How AI Search Is Recommending Air Purifiers: Monthly Trends

Executive Summary

Levoit remains the category leader in AI search recommendations for air purifiers in September 2026, holding 82.8% valid recommendation coverage. Coway holds second place at 73.1%, a 9.7-point gap behind Levoit, and Blueair is third at 62.7%.

Across the three-month series from July to September, IQAir is the largest riser, climbing 10.8 points from 35.0% to 45.8%, a significant move that has held for two consecutive months, including a 5.3-point gain from August to September. Dyson is the largest decliner, falling 13.1 points from 24.6% to 11.5% — a significant move both across the full period and in September alone, when coverage fell a further 7.9 points from August's 19.4%. Molekule also declined significantly across the period, from 2.9% to 1.0% valid recommendation coverage, continuing a two-month downward run.

Levoit's own coverage softened from 85.7% in July to 82.8% in September, a change the benchmark classifies as stable rather than significant, though the decline has now continued for two consecutive months. Levoit's rank-one rate fell more sharply over the same period, from 34.8% to 27.1%.

Each monthly run begins with 800 prompt-surface observations (504 unique questions) in July and 800 prompt-surface observations (545 unique questions) in September across the benchmark's defined AI/search surface universe. Of those, 800 in July and 800 in September mentioned a tracked brand or competitor; 787 were relevant and 13 were irrelevant in July, while 786 were relevant and 14 were irrelevant in September. The public metrics use the 615 qualified observations in July and 598 qualified observations in September that survive both qualification stages. August contributed 700 collected prompts with 546 qualified observations.

Key Findings

Benchmark Context

The report separates the raw collection universe from the qualified analysis set. Brand-level recommendation percentages are calculated within the qualified benchmark set, which is smaller than the raw collection due to relevance filtering and reserved prompts.

Both months recorded qualified observations across six AI surface families. August dipped to five families when ChatGPT did not contribute qualified observations, and September restored breadth to six. The July and September runs offer the cleanest comparability across the three-month series.

Benchmark-Level Metrics

AI Recommendation Trend

A two-brand top tier holds while IQAir and Dyson diverge

Levoit's leadership is intact but has softened across the three-month series, while Coway has gradually narrowed the gap. The clearest structural pattern is the widening distance between Dyson and IQAir: the two brands were 10.4 points apart in July and 34.3 points apart in September, with the gap widening in both months of the series.

The category's movement this period concentrated in three brands: IQAir's rise and Dyson's and Molekule's declines are classified as significant, while Levoit's softening and Coway's gain are classified as stable. Blueair, Rabbit Air, and Winix each reversed a one-month decline from August, though none of these shifts were large enough to be classified as significant, indicating the middle of the category is holding rather than shifting further this month.

What Changed This Month

IQAir: Significant riser, now clear in fourth position

IQAir's valid recommendation coverage rose 10.8 points from 35.0% in July to 45.8% in September, a significant move that places it 4th in the category. In absolute terms, IQAir recorded 274 valid recommendations in September versus 215 in July. The rise has continued for two consecutive months, including a 5.3-point gain from August to September.

The raw mention presence rate rose 13.1 points, from 40.6% (250 of 615 observations) in July to 53.7% (321 of 598 observations) in September. The top-three rate rose 3.0 points, from 18.4% to 21.4%, and the rank-one rate rose slightly, from 7.3% to 8.0%. The presence increase outpaces the increase in top-three and rank-one placement, suggesting the coverage gain is concentrated more in broader mention presence than in top-position placement.

Highest-priority diagnostic: Which surfaces or prompt types are driving the presence gain, and is the rise concentrated in specific recommendation contexts or spread across the full question set?

Dyson: Significant two-month decline in coverage and presence

Dyson's valid recommendation coverage fell 13.1 points from 24.6% in July to 11.5% in September, a significant decline that places it 8th in the category. The sharper move came from August to September, when coverage dropped 7.9 points. In absolute terms, Dyson recorded 69 valid recommendations in September versus 151 in July.

