Where Does Your Supplement Rank in AI Recommendations?

As AI assistants become a more common source of product recommendations, CPG brands in the health and wellness space have a new arena to compete for credibility and prominence. Early adopters already put significant effort into creating content that large language models like ChatGPT and Gemini can use in ranking top-recommended products for end users.

On Sept. 9, AI Shelf Space published its Supplement AI Visibility Index, a free public ranking of which brands are most likely to be recommended across seven AI assistants: ChatGPT, Gemini, Google AI Overviews, Google AI Mode, Microsoft Copilot, Perplexity, and Claude.

The index tracks 185 supplement categories (“shelves”) and the current edition is based on the analysis of 178,786 answers sampled (all of which are read by the human eyes of AI Shelf Space’s team), with more than 1 million individual brand recommendations analyzed. Each shelf is sampled with a battery of prompts and hundreds of repetitions.

High sample rates are critical; AI assistants will answer the same question twice with two very different product rankings. In fact, AI Shelf Space found that fewer than 1% of 3,000 identical prompts returned identical brand lists. The company said other AI visibility tools often run a single check, providing little insight due to the unpredictability of AI assistant responses.

The Supplement AI Visibility Index will continue to synthesize feedback from AI recommendations over time to track changes in brand rankings across categories and keep rankings current.

“Repetition is key. Our public index is free, based on hundreds of repetitions per product category, and any brand can look up its own name to see whether it already has shelf space on any of the top 185 supplement categories at aishelfspace.com,” said Matias Garrido-Garcia, founder of AI Shelf Space.

Users of the Supplement AI Visibility Index can explore headline numbers, rankings for all tracked categories, and compare differences across each of the seven AI platforms included.

For more in-depth analysis, the platform also offers a full brand audit, with insights including one brand’s rank on all shelves across all AI assistants, and the public top five excluding one’s own brand, delivered as a private dashboard within two business days. The yearly intelligence package audits a complete year with quarterly reports, weekly monitoring with shelf space alerts, monthly optimization opportunities for PR and content, AI market intelligence briefings, two competitors tracked on every shelf, and overall brand perception.

In an interview with Nutraceuticals World, Garrido-Garcia discussed his top takeaways from the data, the methodology behind the data collection, commonalities among the top-ranked brands in AI suggestions, and more.

What Company Behaviors or Product Traits Get Attention from AI Assistants?

According to the index, just 25 brands accounted for 48% of all AI recommendations. Garrido-Garcia mentioned a few common threads shared among them, and why AI assistants seem to reward them. However, he noted that only AI model developers have insights into a model’s internal process. “What we can rank is what the assistants say and what they read.”

When it comes to a product’s attributes, Garrido-Garcia said value, budget, and affordability are mentioned in 70% of recommendations.

Second-most important is independent verification; third-party testing appeared in 60% of recommendations, followed by NSF (28%), USP (19%), or some other independent testing service (25%). Twenty of the 25 leaders publish third-party testing claims, Garrido-Garcia said, noting that AI assistants think and talk “like a conscientious buyer.”

Third-most important were formulation attributes, including organic (33%), vegan (31%), allergen-related statements (26%), and non-GMO (24%).

The fourth most important attribute was “clinically studied/clinically proven” (24%).

Manufacturing attributes were also a factor, including GMP (17%) and heavy metal testing (11%). “Made in USA” claims appeared in only 1.2% of recommendations.

Speaking on the commonalities among top-ranked brands, Garrido-Garcia said, “The first thread is almost boring: breadth. Nineteen of those 25 brands have products in more than 100 of the 185 product categories we measure, and only 24 brands in the whole dataset reach that mark. So a brand with a plainly named product for every measured category holds a ticket in almost every draw,” he said, noting NOW Foods and Thorne as key examples.

“However, this is not the whole picture, as many other important brands with very broad catalogs didn’t get as many AI recommendations.”

A thick written record is also important to AI visibility, Garrido-Garcia noted. Half of the list is natural-channel and practitioner brands with lots of literature, such as NOW Foods, Life Extension, Thorne, and Pure Encapsulations. The other half are younger and more online-native brands that are savvy about retailer listings, editorial reviews, testing directories, and online product pages. Brands that dominate brick-and-mortar retail shelves but lack online records overall represent a slim portion of AI recommendations.

“The lesson on digital footprint is that size matters less than location. By our classification, the assistants’ citations go 27% to editorial sites, 21% to retailers, 19% to affiliate review sites, and 18% to brands’ own sites.” Ads and social posts earn little credit, while reviews, retail listings, testing information, and clear product pages earn a lot.

