BACKTIER

Measurement

How we measure AI visibility results.

We do not publish anonymised percentage gains. Numbers without a named client, a fixed prompt set, a measurement window, and a stated baseline are marketing, not evidence. What follows is the method we actually use, so you can judge our reporting before you hire us.

The measurement cycle

01

Baseline

Before any implementation work, we record how AI systems currently answer a fixed set of prompts in your category: which sources they cite, whether your brand appears, and whether what they say about you is accurate. That prompt set is written down and frozen so later readings are comparable.

02

Instrumentation

We track the same prompts across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews on a repeating schedule. Readings vary between runs because these systems are non-deterministic, so we report ranges and trends rather than single observations.

03

Implementation

Entity architecture, structured data, retrievable content, and citation pathways are shipped in tracked increments, so each change can be tied to a date in the measurement record instead of being credited retroactively.

04

Reporting

Each report states what was measured, over what window, against which baseline, and what remains unchanged. Where a movement cannot be attributed to a specific change, we say so.

What we will not claim

  • Answer engines are non-deterministic. The same prompt can produce different sources on consecutive runs, so a single citation is not evidence of a durable change.
  • Model and index updates happen outside your control and can move results in either direction independently of any work we do.
  • No agency can guarantee a citation, a ranking, a knowledge panel, or placement in an AI system's recommendations. Anyone promising those outcomes is describing something they cannot control.
  • Traffic and revenue effects are influenced by factors beyond AI visibility, so we do not present them as directly attributable unless the client's own analytics support it.

If you want to see what AI systems currently say about your brand, the fastest route is a baseline reading of your own prompt set.