From one question to many answers
An assistant almost never gives the same answer twice. So we ask every question on five assistants, five times on each. One question becomes 25 answers, and we report the pattern across them.
Which has more protein, Brand A or Brand B?
One answer is an anecdote. Twenty-five show a pattern.
The size of a measurement
20 questions × 5 assistants × 5 repetitions 500 answers to read and classify
Standard depthMove the dials. 20 × 5 × 5 = 500 is an illustrative configuration, not a package: the scope is agreed in writing before the first question runs.
What counting means
We read and classify every answer. In each one we check three things. Does it name the brand? Does it recommend it? Does it prefer it over the others?
Illustrative example
We count what is written
An answer can suggest a product is healthy without saying so. We only count “healthy” when the answer writes the word. What an answer implies goes into the analysis, never into the count.
Brand A Natural and Brand C Greek are the better buys in Mexico, offering the best balance of low sugar, high protein and good results in quality evaluations…
Types of question, types of conclusion
How a question is worded decides what its answers can prove. Pick a type to see what it measures and what you can do with the result.
Which Greek yogurt brand is the best buy in Mexico?
Measures visibility- Measures
- Whether the brand appears, whether it is recommended, and which sources are cited
- Lets you conclude
- Whether the assistants mention the brand on their own
- What you can do
- Work on the cited sources and cover the gaps where the brand does not appear
What is Brand A Natural?
- Measures
- What the assistants say about the brand
- Lets you conclude
- Whether the description is correct
- What you can do
- Correct wrong descriptions on the page they come from
When the question names the brand, the brand appears in 100% of the answers. That number tells you nothing.
Which has more protein, Brand A or Brand B?
- Measures
- Who wins, why, which sources and which figures
- Lets you conclude
- Which pages decide the comparison
- What you can do
- Work on those pages and correct wrong figures with the label data
When the question names the brand, the brand appears in 100% of the answers. That number tells you nothing.
I'm a mom of young kids. Which yogurt is best for their lunchbox?
Measures visibility for that buyer- Measures
- Whether the brand appears for that buyer, and which sources change
- Lets you conclude
- Whether the assistants mention the brand for each type of buyer
- What you can do
- Content for each buyer; measured in a separate test, after the main measurement
| Measures | Lets you conclude | |
|---|---|---|
| Category, unbranded | Whether the brand appears, whether it is recommended, and which sources are cited | Whether the assistants mention the brand on their own |
| Brand identity | What the assistants say about the brand | Whether the description is correct |
| Head to head | Who wins, why, which sources and which figures | Which pages decide the comparison |
| Buying situation | Whether the brand appears for that buyer, and which sources change | Whether the assistants mention the brand for each type of buyer |
How we word the questions
The wording can decide the answer before the assistant starts. So we adjust some questions before measuring. Each card shows a question as it might arrive, the version we measure, and why.
No verdict built in
As it might arrive
Is Brand A an ultra-processed food I should avoid?
As we measure it
How processed are Brand A's products?
Why
Assistants tend to accept what the question takes for granted. If the question says "avoid it", the answer leans that way.
No judgements the buyer did not ask for
As it might arrive
Which yogurt brands do you consider healthy?
As we measure it
Which Greek yogurt brand is the best buy in Mexico?
Why
Ask for "healthy" and you get more "healthy" labels, because the assistant follows the request. When a buyer really does ask about health, we keep the question the way they would ask it.
Name the country
As it might arrive
Which yogurt is the healthiest?
As we measure it
Which yogurt is the healthiest in Mexico?
Why
Without the country, some assistants answer about other markets, such as the United States or Spain.
One idea per question
As it might arrive
Does a yogurt-and-granola breakfast give me enough protein without breaking my diet or raising my sugar?
As we measure it
Which breakfast is high in protein and low in sugar?
Why
Three conditions in one question get a partial answer to one of them, and we do not know which.
Include what the answer needs
As it might arrive
How many yogurts should I eat to reach my protein goal?
As we measure it
How many yogurts do I need to get 30 g of protein?
Why
Without the goal, the assistant replies with a question instead of an answer, and that reply cannot be counted.
One way to write the brand
As it might arrive
Is Brand A+ healthy? / Is the BrandA Plus line light?
As we measure it
Brand A+ in every question
Why
Another way of writing it is another question. It can change the answer and makes the results harder to add up.
Anatomy of a head-to-head answer
An illustrative answer to “Which has more protein, Brand A or Brand B?”, built on the pattern we see in real answers. Each layer of the answer supports a different kind of conclusion.
Illustrative answer
If you mean plain Greek yogurt, it depends on the product:
- Brand A Plain Greek: about 9.5 g of protein per 100 g.
- Brand B Plain High Protein: about 10.1 g per 100 g.
So in that comparison, Brand B has slightly more protein. But not every version is the same: for example, in a 2025 quality study, Brand A Light had 6.2 g/100 g and Brand B Light 5.8 g/100 g. Always check the nutrition label of the exact product. (consumer-study.example)
The finding is the source, and the figure that can be verified.
What we do with this
Sources to influence
The pages the assistants cite when they answer about your category. A stronger presence there changes what they say.
Claims to correct
Wrong descriptions or figures about your products, each with the page it came from.
Presence nobody asked for
How often the assistants mention your brand when nobody names it. This number tracks visibility over time.
The examples are illustrative and use fictional brands. They follow the pattern of real answers from ChatGPT, Gemini, Claude, Google AI Overviews and Google AI Mode.