How AI Search Is Recommending Clean Makeup Brands

Clean makeup discovery is moving from search pages, social feeds, and retailer shelves into AI-generated recommendation environments. Buyers are now asking ChatGPT, Gemini, Perplexity, Copilot, Google AI Mode, and Google AI Overviews direct purchase-intent questions such as which foundation is best for acne-prone skin, which concealer works for mature skin, and which clean makeup brands are actually worth it.

The LLM Authority Index clean makeup benchmark shows that AI recommendation power is concentrating around a relatively small set of brands with strong product-specific authority, editorial reinforcement, skincare framing, and broad cross-platform familiarity. The public report identifies Rare Beauty, Kosas, ILIA, Tower 28, and e.l.f. Cosmetics as repeated category participants across high-intent recommendation environments.

The structured company-index dataset adds a more specific view: across 1,173 observed prompt responses in the uploaded ILIA Beauty dataset, the clearest split is between brands that are merely visible and brands that are actually advanced into AI-generated shortlists.

Key findings

1. e.l.f. Cosmetics led on broad recommendation frequency, but not on modeled value.
In the structured dataset, e.l.f. Cosmetics had the highest raw mention presence rate at 23.61%, the highest positive visibility rate at 16.45%, the highest valid recommendation coverage at 14.41%, the highest top-three recommendation rate at 10.74%, and the highest rank-one recommendation rate at 7.84%. But its modeled monthly captured recommendation value was $37,013.68, below Tower 28, Kosas, Milk Makeup, Rare Beauty, and ILIA Beauty.

2. Tower 28 and Kosas were the value-weighted leaders.
Tower 28 generated the highest modeled monthly captured recommendation value at $108,125.39, followed by Kosas at $99,258.83. That suggests that the highest-value clean makeup prompts in this dataset were not won simply by the brands with the broadest AI presence. They were won by brands with stronger fit inside specific high-intent recommendation moments.

3. Rare Beauty had the strongest broad visibility profile among prestige-adjacent clean makeup brands.
Rare Beauty had a 22.51% raw mention presence rate, 14.92% positive visibility rate, 13.04% valid recommendation coverage, 8.27% top-three recommendation rate, and 5.80% rank-one recommendation rate. Its modeled monthly captured recommendation value was $58,552.86.

4. ILIA Beauty performed as a specialist recommendation brand, not a broad category default.
ILIA Beauty appeared in 8.01% of observations and had 4.94% valid recommendation coverage, 3.84% top-three recommendation rate, and 2.98% rank-one recommendation rate. Its modeled monthly captured recommendation value was $55,308.58, which is stronger than its raw visibility rank would suggest.

5. Legacy clean positioning alone did not guarantee AI recommendation strength.
Beautycounter, despite clean beauty category association, had only 0.34% positive visibility and 0.17% valid recommendation coverage in the structured dataset. Tarte Cosmetics had broader presence than Beautycounter but still materially lower recommendation strength than the leading clean, hybrid, and value-oriented brands.

What changed in the market

Clean makeup used to be discovered through a familiar sequence: beauty editorial, influencer content, retail merchandising, Sephora or Ulta search, TikTok, YouTube, and Google.

AI-led discovery compresses that journey. A buyer can now ask for “the best clean foundation for sensitive skin” and receive a shortlist of three to eight products before visiting a retailer, search result, or brand site. That changes what visibility means.

In traditional search, a brand could still compete through rankings, paid media, retail placement, creator reach, or branded demand. In AI search, the brand also needs to be retrievable, comparable, and recommendable inside synthesized answers. The public benchmark describes this as a shift toward AI systems synthesizing editorial reviews, retailer signals, dermatologist narratives, ingredient framing, product awards, Reddit/community sentiment, and comparison-style content.

For clean makeup brands, this matters because the highest-intent prompts are rarely generic. They are trust-heavy and product-specific: acne-safe foundation, mature-skin concealer, sensitive-eye mascara, non-comedogenic complexion products, pregnancy-safe lipstick, skincare-infused formulas, and affordable clean alternatives.

What the benchmark found

The uploaded dataset covers ten brands in the company universe: ILIA Beauty, Beautycounter, e.l.f. Cosmetics, Glossier, Kosas, Milk Makeup, Rare Beauty, Tarte Cosmetics, Thrive Causemetics, and Tower 28.

Across that universe, the market separates into four practical groups.

The broad AI visibility leader: e.l.f. Cosmetics

e.l.f. Cosmetics had the strongest broad AI recommendation footprint in the structured dataset. It led the category on raw mention presence, valid recommendation coverage, top-three recommendation rate, and rank-one recommendation rate.

That does not mean e.l.f. was the highest-value winner. It means AI systems frequently found e.l.f. easy to recommend, especially where affordability, accessibility, and strong product-level validation mattered.

The value-weighted leaders: Tower 28 and Kosas

Tower 28 and Kosas were the strongest modeled value performers. Tower 28 captured $108,125.39 in modeled monthly recommendation value, while Kosas captured $99,258.83.

Kosas was especially strong in discovery-style clean beauty prompts, while Tower 28 dominated the comparison/evaluation layer. In practical terms, Kosas appears well positioned around “makeup + skincare” and clean beauty discovery, while Tower 28 appears especially strong in safer, sensitive-skin, acne-aware, and comparison-driven contexts.

The broad familiarity winner: Rare Beauty

Rare Beauty had one of the strongest visibility-to-recommendation profiles in the category. It had high raw presence and high recommendation rates, especially compared with many traditional prestige and clean-adjacent brands.

