How AI Chooses Brands Visibility Experiment Research Library Published Experiments About Get in Touch
Category: Home & Sleep AI Surface: ChatGPT Date: Feb 19–20, 2026 Regions: North America + Thailand Query set: 5 broad + constrained queries Variable tested: Regional context + query framing

Mattresses: Location Changes Which Brands AI Recommends

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.

Core Finding
In one sentence

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.

Experiment Design

How we ran the experiment

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.

01

Query set

Broad and constrained prompts, including "best mattress," "best mattress for side sleepers," "best affordable mattress," and other high-intent recommendation queries.

02

Environment

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.

03

Objective

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.

Regional Variation

The same query surfaced different brands by region

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.

Thai IP context (THB pricing)

Brands surfaced

  • Ashley Sleep Essentials
  • Novilla Hybrid
  • Sealy Dreamlife
  • Zinus Cooling Gel Hybrid
  • Siena Memory Foam
  • PlushBeds Cooling Gel
North American network (USD pricing)

Brands surfaced

  • Helix Midnight
  • Nectar Premier
  • Puffy Lux Hybrid
  • Saatva Classic
  • Tempur-Pedic Adapt
  • Purple Restore

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.

Brand Frequency

No single brand dominated the category

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.

Top brands by query appearances (North America)
Helix Midnight
2/5
2
Nectar Premier
2/5
2
Purple Restore
2/5
2
Tempur-Pedic
2/5
2
Siena Memory Foam
2/5
2
All others
1

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.

Pattern Analysis

What the sample suggests

Across the tested queries and regions, four patterns appeared consistently. These are observed selection behaviours, not universal laws.

Pattern 01

AI organised results into segments, not a single winner

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

Region appeared to act as an eligibility filter

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

Price and intent narrowed the candidate set

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

High-consideration categories resist consensus anchors

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.

Structural Model

A simpler model for mattress visibility

This experiment suggests that mattress recommendations follow a layered eligibility model. Brand inclusion appeared conditional rather than automatic.

01

Regional fit

Is the brand recognizable and relevant in the inferred geography?

02

Segment fit

Does the brand fit the expected bucket for the query — value, premium, cooling, pressure relief, and so on?

03

Intent fit

Does the brand's product narrative align with the specific problem the user is trying to solve?

04

Final selection

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.

Cross-Category Contrast

Why this category behaves differently

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.

Protein powder

Broad queries can support a consensus anchor

Broad queries can often support a consensus-style anchor brand. The recommendation task is narrower, more standardised, and easier to summarise.

Mattresses

Broad queries still require segmentation

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

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.

Implications

What this means for mattress brands

Role clarity

Define the segment you want to own

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.

Geographic visibility

Regional presence affects inclusion

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.

Query-fit messaging

Feature claims need to map cleanly to use cases

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.

Strategy implication

Don't chase one universal ranking

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.

Closing Observation
The takeaway

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.

Run Your Own Experiment

Find out where your brand appears —
and where it gets filtered out

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.