Getting named in AI answers, not just ranked by Google
Assistants kept recommending the competition and nobody could say how often. First measure, then establish the noise floor, then rebuild the content.
41 to 4 %
text position of the first product mention
measured, not guessed
Before
assumptions
After
solid measurement
The problem
AI assistants regularly recommended competitor products, and nobody on the team could say how often that happened or why.
What I built
Measurement that separates neutral questions from ones where the brand already appears. Only the neutral ones tell the truth, and there the honest share was 25.4 percent rather than the more comfortable 33.9 percent. I also established the weekly noise floor so the team would stop reacting to swings that are not movement. Only then were the articles themselves rebuilt.
How it works in detail
The difference between neutral and primed questions is the whole point. Ask a language model about a brand and it will name the brand. That feels good and measures nothing. Only the neutral question about the product category shows whether you appear at all.
The noise floor sits at around one and a half percentage points per week. Without that number every team reads every movement as a win or a setback.
Rebuilding the articles was not about keywords but about order and answerability. A clear answer early in the text can be quoted. A brand story spread over two thirds of the page cannot.
The result
The first product mention moved from 41 percent of the text length to 4 percent, which is where an AI answer actually picks it up. Plus reporting you can defend in a meeting.
What I take from it
The flattering number damages the project. Walk into a meeting with 33.9 percent and you lose credibility the moment somebody questions the method.
Built with
- Neutral-prompt measurement
- Adobe Experience Manager
- Content restructuring