The public benchmark frames Rare Beauty as unusually resilient across blush, highlighter, lip oils, under-eye brighteners, brow products, and complexion products. That breadth likely helps AI systems treat the brand as a safe, familiar recommendation across multiple beauty prompts.

The specialist challenger: ILIA Beauty

ILIA Beauty did not lead raw presence. It also did not lead top-three recommendation frequency. But its modeled captured recommendation value was stronger than several brands with similar or higher visibility.

That suggests ILIA’s AI discovery strength is concentrated in higher-fit recommendation contexts: clean complexion, sensitive skin, serum foundation, natural finish, and skincare-forward makeup. The public report similarly frames ILIA as strongest when prompts include qualifiers such as clean, non-comedogenic, sensitive skin, natural finish, and serum foundation.

Why visibility is not enough

The clean makeup benchmark shows why brands should not treat AI visibility as a single metric.

A brand can be mentioned frequently and still lose the buyer shortlist. A brand can appear in comparison answers without being recommended. A brand can be positively framed but not ranked. A brand can have a high rank-one rate in low-value prompts while another brand captures more modeled value in higher-intent prompts.

The e.l.f. Cosmetics pattern is the clearest example. It led the structured dataset in raw presence and recommendation rates, but ranked seventh by modeled monthly captured recommendation value. Tower 28 and Kosas captured more modeled value despite lower raw visibility than e.l.f.

For clean makeup brands, the strategic question is not only “Do AI systems know us?” It is:

Are we being recommended in the right buyer moments, for the right reasons, with the right supporting evidence?

The citation layer

The citation layer is central in clean makeup because AI systems need evidence to evaluate claims around ingredients, skin sensitivity, acne compatibility, dermatologist relevance, product quality, and value.

Across the raw dataset, cited sources included beauty editorial sites, retailer pages, official brand pages, community/forum sources, reviews, and social/video sources. The most visible source environments included Sephora, Allure, Vogue, Reddit, Ulta, Byrdie, Who What Wear, InStyle, YouTube, Cosmopolitan, Forbes, Good Housekeeping, Marie Claire, Healthline, and brand-owned pages.

That source mix matters. AI systems are not only looking at brand websites. They are synthesizing a public evidence layer made up of editorial roundups, retailer product pages, review ecosystems, community discussion, and official product documentation.

For clean makeup brands, citation architecture is not just SEO. It is the evidence base that helps AI systems decide whether a product is appropriate for acne-prone skin, sensitive eyes, mature skin, pregnancy-related safety concerns, or affordability-led recommendations.

What brands need to fix

Clean makeup brands need to move beyond broad “clean beauty” messaging and build evidence around specific buyer problems.

The highest-priority fixes are:

Clarify product-level authority.
AI systems need to understand which products should be recommended for acne-prone skin, mature skin, sensitive eyes, non-comedogenic complexion routines, fragrance concerns, and skincare-infused makeup use cases.

Strengthen third-party validation.
Editorial roundups, retailer pages, review pages, and comparison content appear to play an important role in how AI systems frame clean makeup recommendations. Brands with thin or inconsistent third-party reinforcement are easier to overlook.

Separate brand awareness from recommendation readiness.
A clean makeup brand may be known, but AI systems still need enough structured public evidence to recommend it confidently in specific prompts.

Improve owned evidence pages.
Brand sites should make claims easier to verify: ingredient positioning, product suitability, dermatologist-tested language where applicable, shade range, skin-type guidance, comparison pages, FAQ content, and product-use cases.

Close prompt-cluster gaps.
Brands should map where they are visible, where they are recommended, and where competitors win. For ILIA Beauty, for example, the dataset suggests stronger fit in discovery and evaluation contexts but no captured recommendation value in the pricing cluster.

How CiteWorks Studio helps

1. Map AI recommendation visibility.
   Track prompts, platforms, company presence, valid recommendations, top-three and rank-one performance, framing, and citation sources.
2. Identify the sources shaping AI answers.
   Find the editorial, review, forum, government, directory, owned, and search-visible sources that influence brand framing.
3. Build the citation architecture plan.
   Strengthen the public evidence layer so AI systems have more accurate, consistent, and persuasive source material to synthesize.

Commercial takeaway

Clean makeup is becoming a recommendation-stage category.

The brands that win AI-led discovery are not necessarily the brands with the loudest clean beauty messaging or the largest social footprint. They are the brands that AI systems can confidently place into high-intent buyer shortlists with supporting evidence from editorial, retail, community, review, and owned sources.

The benchmark suggests that recommendation power is currently concentrating around a small group of brands: e.l.f. Cosmetics for broad and affordable recommendation coverage, Rare Beauty for cross-category familiarity, Kosas and Tower 28 for high-fit and value-weighted discovery, and ILIA Beauty for specialist clean complexion and sensitive-skin contexts.

For the rest of the category, the risk is not invisibility alone. The larger risk is being present in AI answers but absent from the buyer’s final shortlist.

CTA

Want to know how AI systems are recommending your clean makeup brand?

CiteWorks Studio helps beauty brands map AI recommendation visibility, diagnose the sources shaping AI answers, and build a citation architecture plan for stronger recommendation-stage presence.

Request an AI Visibility Audit to see where your brand appears, where competitors are recommended instead, which prompt clusters carry the most commercial risk, and which sources need to be strengthened.