How AI Search Is Recommending Roofing Companies: Monthly Trends

Executive Summary

GAF and Owens Corning are tied for the lead in August 2026, each posting 51.6% valid recommendation coverage on 610 qualified observations. GAF moved up from 48.4% in July and Owens Corning moved up from 47.6% in July; both changes fall within normal month-to-month variation for the category, and neither is flagged as a significant mover in the underlying data. The more consequential story this month came from two significant decliners, not from the tied leaders.

CertainTeed's valid recommendation coverage fell from 44.1% in July to zero in August, a change flagged as significant in the current data. Malarkey moved the same way, from 35.1% in July to zero in August, also flagged as significant. These are the two largest baseline-to-current movements in the category this month and represent a real change in which manufacturers appear in the valid recommendation set, not a rounding artifact.

The remaining brands posted modest gains that stay within normal month-to-month variation: Atlas Roofing rose from 20.4% to 23.1% on 141 valid recommendations, Erie Home and Power Home Remodeling both reached 3.8% coverage on 23 valid recommendations each, IKO edged from 12.0% to 13.0%, and TAMKO held effectively flat at 10.2%. Taken together, the current benchmark shows the top two brands holding steady at the front of the field while the category reorganizes beneath them following CertainTeed's and Malarkey's declines.

Each monthly benchmark run begins with 800 prompt-surface observations across the defined AI/search surface universe. July's run produced 455 unique questions; August's produced 515. All 800 prompts in each month mentioned a tracked brand or competitor. In July, 783 of those prompts were relevant and 17 were irrelevant; in August, 656 were relevant and 144 were irrelevant. The public benchmark is built on 533 qualified observations from July and 610 from August.

Key Findings

Benchmark Context

Research-Scope Funnel

The table below separates the raw collection universe from the qualified analysis set. Brand-level recommendation percentages are calculated within the qualified benchmark set, not the full prompt pool.

The August run added a surface to the qualified set alongside ChatGPT, Copilot, Gemini, AI Mode, AI Overviews, and Perplexity. The increase in qualified observations from 533 to 610 reflects this broader surface coverage, not a change in the underlying prompt pool size.

Benchmark-Level Metrics

AI Recommendation Trend

The Top Two Are Tied at the Front, and a Significant Change Occurred Below Them

The most consequential finding in August is not the leadership tie but the disappearance of CertainTeed and Malarkey from the valid recommendation set. Both brands registered zero valid recommendation coverage in August after holding 44.1% and 35.1% respectively in July. This is a significant category-level change that reshapes the competitive map.

The category-level change in August came primarily from these two declines. CertainTeed's and Malarkey's drops exceeded normal month-to-month variation, while the gains among the stable risers remained within it. The signal this month is in the losses, not the gains.

What Changed This Month

GAF and Owens Corning: A Tie at the Top, Both Steady

GAF's valid recommendation coverage moved from 48.4% in July (258 valid recommendations on 533 observations) to 51.6% in August (315 valid recommendations on 610 observations). Owens Corning moved from 47.6% in July (254 valid recommendations) to 51.6% in August (315 valid recommendations). Both moves remain within normal month-to-month variation, meaning the tie reflects parallel drift rather than a decisive break between the two.

The composition of those moves differs. Owens Corning's top-three rate rose from 40.9% to 42.8% and its rank-one rate from 7.1% to 9.0%, suggesting the brand is being recommended somewhat more prominently. GAF's top-three rate rose slightly from 42.8% to 43.9%, but its rank-one rate fell from 28.7% to 24.9%, meaning GAF is appearing in more recommendation sets but in fewer first positions.

Both brands are present in roughly 92% to 97% of observations, so the story is not about visibility but about which position and phrasing AI systems use once a brand is named. Highest-priority diagnostic: which prompt types and surfaces moved Owens Corning into rank-one more often while GAF's rank-one share declined, and what evidence sources those surfaces cite.

CertainTeed: From Frequent Presence to Absent From the Recommendation Set

CertainTeed's valid recommendation coverage fell from 44.1% in July (235 valid recommendations on 533 observations) to 0.0% in August on 610 observations. This decline is the largest single movement in the category and is flagged as significant in the current data.

The brand's raw mention presence rate was 87.1% in July, meaning it was visible in most AI responses that month. The August data reports no valid recommendation coverage and no raw presence figures for the brand, indicating the change is concentrated in the qualified recommendation set specifically, not necessarily in all mention of the company.

