A serum sold as "Crystal Retinal" lists retinal 37th of its 43 ingredients, 26 places after phenoxyethanol, a preservative that is usually used at 1% or less. Skincare forums have a rule of thumb for that: anything listed after phenoxyethanol is probably at 1% or less too. So is the hero ingredient in a best-selling serum usually near the top of the list, or near the bottom? And how often can you even run that test?
I pulled the ingredient lists of 45 best-selling face serums on Sephora US with Python on 10 November 2026 and parsed them. The short version: vitamin E is in 67% of them and vitamin C in 29%. The ingredient a product names in its title is among the first five in only 22 of 57 cases. And the phenoxyethanol test only applies to 21 of the 45 serums, because the other 24 do not contain phenoxyethanol at all.
The data
The Sephora API returns Sephora US catalogue data as JSON, which makes it a workable skincare ingredients API: /product-details carries the full ingredient text of every size in currentSku.ingredientDesc and regularChildSkus[].ingredientDesc, next to the product name, price, rating and the highlights badges ("Fragrance Free", "Niacinamide").
I made one call to /search-by-category with categoryId=cat60103 (Face Serums, 491 products), sortBy=BEST_SELLING and pageSize=60. From those 60 I dropped a mini size, an eye serum and an eye and lip serum, then called /product-details for the first 45 of the rest. All 45 had a readable ingredient list. The median serum costs $38 at its smallest size.
An ingredient list on a US cosmetic label follows a rule set in 21 CFR 701.3: ingredients above 1% in descending order, then ingredients at 1% or less in any order, then colour additives in any order. That is what makes position worth measuring, and also what limits it.
The code
The script below takes a product id and the ingredient its name advertises, pulls the ingredient list and says where that ingredient sits relative to phenoxyethanol. Put your RapidAPI key in the RAPIDAPI_KEY environment variable and it runs unchanged.
import html
import json
import os
import re
import requests
HOST = "real-time-sephora-api.p.rapidapi.com" # Sephora API on RapidAPI
BASE = os.environ.get("API_BASE", f"https://{HOST}")
def headers():
if "API_BASE" in os.environ:
return json.loads(os.environ.get("API_HEADERS", "{}"))
return {"x-rapidapi-host": HOST, "x-rapidapi-key": os.environ["RAPIDAPI_KEY"]}
def split_inci(text):
"""Split on commas outside brackets, but not inside names like 1,2-Hexanediol."""
parts, buf, depth = [], "", 0
for i, ch in enumerate(text):
if ch in "([":
depth += 1
elif ch in ")]":
depth -= 1
between_digits = text[i - 1:i].isdigit() and text[i + 1:i + 2].isdigit()
if ch == "," and depth == 0 and not between_digits:
parts.append(buf)
buf = ""
else:
buf += ch
parts.append(buf)
return [p.strip(" .*").lower() for p in parts if p.strip(" .*")]
def inci_list(ingredient_desc):
"""Skip the '-' marketing bullets and return the first comma-heavy line as a list."""
text = re.sub(r"(?i)<br\s*/?>|</p>", "\n", ingredient_desc or "")
text = html.unescape(re.sub(r"<[^>]+>", "", text))
text = text.replace("", "").replace("", "") # soft hyphens and BOMs hide in some lists
for line in (ln.strip() for ln in text.split("\n")):
if line and not line.startswith("-") and len(split_inci(line)) >= 5:
return split_inci(line)
return []
def position(inci, pattern):
return next((i + 1 for i, name in enumerate(inci) if re.search(pattern, name)), None)
# product id -> the ingredient its name advertises
SERUMS = {"P427417": "niacinamide", "P525930": "retinal", "P518192": "polyglutamic acid"}
for product_id, active in SERUMS.items():
r = requests.get(f"{BASE}/product-details", params={"productId": product_id},
headers=headers(), timeout=60)
r.raise_for_status()
product = r.json()
product = product.get("data", product) # this endpoint returns the product without a wrapper
inci = inci_list(product["currentSku"].get("ingredientDesc"))
at = position(inci, active)
phen = position(inci, r"^phenoxyethanol$")
if phen is None:
verdict = "no phenoxyethanol to compare with"
elif at and at > phen:
verdict = "after phenoxyethanol: typically 1% or less"
else:
verdict = "before phenoxyethanol"
print(f"{product['productDetails']['displayName'][:45]:45} {len(inci):>3} ingredients, "
f"{active} #{at}, phenoxyethanol #{phen}: {verdict}")
My run printed:
Niacinamide 10% + Zinc 1% Serum for Oily Ski 11 ingredients, niacinamide #2, phenoxyethanol #10: before phenoxyethanol
Crystal Retinal Age-Defying Vitamin A Night S 43 ingredients, retinal #37, phenoxyethanol #11: after phenoxyethanol: typically 1% or less
Super Saturated Hydrating Barrier Serum with 20 ingredients, polyglutamic acid #12, phenoxyethanol #9: after phenoxyethanol: typically 1% or less
The full analysis script does more cleaning than this: it joins lists that break across two lines, stops at "Free of" claims, and maps 32 title words to the INCI names that count for them. It is in the article's folder, and a broader set of Sephora examples is in the GitHub repo.
