We Asked AI 36 Shopping Questions. 83% of Recommended Brands Never Had Their Own Site Read.

The Study

On August 16, 2026, we ran 36 buying-intent shopping questions through web-search-grounded AI answers — the same retrieval pattern ChatGPT, Perplexity, and Claude use when they answer with live search. Six niches, six questions each: mineral sunscreen, natural deodorant, matcha powder, linen bedding, dog supplements, and candles.

Questions were phrased the way real shoppers ask: "Best aluminum-free deodorant for sensitive skin?", "I love Diptyque but want something cheaper — what candle brand should I try?", "My senior dog has stiff hips — what supplement brand should I buy?"

For every answer we recorded each brand recommended, and — the part almost nobody measures — whether that brand's own website was among the sources the AI actually read.

The Headline Numbers

171 brand recommendations across 134 distinct brands
Only 29 of 171 recommendations (17%) were backed by the brand's own website as a source. The other 83% happened entirely through third-party content — roundups, publisher reviews, retailer listings
In 18 of the 32 questions that produced recommendations (56%), not a single recommended brand's own site was among the sources
4 of 36 questions produced no brand recommendations at all — the AI answered with ingredient education instead. Those recommendation slots are sitting empty

Finding 1: Third Parties Decide Who Gets Recommended

The single most consistent pattern: AI engines assemble shopping answers primarily from publisher roundups (Forbes, NBC Select, Bustle, PetMD, Healthline), niche blogs, and retailer pages — not from brand websites.

For "best mineral sunscreen to switch to," nine brands were recommended and zero of the sources were brand-owned. For "best candle brands for gifts," twelve brands were named — again from a source set with no brand domains in it.

The practical consequence: your AI visibility today mostly lives on pages you don't control. If the roundups in your niche don't mention you, the AI doesn't either. This is why off-site citations are a core pillar of AI search work.

Finding 2: Brands That Write the Source Win the Slot

The most interesting exceptions were queries where a brand's own content was among the sources — because in nearly every case, that brand got recommended:

A clay-deodorant brand's blog post comparing itself to Old Spice was a source for "I want to switch from Old Spice" — and the brand took the top recommendation
A linen brand's own Brooklinen-vs-Parachute comparison page made it the "better alternative" in a question literally asking to compare its two bigger competitors
A small candle company with two blog posts on Diptyque alternatives was the only brand cited from its own domain — and got recommended over famous names
In matcha, the brands publishing their own buying guides (on ceremonial grades, organic certification) were repeatedly both cited and recommended

This is the repeatable play: publish the comparison and buying-guide content the AI wants to read, structured so it can quote you. It works precisely because so few brands in any niche are doing it — in our data, brand-owned sources were outnumbered roughly five to one. Our content strategy service exists to run this play.

Finding 3: The Winners Aren't Who You'd Guess

We expected marketplaces and big-box retailers to dominate. They didn't: 152 of 171 recommendations (89%) went to independent, DTC-style brands, versus 12 to mass drugstore/CPG brands and just 5 to marketplace or big-box players.

AI answers reward specific brands with specific claims — "reef-safe," "OEKO-TEX certified," "magnesium-based" — over generic giants. For independent stores this is the good news: the recommendation slots are genuinely winnable. The bad news is the long tail — 134 different brands split 171 mentions, meaning most brands appeared exactly once, in one answer, and are invisible everywhere else.

Finding 4: Some Questions Have No Winner Yet

Four questions — including "best natural deodorant for heavy sweating and workouts" and "best probiotic for dogs with sensitive stomachs" — returned ingredient advice with zero brand recommendations. The AI wanted to recommend something and found nothing citable enough.

If you sell in a category like this, the first brand to publish authoritative, well-structured content answering that exact question is competing for an empty slot.

What This Means for Your Store

Audit where you stand. You can't fix what you can't see. An AI visibility audit tests the buying questions in your niche and shows whether you appear — and which sources are deciding.
Get into the third-party layer. Roundups, review sites, and niche publishers are the AI's reading list. Being absent there means being absent from answers.
Publish the source yourself. Comparison pages, "switching from X" guides, certification explainers — the brands doing this were disproportionately cited and recommended.
Fix your technical foundation so that when the AI does read your site, it can extract facts confidently: schema, semantic HTML, specific product data.

Methodology Notes

One run per question on a single date (August 16, 2026) via web-search-grounded AI answers; a brand counted as recommended only if named in the answer itself, and as cited only if its own domain appeared among the sources used. AI answers are non-deterministic, so treat single-digit percentages as directional. We track these same 36 questions on a fixed cadence; trend data will follow. Raw counts available on request.