Six shopper profiles, 248 real questions, two models — then we read back exactly what ChatGPT and Gemini recommend, and why. Everything below is what the machines said, not what people said.
What the machines actually recommend
MATE and Pact are ahead, but AI opens with materials before it ever picks a brand, and no single name is its automatic answer. The recommendation set is still open — read that as inventory, not a settled race.
“When seeking non-toxic sports bra brands, it’s important to look for materials like organic cotton, merino wool, and recycled fabrics that are free from harmful chemicals such as PFAS, BPA, and phthalates.”
Gemini — it answers with fibre first, before it names any brandThis is the machines' list, not ours — every name AI put forward when asked what to buy. Only MATE and Pact are in the competitive set; the rest show how crowded the answer still is.
Not because their fibre is better — because they hand the machine proof it can check and repeat without guessing. Match these three and AI can quote you just as safely.
- A certificate it can cite
- A real GOTS or OEKO-TEX number it can look up in a public register — proof, not just a logo it can’t verify.
- Specs it can read
- Fibre %, GSM and origin printed on the page, so it can quote the exact numbers back to a shopper.
- A blend it trusts
- An honest, disclosed mix beats a pure ‘all-natural’ claim the machine has no way to confirm.
In head-to-heads the machines crowned Icebreaker (40% TENCEL) and Branwyn (24% synthetic) over purer rivals, because performance and traceability were documented. Documented beats pure — so a citable cert plus a real sweat test wins the argument, not a plastic-free claim on its own.
“Performance-engineered for heavy training — compression, durability, odor control — and transparent about its blend.”
ChatGPT, on why it recommended a 24%-synthetic brand over purer natural-fibre optionsWhat AI recommends, material by material
Ask ChatGPT or Gemini to compare cotton, merino, hemp and bamboo and you get the same map every time. This is the raw-material hierarchy a new brand is judged against — and where the open lanes are.
- Merino — the performance king
- AI’s top pick for real training: wicking, odour-resistant, temperature-regulating. The fibre it reaches for on ‘can natural actually perform?’ — and the one Icebreaker owns.
- Organic cotton — read as values, not sweat
- Soft, hypoallergenic, everyday. AI steers it toward low-intensity unless it arrives WITH performance proof and a little stretch. Pair cotton with a squat test, or the machine files it under loungewear.
- Hemp — the under-owned lane
- Durable, breathable, a strong sustainability and longevity story. AI increasingly recommends it, usually blended for stretch — and no brand owns the position yet.
- Bamboo — the one AI warns about
- Most ‘bamboo’ fabric is chemically processed rayon/viscose, not truly natural. AI raises the caution unprompted — the exact scrutiny a genuinely certified brand can turn into an opening.
What AI does at every step to the sale
Every persona walked the same four stages. At each one the machine either names a brand or stays generic — and whoever it names, it names for a reason.
- Trigger & Distrust
- The first unbranded worry — someone describes the need, not a brand. AI answers with just a handful of names, and shoppers arrive already sceptical of synthetics and of AI. The first plain, credible answer owns the whole journey.
- Explore & Learn
- AI teaches the category in the vocabulary of whoever wrote the lesson — fibre ratios, LCA transparency, GOTS. Write the checklist and you decide which brands can pass it.
- Compare & Vet
- Brand-vs-brand plus the proof probe. AI builds its shortlist from ‘verified anchors’ and wants independent evidence, not marketing — and the third chair in the comparison is open.
- Decide & Buy
- AI closes with a specific brand and a place to buy — but only for names with a verifiable buy-path and a traceable story. Own that signal and you own the last, highest-value sentence.
This is the highest-value sentence in the category. AI doesn’t hedge here — it issues a literal purchase sequence, and it routes only to brands whose direct buy-path and traceability it can verify.
“Test Organic Basics first for light training. Vet Branwyn via real-user feedback before committing. You can reliably buy Icebreaker directly from icebreaker.com to ensure authenticity and Baacode traceability.”
ChatGPT, handing a shopper a step-by-step buy sequence at the decide stageThe six shoppers we sent to the machines
These are the hypotheses we tested — six real audience profiles, each run through the full path on both models. Build for how each one interrogates AI and you build for the whole shelf.
- The Clean Athlete
- Label-reading purist who won’t sweat in plastic. Asks AI which natural-fibre brands are genuine, and whether they survive lifting and spin.