This is the distinction between visible and recommended: a brand can be named in AI responses without being surfaced as a valid recommendation in the qualified set. The data does not establish whether CertainTeed is being discussed in contexts that no longer produce recommendation credit or is simply appearing less. Highest-priority diagnostic: which prompts previously produced CertainTeed recommendations now yield a different brand or no recommendation at all, and what sources those answers draw from.

Malarkey: A Sharp Change After a Strong July

Malarkey's valid recommendation coverage fell from 35.1% in July (187 valid recommendations on 533 observations) to 0.0% in August on 610 observations. This decline is flagged as significant in the current data and is the category's second-largest movement.

In July, Malarkey had a raw mention presence rate of 50.3% and a top-three recommendation rate of 5.8%, with a notably high net sentiment score of 0.9. The August data shows zero valid recommendation coverage, with no supporting presence figures reported for the brand.

The pattern here differs from CertainTeed's: Malarkey was a mid-tier brand with positive sentiment in July that still rarely reached the top three. Its move to zero suggests the contexts in which it was previously recommended now produce different outcomes. The benchmark cannot distinguish platform behavior from measurement effects here. Highest-priority diagnostic: whether this is a surface-specific pattern, a shift in which prompts are being asked, or a change in the sources AI systems draw from when discussing metal and specialty roofing options.

Atlas Roofing: Quiet Mid-Tier Gains

Atlas Roofing rose from 20.4% in July (109 valid recommendations on 533 observations) to 23.1% in August (141 valid recommendations on 610 observations), a gain that remains within normal month-to-month variation.

The brand's raw presence rate held roughly steady, moving from 37.0% in July to 36.1% in August. Its top-three rate rose from 0.8% to 1.3%, and its rank-one rate remained at zero. Atlas Roofing is being named in more recommendation contexts, but almost never as a top-three or first recommendation.

This is the visible-versus-recommended distinction in the other direction: the brand's presence is high relative to its recommendation position, meaning it is discussed but not prioritized. Highest-priority diagnostic: which surfaces and prompts name Atlas Roofing without recommending it, and what evidence would move it into a top-three position.

Buyer-Intent Interpretation

In both July and August, all qualified observations fell into the Brand Recommendation cluster. No qualified observations landed in Pricing & Value or Multi-Brand Comparison, meaning the public benchmark can currently answer which brand AI systems recommend first, but not how those systems position brands on price, value, or head-to-head comparison.

The recommendation-shaped answer share rising from 24.0% to 29.3% suggests AI systems are being asked more directly for recommendations, but the absence of pricing and comparison observations means the commercial picture is still partial. Buyers comparing roofing brands on cost are a segment this benchmark does not yet cover.

Brand Opportunity Summary

The benchmark identifies where attention is warranted; a company-level analysis is needed to explain why.

Evidence Layer

The aggregate metrics are built from prompt-level observations capturing the query, the AI surface, the recommendation outcome, the ranking position, sentiment, and citations where exposed. Company-level analysis can go deeper into patterns by prompt, competitor, surface, and evidence source. Source presence in an AI response is not automatically treated as proof of causation; it is a signal that the source shaped the answer, not that it caused the recommendation outcome.

Interpretation Notes

• Small-count movements: DECRA's 0.2% coverage rests on 1 valid recommendation in August; Erie Home and Power Home Remodeling each rest on 23. These figures are directionally indicative, not statistically robust.
• Qualified denominator vs. raw collection: percentages are calculated against the qualified benchmark set (610 in August), not the 800 raw prompts, so changes in the qualified set can affect rates even when raw prompt counts are stable.
• The benchmark identifies where change occurred; it does not by itself establish cause. CertainTeed's and Malarkey's declines may reflect prompt mix, surface coverage changes, or source shifts, and the data alone cannot distinguish among them.

About This Benchmark

This report is part of the CiteWorks Studio AI Industry Market Discovery research program.

• AI Industry Market Discovery
• AI Industry Market Discovery Methodology
• AI Industry Market Discovery Metrics
• AI Industry Market Discovery Standards

Next Step

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

The aggregate percentages in this report do not reveal which high-intent prompts are won or lost, which competitor takes the recommendation when a brand loses, what attributes AI systems associate with each roofing option, or which external sources shape those answers. A company-specific audit answers those questions at the prompt level, tracing each surface's response back to the evidence sources that produced it.

A company-level AI visibility audit maps those prompt, surface, competitor, ranking, sentiment, and evidence-source patterns into a prioritized visibility strategy. It identifies which gaps are addressable through owned content, which require third-party source cultivation, and which surfaces deserve attention first.

Request an AI visibility audit