The calculation that marketing managers have made for years was: the higher on Google, the more visits, more leads. In 2026, this calculation is incomplete. A large part of the questions is answered within the AI tool itself, and the user never reaches the website. When the buyer asks, "what is the best company for X in my case," the answer comes from three or four names chosen by those who have experienced the model training and what is published about them.
This is where Share of Model comes in. The idea is simple: if the response mechanisms mention 5 brands and yours appears in 2 of those 5 answers, your presence is 40%. It’s not a vanity metric — it’s the 2026 version of share of voice.
Why this matters more than ranking
Ranking measures position on a list. Share of Model measures presence in the decision. A customer who asks ChatGPT about your category receives a summary with names, advantages, and objections. If your brand is not there, you lose the conversation before the first click — and you don’t even appear in Analytics for you to notice.
The most revealing data of the year: most companies say they will optimize for AI search, but almost all have zero recorded mentions. In other words, the field is open for those who start measuring now.
What fuels the citation
- Consistency of brand description on third-party sites: same proposal, same name, same services.
- Content that answers complete questions, with definition, example, and caveat.
- Editorial authority: publications, awards, proprietary data, and disclosed cases.
- Structured data on the page (schema) that explains who you are and what you sell.
How to calculate in practice
Compile a fixed set of 20 to 30 questions that a buyer would ask in your category — including uncomfortable ones, like "what are the risks of hiring company X?" Run the same set every month on ChatGPT, Gemini, Perplexity, and Google’s overview. Note: did the brand appear? Was it mentioned with a link? Was it mentioned with an error or outdated information?
The calculation: (responses with your brand ÷ total questions) × 100. Repeat for the three main competitors. The comparison is the number you take to the meeting.
Common measurement mistakes
- Always asking the same thing. Small variations in prompts change the response. It’s better to fix the set and record the date.
- Confusing mention with recommendation. Being cited as "an option" is different from being cited as "the most recommended."
- Ignoring factual errors. Sometimes the brand appears, but with an old price or discontinued service. This is damage, not victory.
What to do with the number
Low Share with good content indicates a distribution problem: lack of external mention. The correction is to gain presence in guides, comparisons, interviews, and industry directories — the places from which AI draws.
Good Share and poor conversion indicate an expectation problem: the answer that AI gives does not match the page that receives the visitor. Then the correction is in messaging, not SEO.
Also read:
- SEO consulting
- Share of model as part of brand work
- Content with data that AI cites
- Advertising within the conversation with AI
- GEO: appear in AI responses
Start small, but start
It’s not necessary to buy an expensive platform in the first month. A spreadsheet, a fixed set of questions, and 40 minutes a month can already show the trend. What you can’t do is continue measuring only clicks when the buyer's decision happens before the click.
Kaizen Agency works with performance marketing and data-driven content for companies that want to be found where the decision happens. Talk to us to create your first visibility diagnosis in AI.
Also see: AIO (AI Optimization): How to Teach AI to Recognize Your Brand.




