We tested how AI systems recommend mattress brands across a set of buyer-style queries, running the same queries from different locations. The goal was to observe how context affects which brands are included in responses.
Across the queries we ran, the same question often returned different brands depending on location. This suggests that AI recommendations are influenced by contextual factors — and that visibility is not consistent across users.
In this sample, ChatGPT did not behave as if it were searching for one universal "best mattress." It appeared to sort brands into recommendation buckets shaped by region, price tier, and functional need — then select within those buckets.
That matters because it points to a different visibility model than lower-consideration categories. In mattresses, repeated appearance did not come from broad reputation alone. It appeared to come from whether a brand fit the right segment for the query being asked.
This was a small, directional experiment designed to observe recommendation behaviour, not to produce a definitive category ranking. The same set of mattress queries was tested across two regional contexts to see whether brand selection stayed stable or changed with geography.
Broad and constrained prompts, including "best mattress," "best mattress for side sleepers," "best affordable mattress," and other high-intent recommendation queries.
ChatGPT only, tested across two network environments: North America and Thailand. The focus was the model's first-pass recommendation behaviour, not search rankings or retailer results.
Understand whether AI defaults to the same brands across mattress queries, or whether visibility is filtered by region, price, and functional fit before selection occurs.
Because this page reflects a limited sample, the findings should be read as directional evidence of category logic, not as a universal rule for every mattress query.
One of the clearest findings in this experiment was regional variation. Identical queries produced materially different brand sets depending on inferred geography, even when the overall recommendation structure remained similar.
The brand pools changed, but the recommendation structure did not. In both regions, ChatGPT still organised answers around the same kinds of roles — general pick, value pick, cooling pick, pressure-relief pick, and so on. In this sample, geography appeared to influence which brands were eligible before the model chose among them.
Unlike a commodity-style category where one default brand can reappear across most queries, mattress visibility in this sample was distributed. No single brand emerged as a universal winner across the query set.
In a fragmented, high-consideration category, even 2 out of 5 appearances can indicate a stable role. The key pattern here is not dominance — it is distributed eligibility.
That is an important difference from categories where one brand clearly anchors the recommendation set. In mattresses, repeated appearance seems to reflect contextual fit within a segment rather than broad category ownership.
Notably, several familiar brands did not appear in this sample at all, including Casper, Tuft & Needle, and Leesa. Their absence does not imply weak products; it suggests they were not the model's most obvious fit for the prompts tested here.
Across the tested queries and regions, four patterns appeared consistently. These are observed selection behaviours, not universal laws.
Pattern 01
Even broad mattress queries were rarely answered with one dominant brand. Instead, ChatGPT tended to structure recommendations into labelled categories such as "best overall," "budget," "cooling," "pressure relief," or "premium." In this sample, visibility was distributed across roles rather than concentrated around one consensus brand.
Segmented answer structures appeared across both regional contexts, even when the specific brands inside those segments changed.
Pattern 02
The same query produced meaningfully different brand sets across network contexts. What stayed stable was not the brand list, but the shape of the recommendation. This suggests that geography may influence which brands are considered first, before the model assembles the final ranked or segmented answer.
For the same style of query, Thai-context responses surfaced Novilla, Zinus, and Sealy, while North American responses surfaced Helix, Nectar, Puffy, and Purple.
Pattern 03
Queries with a clear price or functional angle did not simply reorder the same brands; they often pulled in a different subset entirely. Affordable queries favoured lower-priced brands, while cooling, side-sleeper, or pain-related prompts favoured brands whose positioning or construction narrative more clearly matched that need.
Budget-oriented recommendations and cooling-oriented recommendations showed little overlap, indicating that query framing materially changed which brands qualified.
Pattern 04
In a category like mattresses — expensive, personal, and feature-heavy — ChatGPT did not appear eager to endorse one universal winner. The model behaved more like a sorter than a crowner, matching brands to roles instead of converging on a single default answer.
Maximum observed frequency was 2/5 queries per brand, far below the kind of repeated dominance seen in simpler product categories.
This experiment suggests that mattress recommendations follow a layered eligibility model. Brand inclusion appeared conditional rather than automatic.
Is the brand recognizable and relevant in the inferred geography?
Does the brand fit the expected bucket for the query — value, premium, cooling, pressure relief, and so on?
Does the brand's product narrative align with the specific problem the user is trying to solve?
Only after those filters appear to narrow the pool does the model choose which brands to surface.
For brands, that means visibility may depend less on broad awareness alone and more on whether the model can place you clearly inside the right recommendation role.
This experiment becomes more useful when contrasted with simpler recommendation environments. In some categories, AI can rely on a stable default brand. In mattresses, the sample suggests it cannot do that as easily.
Broad queries can often support a consensus-style anchor brand. The recommendation task is narrower, more standardised, and easier to summarise.
Broad queries still require segmentation. Region, price, sleeping style, and construction features all shape the recommendation set, making a single universal winner less stable.
The broader implication is that AI visibility strategy is category-specific. A brand cannot assume that what works in a lower-consideration or more standardised market will also work in a layered, high-consideration purchase category.
If AI organises recommendations into buckets, brands need a clear role inside that structure. Being vaguely "good for everyone" may be less effective than being strongly legible as a premium hybrid, cooling specialist, pressure-relief option, or best-value pick.
A strong brand in one market may not be visible in another if its regional signals are weaker. Distribution, retailer presence, pricing context, and local brand familiarity may all influence whether the model treats the brand as eligible.
If your brand wants to appear in "side sleeper," "cooling," or "back pain" prompts, your product language has to make that fit obvious. Ambiguous construction narratives create recommendation risk.
In this type of category, the goal may not be to become "the" AI-approved mattress. It may be to become one of the brands that reliably qualifies in the right high-intent contexts.
In this sample, ChatGPT did not treat mattresses as a winner-take-all category. It appeared to distribute visibility across regionally and functionally appropriate segments, then select brands within those segments. For mattress brands, that suggests AI visibility is shaped by eligibility, fit, and role clarity — not reputation alone.
Every category has its own recommendation logic. We run controlled experiments to show which prompts include your brand, which competitors replace you, and what signals seem to shape AI selection.