What is in a best-selling serum
The most common active is not the one on the front of the box. Vitamin E, as tocopherol or tocopheryl acetate, is in 66.7% of the 45 serums, often as an antioxidant for the formula itself. Some form of hyaluronic acid is in 62.2% and niacinamide in 46.7%. Vitamin C, which has a whole serum category named after it, is in 28.9%, and retinoids in 8.9%.
That lines up with the wider picture. Across 2,328 Sephora skincare products, vitamin E was the most common active on our Sephora skincare ingredients stats page, at 65.5%. Serums are a little different on fragrance, though: 7 of the 45 (15.6%) list fragrance or parfum, so 84.4% are fragrance-free against 66.5% of all Sephora skincare.
The lists themselves run from 11 ingredients (The Ordinary Niacinamide 10% + Zinc 1%) to 61 (Clarins Double Serum Light Texture), with a median of 33. Water comes first in 41 of 45.
Where the advertised ingredient sits
38 of the 45 product titles name at least one ingredient, 61 names in all: "Niacinamide", "Hyaluronic Acid", "15% Vitamin C", "PDRN", "Peptides". I matched each to the INCI names that count for it (sodium hyaluronate for hyaluronic acid, tetrahexyldecyl ascorbate for vitamin C, hydrolyzed DNA for PDRN) and found 57 of the 61 in the list.
The median position of those 57 is 9th, a quarter of the way down the list. Only 22 of the 57 (38.6%) are among the first five ingredients. That is not a scandal: water, glycerin, butylene glycol and the thickeners usually take the first places, and many actives work at a few percent.
The phenoxyethanol test settles less than it promises. Only 21 of the 45 serums (46.7%) contain phenoxyethanol, so for 32 of the 61 named ingredients there is nothing to compare with. Where it is present, it sits low: at a median position of 20th, about 91% of the way down, with a median of 2 ingredients after it. So the line it draws usually covers only the tail of the list.
Of the 25 named ingredients in serums that do use it, 22 come before phenoxyethanol and 3 after: polyglutamic acid in Experiment's Super Saturated serum, and retinal in two retinal serums (Medik8 Crystal Retinal and Dieux Firm Revival HydroRetinal). The retinal results are no red flag. Retinoids are potent at fractions of a percent, so a retinal listed after the preservative is what you would expect.
Remember the test works in one direction only. After phenoxyethanol typically means 1% or less. Before phenoxyethanol does not mean above 1%, because the "1% or less" section can be in any order and may start well above the preservative.
What the other 24 use
The 24 serums without phenoxyethanol mostly preserve with 1,2-hexanediol (17) and ethylhexylglycerin (16), with caprylyl glycol, sodium benzoate and potassium sorbate behind them. 1,2-hexanediol can be used at more than 1%, so it is no substitute marker. One serum, Caudalie's Vinopure, lists none of the preservatives I searched for; alcohol is its third ingredient.
When the title gives a percentage
Ten title claims give a number. Every claim above 1% sits within the first five ingredients, which is what the label rule predicts.
