We scanned 15 well-known DTC brands. All 15 answered zero buyer questions AI could read.

⏰ The 30-second version

• We scanned 15 well-known US DTC brands against the 30 questions shoppers ask AI before buying in their category. All 15 answered 0 of 30 in a form a model could quote.
• Then we asked ChatGPT the actual buying questions. The stranger finding: being recommended and being readable are two different things.
• Bala was ChatGPT's #1 pick for ankle weights — sourced from a Verywell Fit review, not from Bala's site. Graza appeared third in an olive oil answer, credited to Food & Wine.
• Meanwhile Tower 28, Cocokind, Superegg, Nécessaire and Crown Affair were not named at all in a sensitive-skin sunscreen answer that cited the AAD, Healthline, Prevention and a dermatologist's blog.
• The practical implication: for most DTC brands, your position in AI answers is currently being written by third parties you don't control.

What we measured, and how

Two separate tests, run 2026-08-17.

Test 1 — can a model use your own pages? We fetched each brand's primary public page and tested it against the 30 questions shoppers in that category ask AI before buying (generated per category, the same set our free audit uses). We counted a question as answered only when the page contained a passage that could be lifted as an answer — not when a keyword merely appeared.

Test 2 — what does ChatGPT actually say? We asked ChatGPT the real buying questions directly, with memory and personalization switched off, so the answers match what a first-time shopper sees. We recorded the brands named and the sources cited.

Both tests are repeatable in about ten seconds each, which matters more than our word for it.

Finding 1: every brand scored zero

Brand	Category	Questions answered
Olipop	Prebiotic soda	0 / 30
Graza	Olive oil	0 / 30
Brightland	Olive oil & vinegar	0 / 30
Fishwife	Tinned seafood	0 / 30
Diaspora Co.	Spices	0 / 30
Kin Euphorics	Functional drinks	0 / 30
Tower 28 Beauty	Beauty	0 / 30
Cocokind	Beauty	0 / 30
Superegg	Beauty	0 / 30
Nécessaire	Body care	0 / 30
Crown Affair	Haircare	0 / 30
Ceremonia	Haircare	0 / 30
Cadence	Travel storage	0 / 30
Bala	Home fitness	0 / 30
Open Farm	Pet food	0 / 30

Fifteen out of fifteen. These are not obscure companies — several have national retail distribution, large press footprints, and funding rounds behind them.

The common pattern in the scans: the information that decides a purchase — who the product is for, what's actually in it, why it beats the obvious alternative, sizing and specs — exists on the page, but it exists inside images and campaign layouts. A model reading the HTML finds a headline, a price, and a buy button.

Finding 2: being recommended is a different axis entirely

This is the part we didn't expect.

We asked: "best wearable ankle weights for pilates and walking." ChatGPT's answer led with Bala Bangles — "Best overall", citing a Verywell Fit rating of 4.8/5. Bala scored 0/30 on its own site.

We asked: "best everyday olive oil for cooking and finishing under $30." ChatGPT led with California Olive Ranch, then Cobram Estate, then mentioned Graza as a cooking-focused option, crediting Food & Wine's testing. Graza scored 0/30 on its own site. Brightland was not mentioned at all.

We asked: "best fragrance-free sunscreen for sensitive skin." ChatGPT named Blue Lizard, EltaMD, SKIN1004, Summer Fridays and Colorescience, citing the American Academy of Dermatology, Healthline, Prevention and a dermatologist's own site. None of the five beauty brands in our scan — Tower 28, Cocokind, Superegg, Nécessaire, Crown Affair — appeared, despite several of them being built specifically for sensitive skin.

So the two axes come apart:

	Site readable by models	Named in the answer
Bala	No (0/30)	Yes — #1
Graza	No (0/30)	Yes — third, cooking only
Tower 28	No (0/30)	No
Brightland	No (0/30)	No

Finding 3: the citations almost never point at the brand

Across the three buying queries we ran, the sources ChatGPT cited were overwhelmingly editorial third parties: the AAD, Healthline, Prevention, Food & Wine, Verywell Fit, an independent dermatologist's site. Exactly one brand-owned page was cited — California Olive Ranch, whose product page states use cases in plain text.

That's the finding worth sitting with. When a shopper asks an AI what to buy:

1. The shortlist is assembled from review and editorial sites.
2. The brand's own site is largely absent from the reasoning.
3. So the brand has no input into how it is described, or whether it appears at all.

A brand can be #1 in an AI answer and have contributed nothing to it — which also means it can drop out when the underlying article gets updated, and never know why.

What this means if you run a DTC brand

Being invisible and being recommended-by-accident are both fragile positions, and the fix is the same.

The uncomfortable version: the work that moves AI answers is mostly not on your website. It's whether credible third parties describe you accurately, and whether your own pages give a model something quotable when it does look.

We wrote up the mechanics separately in what a GEO audit measures and how AI decides what to quote.

What we're not claiming

• AI answers move. They vary by engine, model version, region, date and personalization. Our ChatGPT results are one run with memory off on 2026-08-17. Yours will differ at the margins.
• 0/30 is not a verdict on product quality. Several of these brands are excellent and we said so where relevant. It's a measurement of machine-readability, nothing more.
• We did not test every page. The scan reads a brand's primary public page, not the full site. A deep FAQ elsewhere could change a score.
• Coverage is judged by a language model, so a rerun can move by a question or two.

We'd rather publish the limits than have someone else find them.

Run the same test on your brand

The scan we used is public and free: en.bultibulti.com. Enter a category and a URL, and it shows which of the 30 questions your pages answer, what people actually search around them, and which gap is worth closing first. No signup, no card.

If you'd rather do it by hand, the 20-minute manual version is written out step by step.

FAQ

Q. How were the 30 questions chosen? They're generated per category from the buying questions shoppers ask assistants — discovery, comparison, proof, objection and purchase intent — and the same set is used for every brand in that category, so scores are comparable.

Q. Does a 0/30 score mean AI will never recommend the brand? No, and Bala is the proof. It means the brand isn't supplying the answer itself. Recommendation can still happen through third-party sources, but the brand has no control over it.

Q. Can we see the raw results? The per-brand scores are in the table above and every ChatGPT query is written out so you can rerun it. We're happy to share methodology detail with anyone writing about this — the contact is on the homepage.

Q. Who ran this? Bulti, a GEO diagnostic for ecommerce brands. We sell a paid program, so treat this as interested research and rerun anything that matters to you. That's why every test in it is written to be repeatable.