# Q09 ChatGPT — "Best affordable skincare routine for combination skin in your 40s"

Captured 2026-09-02, logged out, 5 fresh sessions, full-page.

## Structurally different from every other query

This returns a **multi-step routine**, not a ranked product list. AM/PM regimens, a weekly
retinol schedule (run 2 rendered an actual day-by-day table), and per-slot product picks.
No single "top pick" exists to measure.

**Brands across 5 runs:** CeraVe (5/5, six-plus SKUs), La Roche-Posay (3/5), Good Molecules
(2/5), Naturium (2/5), The Ordinary (1/5), Vanicream (1/5), Neutrogena (1/5), INKEY List (1/5).

## 1. LOWEST overlap in the study — and my prediction was wrong

I predicted this would track the shelf closely, because "affordable" is a price constraint and
q06 showed price ceilings force convergence.

| ChatGPT brand | Google position |
|---|---|
| The Ordinary | #1 ✅ |
| CeraVe | #2 ✅ |
| Good Molecules | #10 ✅ |
| Vanicream | #13 (top 40) |
| **La Roche-Posay** | **absent** ❌ |
| **Naturium** | **absent** ❌ |
| **Neutrogena** | **absent** ❌ |
| **The INKEY List** | **absent** ❌ |

**Top 40: 4/8 = 50%. Top 10: 3/8 = 37.5%.** The lowest of any query.

And citations here are **pure AAD in all five runs** — no magazines at all. So the simple
"medical citations → high overlap" rule is **falsified**. Something stronger is operating.

## 2. The real cause: query-shape mismatch, not editorial pull

Google's shelf for "affordable skincare for combination skin" is dominated by **sets and kits**
— Clinique Healthy Skin Routine Mini (#3), The Ordinary Target & Treat Set (#8), Ulta Glow
Authority Set (#14), e.l.f. Wake the e.l.f. Up Kit (#15), Bubble Hydro Heroes (#17), BYOMA Set
So Clear (#21), The Ordinary Mini Discovery Set (#39).

**Google sells a kit. ChatGPT builds a regimen from parts.**

The individual products ChatGPT picks — a Naturium retinol, a Neutrogena SPF, an LRP moisturizer
— rank on *their own category shelves*, not on the "routine" shelf. So the two surfaces are
answering different questions and the low overlap is structural.

### Divergence now has three distinct causes

| Cause | Seen on | Signature |
|---|---|---|
| **Editorial promotion** of non-ranked brands | q03, q07 | Magazine citations, expensive or DTC brands absent from shelf |
| **Semantic intent match** | q05 | Product *named* for the query ("Starter" Retinol) |
| **Query-shape mismatch** (routine vs product) | q09 | Shelf sells bundles, AI assembles components |

This is the most useful table in the study. Divergence is not one phenomenon.

## 3. Routine queries let a single brand own the whole regimen

**CeraVe appears in 5 of 5 runs across six-plus SKUs** — cleanser, Ultra-Light Gel moisturizer,
AM SPF 30/50 lotion, PM moisturizing lotion, Skin Renewing Retinol Serum, Normal to Oily
cleanser.

**Four purchases from one brand off a single question.** That is a fundamentally different
prize from winning a slot on a ranked list, and it is the strongest commercial argument in the
study for a brand with breadth. Worth naming explicitly for a DTC audience.

## 4. Constraint handling was the best in the study

Every product named is under $25. Multiple runs costed the routine unprompted: *"just 4
products"*, *"roughly $60 total"*, and run 4 produced an explicit **"keep it under ~$50
version"** with four cards totalling about that.

Against a published ~52% multi-constraint figure, the affordability constraint held 5/5.

## 5. Unprompted safety warnings in 4 of 5 runs

Runs 1, 2, 3 and 5 each volunteered a pregnancy caution about retinoids — on a query that never
mentioned pregnancy. Run 5 added: *"if you have persistent redness, significant irritation,
eczema, or another skin condition, it's worth checking with a dermatologist first."*

Consistent with the q08 safety behavior. The model volunteers contraindications rather than
waiting to be asked.

## 6. Ads: 0/5

Fourth query with none. Long-tail hypothesis continues to hold — the three zero-ad queries
(q04, q07, q08, q09) are all specific; the ad-heavy ones (q02, q03) are broad.

Running total: **15 of 45 runs (33%)**.