| Product | Title says | Listed as | Position | Phenoxyethanol |
|---|---|---|---|---|
| The Ordinary Niacinamide 10% + Zinc 1% | niacinamide 10% | niacinamide | 2 of 11 | 10 |
| Experiment Super Saturated | glycerin 30% | glycerin | 2 of 20 | 9 |
| Medik8 C-Tetra | vitamin C 20% | tetrahexyldecyl ascorbate | 2 of 31 | 28 |
| Sunday Riley C.E.O. | vitamin C 15% | tetrahexyldecyl ascorbate | 2 of 26 | 12 |
| INNBEAUTY PROJECT Pro C | vitamin C 15% | tetrahexyldecyl ascorbate | 3 of 55 | none |
| EADEM The Mark | tranexamic acid 5% | tranexamic acid | 5 of 33 | none |
| The Ordinary Hyaluronic Acid 2% + B5 | hyaluronic acid 2% | sodium hyaluronate | 2 of 23 | 22 |
| The Ordinary Multi-Peptide + Copper Peptides 1% | copper peptides 1% | copper tripeptide-1 | 4 of 49 | 48 |
| Paula's Choice 1% Retinol Treatment | retinol 1% | retinol | 17 of 53 | 52 |
| EADEM The Mark | kojic acid 1% | kojic dipalmitate | 11 of 33 | none |
The 1% claims are spread out: copper tripeptide-1 is 4th, retinol 17th. Both are still before phenoxyethanol, so nothing contradicts the titles.
Pitfalls in the data
The text mixes marketing and INCI. 38 of the 45 ingredientDesc fields open with bullets that start with "-", such as "-Niacinamide (Vitamin B3): Reduces the appearance of...". Count those and you double count the hero ingredients. YTTP's Superfood Skin Drip says "Niacinamide 5%" in a bullet, not in its title. innisfree's "Green Tea PDRN" is defined in a footnote as a lactobacillus ferment lysate, and the list has no DNA ingredient.
"Alternative" means absent. Caudalie's "Vitamin C Alternative" and "Retinol Alternative" serums contain no ascorbate and no retinoid. A keyword search on titles counts both as vitamin C and retinol products.
Names on the box are not INCI names. The Ordinary's Hyaluronic Acid 2% + B5 "with Ceramides" lists phospholipids and sphingolipids but nothing called ceramide. Caudalie's Resveratrol-Lift has no ingredient named resveratrol. YTTP's "Peptides" serum has no INCI name containing "peptide" (it has carnosine, a dipeptide).
Synonyms everywhere. 28 serums contain a hyaluronic acid, in seven different INCI names, and 20 of them list sodium hyaluronate without ever writing plain "hyaluronic acid". Vitamin C shows up as seven different strings, and vitamin E once as "Tocopherylacetate" with no space. Match on stems (hyaluron, ascorb, tocopher), not exact names.
Titanium dioxide is not always a sunscreen. It is in 2 of these serums, neither sold as sun protection. Medik8 Crystal Retinal lists it 12th, right after phenoxyethanol. YTTP Superfood Skin Drip lists "CI 77891/Titanium Dioxide" next to tin oxide and synthetic fluorphlogopite, the usual shimmer pigments. A sunscreen counter that searches for titanium dioxide would count both.
The order after 1% can be alphabetical. YTTP's list runs from "Ammonium Acryloyldimethyltaurate/VP Copolymer" to "Xanthan Gum" in A to Z order, 14 ingredients from 29th to 42nd. That run is the "1% or less, any order" section, and position inside it means nothing.
Text hygiene. One list hides soft hyphens inside words ("Sodium Hyaluronate" will not match). Two lists break across two lines mid-list. One size of Dieux's Deliverance ends with "Free of parabens, ... retinyl palmitate", which a naive parser reads as an ingredient. Sunday Riley C.E.O.'s 1 oz list has "Chlorphenesin Glycerin" with the comma missing.
Sizes can carry different lists. 25 serums come in more than one size, and 3 of them show different lists by size. For MERIT it is only "Water" against "Water (Aqua/Eau)". For Sunday Riley Good Genes, the 3.4 oz size lists no lactic acid at all and tetrahexyldecyl ascorbate second, while the other three sizes have lactic acid 4th. Sunday Riley C.E.O. shows three different lists across its four sizes. Read the list of the SKU you care about.
"Fragrance Free" badges are sparse. Only 11 of the 38 serums with no fragrance in the list carry the "Fragrance Free" highlight. Missing badge does not mean scented; read the list.
What I would do next
Run this across the whole category, not one best-seller page, and track lists over time to catch reformulations. The full picture for 2,328 Sephora skincare products, including mineral against chemical sunscreens, is on the stats page; parsed ingredient extracts for the full catalogue are available on request with the Sephora US product dataset.
To pull the lists yourself, get a key for the Sephora API on RapidAPI. For a tour of everything else it returns, from prices to store stock, see my LinkedIn article on Sephora product data. Questions or a custom extract: get in touch.



Top comments (0)