a permanent fact.

Research agenda

Studies in progress.

No findings are published yet. These are the questions we are testing first, and we will publish the method alongside every result.

Planned

How consistent are AI answers?

How often does the same buyer question return the same brands across repeated runs and across engines?

Planned

Which sources get cited?

For category-comparison questions, how often are brand sites, review sites, publications and forums cited?

Planned

Does structured data change citations?

Do pages with matching FAQ or organization markup get cited more often than similar pages without it?

Get the first study.

One email when we publish. No spam.