AI Visibility Academy · Chapter 02

Why every fix starts with a baseline

2 min read · · CiteCue Team

Because AI answers change even when you don't. The same question returns different brands from one run to the next, the engines update their systems, and rivals publish new pages. If you ship a fix and only then start measuring, a better number afterwards proves nothing: you can't separate your change from the background movement. A baseline is the dated record of before that gives your after a meaning.

What a baseline records

A baseline is more than a score. Keep enough to explain the score later:

  • the exact questions, word for word, and the date you ran them
  • which assistants you used, and how many times you ran each question
  • for every answer: whether you were named, which of your URLs was cited, and who else appeared
  • the full answer text

The last item is the one people skip and later regret. When a number moves two months from now, the saved answers show what actually happened: a rival took your place, a review site replaced your page as the source, or the question simply started getting a different kind of answer.

What it lets you do

Attribute. With a before and an after on the same questions, you can connect a movement to a change you made, especially if you also log the date of every change you ship.

Choose the first fix. The baseline shows where you're missing. A question where rivals are named and you aren't is a different job from one where you're named but described wrongly.

Read movement honestly. Even the platforms are cautious here. Microsoft's AI Performance report for Copilot and Bing notes that citation trends can shift with the questions people ask, with updates to your content, or with system and model updates, and that a trend on its own can't be pinned to any single cause. A baseline doesn't remove that uncertainty, but without one you can't even see its size.

When to take it

Take it before the first fix, and again whenever you change the question set. Adding or rewording questions starts a new series, and comparing it with the old one mixes two different measurements.

Take it over several days rather than in one sitting, since a single pass can land high or low by chance. How many runs make a reliable reading covers how much repetition that takes, and mention rate, citation rate and share of voice covers reading the numbers once you have them.

Where to start

Before you edit a single page this month, run your questions, save the answers and write today's date at the top. Beside it, start a plain change log with three columns: date, page, what changed. Those two files are the whole baseline, and every result you report later will rest on them.

Source

See where AI cites you, and where it doesn’t