Data you can trust because AI never does the counting
Every figure carries its count and its range and the answers it was built on. Significance is stated where it can be so that a reading can be checked.
Six promises
Every reading is made the same way and these are the rules it is made to.
AI never calculates
The models answer and the platform counts. Every share and every range is worked out outside the AI from the answers themselves and stored with the version of the method that produced it so that the same reading can be reproduced later. The tools are AI and platform agnostic and read any model and any search engine the same way.
How sure we are
A model gives a different answer each time it is asked so every question is asked several times and the range is kept beside the share. Raising the number of runs narrows the range and takes a reading from directional to significant. The chart shows the range closing as the runs increase while the share itself hardly moves.
Known versus found
Every question is asked with web search off and with it on. Search off shows what a model has learned about you and search on shows what it can find today and the two need different fixes. A fact the model has learned wrong needs the sources corrected and time. A fact it finds wrong needs the page it finds fixed today. The chart shows each company's share with search off against search on.
What the model has learned about the brand. A wrong fact needs the sources corrected and time.
What the model finds on the web today. A wrong fact needs the page it finds fixed today.
Five models asked three times. The share of answers that say yes is the reading.
Readings over time
Each reading is kept with its date so that change is measured against the range and a move inside the range is never called a change. The chart shows one company across eight readings with its range.
Cleaned and managed for you
Brand names are resolved so that a lookalike is told apart from the brand and a misspelling is counted with it. Duplicates are removed and every run is kept with its date and model version and nothing is overwritten. The result is one consistent and validated data layer that your decisions and your dashboards and your agents all read from.
Questions people ask us
Can we check a Model Research reading ourselves?
Yes. The AI answers the reading was built on come with it and every figure says how many answers sit behind it and what its range is. We would rather you checked than took our word for it.
How many times is each question asked?
Three times on each of five models with search off and on as the standard and more when a client wants a tighter range. The chart above shows what more runs buy.
What does a Model Research reading include and how often is it produced?
A reading is a dated report on one company and its rivals with the full set of AI answers it was built on. It can be a one-off snapshot or repeated at the interval you choose and repeating it builds the record that shows how fast the answers are changing.
Where is the method itself described?
In words on the Methodology page. The frameworks are published and versioned and the prompts are not because the prompts are the instrument.
Questions about the method are welcome. Our details are on the contact page.