A brand that is easy to recommend when someone asks for the best product in a category can be absent when the question carries a price ceiling or a price floor. This post measures how often that happens in our sample, and how much weight the number can bear. It does not say why it happens, because the design cannot separate the causes.
What did we compare?
Every category in the October 2026 edition has five prompts. Two of them matter here: a budget prompt with a price ceiling and a premium prompt. Each prompt was asked once on each engine, OpenAI's API with web search (gpt-6-luna) and Perplexity's Agent API with the Sonar model (perplexity/sonar), so a category usually has 2 budget answers and 2 premium answers. That holds for 155 of the 160 categories. In the other 5, a budget or premium answer is missing because a Perplexity request failed.
Here is what a pair looks like for air fryers. The category page lists both prompts: "best air fryer under $80" and "best premium air fryer oven, $250 or more". The prompts name different price bands. In 118 of the 120 categories whose prompts have not changed since October, the premium prompt states a price floor, and all 120 budget prompts state a ceiling. (The other 40 US categories get new premium wording from the November 2, 2026 run, and we did not count their prompts here; this post covers only the October edition.)
For each category we took every brand recommended in its budget answers and every brand recommended in its premium answers. We divided the brands on both lists by the brands on either list, a measure called Jaccard overlap, and averaged over categories with equal weight. The methodology explains how brands are extracted and counted.
How much did the two lists overlap?
The mean overlap between a category's budget list and its premium list was 8.2%. In 70% of the 160 categories, which is 112 of them, the two lists had no brand in common. In the other 48 categories the lists shared at least one brand, and there the overlap averaged about 27%.
Short lists make the measure jumpy. Two lists of four brands that share one brand score 14.3%, so a single shared brand moves a category a long way.
For context, the chart sets the budget and premium figure beside two neighboring comparisons from the same edition.
The price pairs sit well below the engine figure, but the bars are not like for like: the engine figure compares lists built from about 4 answers per engine on identical questions, and each price pair compares two answers with two on different questions, which scores lower on shorter lists. We did not test how much of the gap comes from the price band, the wording, list length or ordinary run-to-run variation.
Why should you read the number carefully?
- The questions asked for different price bands. A ceiling and a floor ask for non-overlapping products by construction. The data cannot say how much of the 8.2% is the price band and how much is the assistants' reading of "budget" and "premium".
- There is one answer per prompt per engine. Answers vary from run to run, so a category's pair is an example, not a finding about that category.
- The 160 categories are ours. We chose them across three markets, they are not a random draw of all products, and this edition pools the US, Mexico (in Spanish) and Canada. We do not report the overlap by market.
- It is the API, not the app. The apps can differ in personalization, memory, location and product interface.
- It describes answers, not shoppers. We sampled answers to prompts we wrote. We did not observe what anyone searched.
How many brands are recommended for each kind of question?
Each answer recommends few brands, so each list is short and one shared brand moves the score a lot. The table gives brands per answer by question type, for both engines together and all three markets, October 2026 edition. The alternative-to-a-brand question is left out because it names a brand in the prompt and lists far more brands.
| Question type | Answers | Mean brands | No brand | 1 to 3 brands | 4 to 6 brands |
|---|---|---|---|---|---|
| Budget | 318 | 2.2 | 0.3% | 93.4% | 6.3% |
| Premium | 317 | 2.6 | 0% | 84.9% | 14.5% |
| Best overall | 318 | 2.4 | 9.1% | 72.3% | 17.9% |
| Specific use case | 318 | 2.2 | 4.1% | 87.1% | 8.5% |
Fewer than one answer in a hundred, for any of these question types, recommended 7 or more brands. Two readings are safe. A price-limited question almost always gets brands recommended, while the open best-overall question was the most likely to get an answer that recommended no brand at all (9.1% of its 318 answers). And a premium answer recommended a few more brands than a budget answer, 2.6 against 2.2 on average, which the data does not explain.
For a brand, the practical meaning is the slot count. With a mean of 2.2 brands in 318 budget answers, one budget answer has about two places, so a brand has few chances to be recommended.
What can a brand do with this?
Nothing here proves that any change will get a brand recommended. These steps use the numbers to decide what to measure and what to write.
- Pin down your bracket. Write the price a shopper would type as a ceiling or a floor for your product, in your market's currency. If your product sits between the two bands, check both questions anyway: the data does not show where such a product lands, and a price that fits neither the ceiling nor the floor in your market's currency may fit neither question.
- Ask each question separately. Run the best-overall, budget, premium and use-case questions for your category on each engine, several times, and record the brands in each list. The measurement method describes a manual way to do this.
- Compare the lists, not one run. If you appear for best overall but never for budget, your price or your price wording may be the gap. If you appear for neither, the problem is earlier, and the October overview shows how few slots exist.
- Make the price band readable on your pages. State your price as text with its currency, and say what the product is for at that price. This is reasoning, not evidence: the answers we sampled came from web search, and pages that state the facts a question asks for are easier to retrieve and read. We have no measurement that this changes a recommendation.
- Track each bracket on its own. A single "share of answers" figure that blends the two hides the pattern above.
The full edition is in the October 2026 report.
Frequently asked questions
Does this mean a premium brand cannot be recommended for a budget question?
No. In 48 of the 160 categories the two lists shared at least one brand. The data shows that shared brands were uncommon in this sample, not that they are impossible.
Is the overlap the same in Mexico and Canada?
We do not know from this edition. We computed the budget and premium figure over all 160 categories and did not split it by market. Mexico's prompts are in Spanish, with different categories and prices in pesos, so a market figure would not be like for like.
How we write: drafted with AI tools from Dopbot Index data and public sources; every number and outside fact is traced to its source in a claims record and checked before publishing. More about how we write; corrections to hello@dopbot.com.