Research · Published Oct 7, 2026 · Data from Oct 6, 2026
We checked 21 online brands. An AI model named 19 of them in none of its answers.
More shoppers now ask an AI assistant what to buy instead of searching. We wanted to see who it recommends, so we took 21 real online brands, wrote the 10 questions a shopper in each category would ask, and put them to one AI model.
Small sample. 21 brands, one AI model, one run each, on Oct 6, 2026. This describes what happened to these 21 brands in that run. It is not a measure of what every AI assistant does.
Key findings
- Across all 210 answers, a brand we checked was named in 4 (about 2%). With only 21 brands the 95% interval on “share of brands named” is wide: 2 of 21 is roughly 3% to 29%.
- The answers went mostly to brands everyone already knows: Lululemon and Gymshark for leggings, Lodge and All-Clad for cookware, Patagonia and Icebreaker for wool.
- The 2 brands that were named showed up on specific, problem-shaped questions, not on generic “best X” ones.
- In 6 cases a direct competitor was named while the brand we checked was not, so it was not only the giants that were named.
The results, brand by brand (anonymised)
Grey: brand named in 0 of 10 answers (19). Purple: brand named in 2 of 10 answers (2). n = 21 brands.
| Category | Brand named | Competitor named |
|---|---|---|
| Baby sleepwear | 2/10 | 0/10 |
| Hunting packs | 2/10 | 1/10 |
| Chai | 0/10 | 2/10 |
| Skincare | 0/10 | 2/10 |
| Home fragrance | 0/10 | 2/10 |
| Streetwear | 0/10 | 1/10 |
| Outdoor wool apparel | 0/10 | 1/10 |
| Baby care | 0/10 | 1/10 |
| Activewear | 0/10 | 0/10 |
| Cookware | 0/10 | 0/10 |
| Kids clothing | 0/10 | 0/10 |
| Cashmere | 0/10 | 0/10 |
| Hemp accessories | 0/10 | 0/10 |
| Beauty-professional supplies | 0/10 | 0/10 |
| Home goods | 0/10 | 0/10 |
| Garden centre | 0/10 | 0/10 |
| Golf equipment | 0/10 | 0/10 |
| Pool accessories | 0/10 | 0/10 |
| Seafood | 0/10 | 0/10 |
| Bathroom fixtures | 0/10 | 0/10 |
| Electric mobility | 0/10 | 0/10 |
Who the AI named instead
For seven of the categories we kept the brands the AI named most across the 10 answers (extracted automatically from the answer text). These are well-known brands the AI recommended, not the brands we checked.
| Category | Brands the AI named |
|---|---|
| Activewear | CRZ YOGA, Lululemon, Athleta, Gymshark |
| Cookware | Lodge, Nordic Ware, Silpat, Caraway, All-Clad, Made In |
| Chai | The Chai Box, Wagh Bakri, Society Tea, Vahdam, Twinings, Bigelow A similar-size competitor was named 2 of 10 times; the brand we checked, 0 of 10. |
| Outdoor wool apparel | Icebreaker, Smartwool, Patagonia, Ibex A direct competitor was named 1 of 10 times; the brand we checked, 0 of 10. |
| Kids clothing | J.Crew Crewcuts, Carter’s, Cat & Jack, Old Navy |
| Cashmere | Naadam, Quince, Vince, Everlane, Jenni Kayne |
| Skincare | Elemis, DHC, Clinique, La Roche-Posay, The Ordinary A direct competitor (Elemis) was named 2 of 10 times; the brand we checked, 0 of 10. |
The two brands that were named
Both were named only on narrow questions that describe a problem. These are the four questions, word for word:
Baby sleepwear (named 2 of 10)
“What should I buy for a baby who constantly breaks out of traditional swaddles?”
“What is the best swaddle for a newborn who prefers sleeping with arms up?”
Hunting packs (named 2 of 10)
“Which pack design offers the best quick-access rifle storage for mountain hunting?”
“What should I get for a multi-day precision shooting competition requiring heavy gear transport?”
Two brands is too few to call a pattern. It is the one thing in this data that points somewhere useful, so we are publishing it as a lead to test, not as a finding.
Method
- Brands. 21 online brands across the categories above.
- Questions. 10 per brand, written as a shopper in that category would ask an AI assistant: “best X for…”, “what should I get for…”, with the use case, budget and experience level varied. Each set comes from a curated category question bank where one exists, or is written by an AI model, and never names the brand being checked.
- Model. Every question went to the same AI model (Anthropic’s Claude Haiku 4.5, a small, fast model) once, as a plain shopper question. Answers were capped at about 400 tokens.
- Scoring. An answer counts as naming the brand when its text contains the brand name or the brand’s domain name. We did the same for one direct competitor per brand. Nothing is judged for tone or accuracy in this study.
Example questions
- “What should I get for a beginner looking for affordable, squat-proof leggings?”
- “What is the best chai blend for someone who prefers a strong, spicy kick?”
- “Which chai blend pairs best with oat milk for a creamy texture?”
- “What is the best non-stick pan for a beginner learning to sear steaks?”
- “What is the best investment piece for someone building a high-quality capsule wardrobe from scratch?”
Want the full question set to test your own category? Email hello@arenza.ai and we will send it.
Caveats
- Small sample. 21 brands, one run each, on one day.
- One model. Other assistants, and the same model with extra features such as web search, may name different brands.
- Capped answers. Answers were limited in length, so a brand mentioned late in a longer answer could have been missed.
- Name matching. We match on names and domains. A brand referred to by a nickname or product name alone may not be counted.
- Not about products. This study does not look at prices, links, what shoppers see on screen, or sales. It measures only whether a brand was named.
What this means for a small Shopify brand
- Check your own category first. Ten questions and ten minutes will tell you more about your store than any average.
- Start from your customers’ problems. The two brands that were named were named on problem questions. Pick the three problems customers describe in reviews and messages, and make sure your product pages answer them in the same plain words.
- Put the facts in text. Size, materials, price and who the product is for belong in plain text on the page, not only in images.
- Look at competitors, too. In 6 of 21 cases a direct competitor was named when the brand was not. That is worth seeing for your own category.
FAQ
How many brands did the study cover, and which AI model?
We checked 21 online brands on October 6, 2026, with 10 shopper questions each (210 answers), using one AI model, one run per brand. The results describe those 21 brands on that day, not every brand or every AI assistant.
How many of the 21 brands did the AI name?
19 of 21 were not named in any of their 10 answers. 2 were named in 2 of 10 answers each, both times on narrow problem questions. Across all 210 answers, a brand we checked was named in 4.
Does this prove AI assistants ignore small brands?
No. It is a small sample from one model and one run per brand. With 2 of 21 brands named, the 95% interval for the true share is wide (about 3% to 29%). It is a starting point for checking your own category, not a general rule.
Can I run the same test on my own store?
Yes. Write the 10 questions a shopper in your category would ask without naming your brand, put them to an AI assistant, and record which brands it names. The questions we used are available on request at hello@arenza.ai.
See what an AI model says about your store
Arenza is a Shopify app that asks AI assistants the questions your shoppers ask and shows you whether your store was named.
Related: how Arenza scores AI answers · Arenza guides. Data: aggregate JSON (no brand names).