The raw mention presence rate fell 14.6 points, from 34.6% (213 of 615 observations) to 20.1% (120 of 598 observations). Dyson's placement data also softened: the top-three rate fell from 7.6% to 4.2%, and rank-one placements fell from 6 in July to 1 in September.

Highest-priority diagnostic: Which prompts stopped surfacing Dyson entirely, and which brand appears in those contexts instead?

Molekule: Continued decline toward near-zero visibility

Molekule's valid recommendation coverage fell 1.9 points from 2.9% in July to 1.0% in September, a significant move on a small base. In absolute terms, Molekule recorded 6 valid recommendations in September versus 18 in July. The raw mention presence rate fell 2.5 points, from 5.5% to 3.0%, and Molekule recorded zero top-three placements and zero rank-one placements in September, down from 6 and 1 respectively in July.

Molekule appeared in 18 of 598 observations in September. The net sentiment score also declined to 0.2 from 0.4 in July, indicating that when Molekule is discussed, the tone is less favorable than earlier in the series.

Highest-priority diagnostic: Are there any remaining prompt contexts where Molekule surfaces as a viable recommendation?

Levoit's softening rank-one position

Levoit's overall coverage declined 2.9 points from 85.7% in July to 82.8% in September, a stable classification, but its rank-one rate fell 7.7 points, from 34.8% to 27.1%. The brand recorded 162 rank-one placements in September versus 214 in July. Its top-three rate declined from 68.8% to 64.9%.

Levoit's raw mention presence held close to steady, at 92.8% in September versus 95.0% in July, meaning the brand still appears in nearly all answers even as its rank-one rate softened. Coway's rank-one rate rose 2.8 points to 22.2% over the same period, and Blueair's rank-one rate rose 4.7 points to 14.9%.

Highest-priority diagnostic: Which prompt types moved away from Levoit as the first recommendation, and which brands are capturing those rank-one placements?

Buyer-Intent Interpretation

In September 2026, all 598 qualified observations fell into the Brand Recommendation cluster. No observations qualified for the pricing and value or multi-brand comparison clusters, matching the July and August profiles where all observations also fell into the brand recommendation cluster.

The public benchmark therefore answers the question of which brands AI systems recommend for air purifier needs. It cannot yet speak to how brands are positioned on price or value, nor which brand wins in explicit head-to-head comparisons. Those commercial questions require additional research beyond the current benchmark scope.

Brand Opportunity Summary

The benchmark identifies where attention is warranted across the ten tracked brands; a company-level analysis is needed to explain why the observed patterns are occurring.

Evidence Behind the Benchmark

The aggregate metrics are built from prompt-level observations that capture the query, the AI surface, the recommendation outcome, the rank position, sentiment, and citations where exposed. Company-level analysis can go deeper into prompt, competitor, surface, and evidence patterns. Source presence is not automatically treated as proof of causation.

Interpretation Notes

• Small-count movements: Brands with low absolute valid recommendation counts (Molekule at 6, Honeywell at 61) can show percentage movement from small changes in the underlying count. Molekule's 1.9-point decline reflects a drop from 18 to 6 valid recommendations.
• Qualified denominator: All percentages use the 598 qualified September observations, not the 800 raw prompts collected. The July baseline used 615 qualified observations.
• Surface breadth variation: August dipped to five AI surface families before September restored breadth to six, which affects the comparability of absolute counts across the middle month.
• Movement classification: The benchmark's significance classification identifies which changes are large enough to track across months; it does not by itself establish the cause of those changes.

Next Step

The Public Benchmark Shows Where a Brand Is Winning or Losing. A Company-Level Audit Shows Why.

The aggregate percentages only reveal the surface. Beneath them sit the questions that matter: which high-intent prompts are won, which competitor takes the recommendation when a brand loses, what attributes AI associates with each option, and which external sources shape those answers.

Dyson's decline across the period could stem from changes in a small number of influential evidence sources or from a broader shift across many prompts. IQAir's rise could be concentrated in a handful of high-visibility contexts or spread across the full question set. The public benchmark cannot distinguish between these explanations. A company-specific AI visibility audit maps prompt, surface, competitor, ranking, sentiment, and evidence-source patterns into a prioritized visibility strategy.

Request an AI visibility audit