Answer-engine visibility is a measurement problem

When a buyer asks ChatGPT or Perplexity what to buy, there is no results page. There's an answer, it names two or three brands, and there is no position eleven to climb from — you're in the answer or you're absent.

That breaks the entire rank-tracking habit, and most of the industry's response has been to keep the habit and change the label. Here's the measurement problem stated honestly:

The answers are probabilistic. Ask the same engine the same question twice and the brands named can differ. One sample is an anecdote. A measurement needs repeated runs of the same question set, stored, so movement between months is real movement and not noise.

The sources churn. The pages an engine cites this month are not the pages it cites next month. A one-time "AI SEO audit" is a photograph of weather. That churn is the argument for measuring monthly — it is not, and never will be, an argument that anyone can promise you a mention.

The unit of measurement is a question, not a keyword. Buyers ask "best X for Y who hates Z" — whole situations. The instrument has to generate the questions your buyers actually ask, hold that set constant, and score every answer against it.

This is why we built the measuring device before we sold the service: SolvedAgain asks a fixed question set across the major answer engines, stores every grounded answer, and traces every metric to a stored answer you can read. The monthly delta report — same questions, new answers — is the receipt that the work moved something.

No mention rates or Share-of-AI-Voice figures in this post, on purpose: this journal prints a number only when a stored run backs it, and the dogfood case is still accumulating months. The mechanism is the claim.

Find out what the machines say about you: run the audit.