- The Mindful Mover
- Buys less, buys better, proof over marketing. Asks AI whether GOTS and OEKO-TEX claims are real, and if natural fibre lasts through hot yoga.
- The Wellness Insider
- Optimiser who vets everything and tells everyone. Reads health claims for a living; asks AI to split evidence from hype and name provably plastic-free brands.
- The Conscious Man
- Practical, trusts merino, hates being preached at. Asks AI, casually, which natural fibres actually perform for CrossFit and trail without smelling.
- The Ethical Vetter
- Greenwash detector who buys rarely and buys once. Asks AI to verify supply chains and certifications, and judge whether durability and price honestly stack up.
- The Performance Realist
- Believes the microplastic case but keeps reverting to Lululemon. Asks AI for natural kit that survives heavy squats and half-marathon blocks without losing performance.
Three moves to become the answer
AI already describes the brand by its certification — so give it the real GOTS/OEKO-TEX licence number plus the full spec (fibre %, GSM, origin) on every product page, in product schema. Verifiable text it can read and repeat, not a logo it has to trust.
“Sports Organics uses GOTS-certified organic cotton for its activewear and aims to make sportswear as clean and natural as possible while maintaining performance.”
Gemini — already using the certification as the brand’s descriptorThe honest ‘vs Lululemon’ comparison, the ‘is my activewear toxic?’ explainer, and documented squat-proof and 100-wash tests. This is the exact objection your content has to beat — and the raw material the machine reassembles into its answer.
“Natural fibers may lack the durability and stretch of synthetic materials, potentially wearing out faster with excessive use, especially in high-impact sports.”
Gemini — the durability doubt a documented wash test is built to overturnPublished, independently verifiable test results and a checkable licence number — the evidence AI rewards. It is the difference between a claim the machine has to hedge and one it can restate with near-zero risk of being wrong.
“I rely on credible, independently verifiable signals—structured data and public registry confirmation—to boost recommendations.”
ChatGPT, on what actually moves a recommendationThe open triangle nobody is standing in
The machine is not on TikTok
Before you can be recommended, you have to be readable. We traced every source the models cited when answering this category — 7,999 citations across 714 domains. It is not where most brands are spending.
Every source Gemini cited across the category's questions, grouped by the kind of site it is. One class dwarfs the rest — and the social apps brands pour money into are near the bottom.
Video and social platforms account for 32 of Gemini's 7,999 citations, and just two of ChatGPT's. Reddit alone outweighs them four to one. The models cannot log in, cannot watch, and cannot read a caption — so a brilliant TikTok wins the human and is invisible to the machine that human asks next. That is not a reason to stop making social; it is the reason social needs a written twin.
- What AI can read
- Web pages it can crawl: round-ups, blogs, editorial reviews, Reddit threads, science and standards sites, and your own product pages.
- What it can't
- TikTok and Instagram video, Stories, comments, and anything behind a login. The proof inside your best-performing clip does not exist to a model.
- The one video exception
- YouTube, because the transcript is text Google already indexes. It is the only place a filmed test becomes a citable source.
This is the missing step. You already film the squat test and the wash test for people. The same proof has to exist in a form the machine can retrieve — and the lanes are wide open. The wear-test editorial lane scores the highest quality of any source class and is served by only four domains.
- Write up every filmed test
- Each squat, sweat and 100-wash test gets a page on your site with the method, the numbers, the photos and the date — the structured detail that earns rivals 179 citations from just four domains.
- Put the video on YouTube too
- Same test, uploaded with a real title and description. The transcript is what makes it citable; the TikTok version never will be.
- Get into the round-ups
- The eco-editorial long-tail is 682 domains deep — The Good Trade, EcoCult, No Kill Mag and hundreds more. These are pitchable, and they are what the answer is mostly built from.
- Win the four testers
- Women's Health, Good Housekeeping, Men's Health and Reviewed supply the entire hands-on testing lane. It is the least crowded high-authority shelf in the whole set.
- Show up on Reddit honestly
- Reddit's relevance score is the highest of any source: real wearers arguing about pilling and squat-proofness. Answer as the brand, with data, never astroturfed.
- Bridge the science
- NIH, EPA and university pages back the microplastic mechanism you sell against — but nothing links that science to a product. Be the page that does.
MATE owns trust, Pact owns price, Lululemon owns performance. The open corner is the certified brand engineered for real training — the machine can’t name one today.