Methodology · AI Visibility

The method: how the number on your dashboard is made

By Kristian Stig Henriksen · Dear Future

This page is the full answer to a fair question: where does the number come from? It is written so you can check every link in the chain. Short version: neutral buyer questions, asked every day under the same conditions, read out by name and reported with the uncertainty visible.

The questions: a market survey, not a keyword list

The foundation is a baseline set of questions buyers in your category could actually ask, balanced across topics and buying stages. It is built like a market survey: broad and fair, not by search volume, and never with questions that mention or flatter your brand. On the front page we call them the guides.

On top of that you can point custom prompts at the places you want to win: one topic, one product line, one corner of the market. That is the task force. How many questions your audit runs with depends on your package; the principle is the same at any count.

Asked every day, under the same conditions

• The same questions every day, so the trend is comparable. The models change continuously; a one-off audit is a photo, a daily one is a time-lapse.
• Real web search switched on, asked from your country and in your language, so the answers mirror what real buyers are shown. A model answering from memory cites almost nothing.
• Named model versions on every data point. An unnamed model can be swapped out without your trend line ever telling you.

What we read out of every answer

In each answer we detect who is named, who is cited (linked), in what order, and in what tone: positive, neutral or negative. Mentions are matched on brand names and their variants, not just the domain, so “Your Brand Inc.” counts even when the link is missing. Every raw answer is stored in the EU for two years, and your team can open the one behind any number.

Uncertainty, shown honestly

Any share measured on a sample carries uncertainty, and we show it instead of hiding it. Every estimate has a 95% confidence interval computed with Wilson's method, which holds up better than the classic approximation when the sample is small or the share sits near 0% or 100%. That is exactly the situation of a brand that is rarely, or almost always, named.

The shaded band on the charts is normal variation. A move only counts as a development once it stays outside the band for several weeks. We never call a single day's swing a development, and we do not react to one.

What we deliberately do not do

• No questions designed to make your brand look good.
• No weighting by search volume; an LLM answer is one answer, not a ranking.
• No unnamed models, and no from-memory answers presented as search.
• No weekly samples drawn up as daily curves.
• No number without a stored raw answer behind it.

What the field is called

What we measure goes by several names: LLM SEO, AI SEO, GEO (generative engine optimization) and AEO (answer engine optimization). The names cover the same question: how visible is your brand in the answers AI assistants give?

Also read our LLM SEO guide and the guide to Share of Voice in AI, or start on the front page.

← Back to the front page