Brands reputed as thought leaders on quality and identity standards can boost AI visibility. “On our six mushroom categories, between 70% and 97% of answers judge products on a single technical test: is the supplement made from the mushroom itself, or from mycelium grown on grain, and is the beta-glucan content disclosed? The brand that has spent years publishing about exactly that test, Real Mushrooms, is named in 98% of reishi answers. The assistants did not reward the loudest advertiser. They rewarded whoever wrote the exam.”

Sales performance holds noticeably weak influence on AI recommendations, with many lower-performing brands punching above their weight; recommendations read like those of a well-read, emotionally detached buyer who doesn’t take popularity into consideration, Garrido-Garcia noted.

“Nature Made is the exception that proves the rule: sixth overall and the leading brand on 11 categories … It is the one drugstore giant that carried its reputation across.”

Meanwhile, many brands with minimal or nonexistent brick-and-mortar footprints and small advertising budgets are mentioned surprisingly often, such as the 2012-founded, online-native brand Nutricost, ranked number five; Double Wood Supplements, founded in 2013, which lands 13th overall and leads spermidine with 93% of answers; and the 2016-founded Transparent Labs, the top pre-workout recommendation and 11th overall.

Specialist brands can dominate in a single category, Garrido-Garcia noted, with hair growth specialist Viviscal appearing in 99% of hair-growth answers, electrolyte experts LMNT appearing in 93% of electrolyte answers, and Ddrops in 93% of infant vitamin D drop answers.

“You can be everywhere, or you can own one shelf outright.”

Short-Term and Long-Term Plays

Garrido-Garcia noted that what works today in digital marketing might not work in the near future.

AI assistants today are still being trained on live web searches, so search engine optimization (SEO) strategies are still influential, meaning that brands need to balance AI visibility and SEO with equal effort. For that reason, products with straightforward names (ingredient, format, and dose in the name) stand out. “A brand whose magnesium is called something clever gets skipped on the magnesium shelf.”

But in the future, SEO will be replaced, he added. “AI agents are unemotional. They are not interested in your packaging, your persuasion copy, or your enchanting website. They recommend facts and on the proof behind the facts. So in the long run, marketing will be all about product design: build products that offer real value in quality, price, convenience, or any combination of them,” because AI agents are “almost blind to the ‘décor’ of shopping.”

Brands seeking to disrupt can perform better in AI recommendations by focusing on a few things, rather than trying to be a jack of all trades. “AI assistants are much better at unearthing that proof of real value, that product truth, than pre-AI Google ever was.”

When it comes to what’s easiest to win right now, recommendations related to health applications, and not ingredient types, are open ground that are up for grabs. None of the 40 condition-specific categories AI Shelf Space tracks have a unanimous leader; the most contested have up to 48 different brands across the top 70 slots.

Each AI assistant has a different “personality,” noted Garrido-Garcia. “ChatGPT reads like a cautious pharmacist: certification bodies, testing services, government fact sheets, and the brands’ own pages. Google’s AI Overviews and AI Mode read like a shopper with a magazine in hand: health publishers and retailer listings in roughly equal measure, plus the video and forum pages that no other assistant touches. Copilot reads the retailers, half of everything it cites. Gemini reads the health publishers, and Perplexity and Claude lean hard on review sites.”

Ultimately, Garrido-Garcia said that provable matters, but it also needs to be SEO-optimized, for as long as AI Assistants are trained on web search results. 

Variability

While AI Shelf Space uses a battery of different prompts phrased in ways that people ask for recommendations, “we do not attempt to capture the entire universe of prompts, simply because it is impossible.

“People already talk to assistants the way they talk to people, in long blurts and multi-turn conversations, not single, clean prompts, and soon the agent will not wait for a question at all. It will notice that Laura is pregnant, tell her that her old multivitamin is not the one for her now, and go find a safer alternative.”

That sort of layered decision-making and personalization can’t realistically be accounted for, he said.

Tracking day-to-day chatbot volatility based on prompts is more straightforward. “Exact percentages move between days. When we ran the same battery on the same category on two different days, a brand’s share of answers moved by 16 points on average, eight points on ChatGPT, and 22 on Gemini. However, ranks barely moved. The brand that led on Thursday led on Sunday, which is why the index ranks inside each category’s top five.”

Garrido-Garcia said that what surprised him most about developing this edition of the Supplement AI Visibility Index was how often different AI models disagreed with one another.

“Only eight in 185 categories had the same number-one across AI assistants, and as many as 194 different brands hold the 1,295 number-one spots in this edition. Within one assistant, the leaders repeat across hundreds of answers, so the divergence across assistants is not due to randomness but due to genuine differences of opinion.”

This is important, he said, in that it shows how quickly SEO will be phased out of the equation. “If AI visibility were just SEO, assistants that were presented the same search results would reach the same conclusions … Even Google’s two AI search products, fed by the same search engine, name the same number-one on barely more than half of the categories. The model’s own pre-existing knowledge and its own way of judging proof matter enormously.”