Hello!
After testing Sparky, Walmart's AI agent, I wanted to see what Sephora is doing on its side. On its US site, the retailer offers AI Beauty Chat, an agent that compares products, fills the basket and recommends brands. And it seems to be winning people over: according to Nadine Graham, GM of e-commerce at Sephora North America, more than 20% of customers who use it buy on the same day (Glossy, June 2026).
The agent is available in two places: a tab in the header, next to the search bar, and a floating bubble on every product page. When you open it from a product page, it already knows which product you are looking at.
You don't need an account to use it. You can, however, log in with your Beauty Insider account, Sephora's free loyalty programme (3 tiers: Insider, VIB, Rouge). The agent then takes the beauty preferences in your profile into account, such as skin type or concerns. This test was done without an account.
I first asked the agent to choose, for dry and sensitive skin, between Drunk Elephant's Protini ($72) and Tatcha's Dewy Skin Cream ($74). It answers without hesitation: the Tatcha, because its texture is richer. It gave me the same verdict in a second session.
When you add that your skin reacts to fragrance, it checks the ingredients, points out that the Tatcha contains fragrance and suggests fragrance-free creams from $32, such as AESTURA's. I checked the product pages: its information is accurate.
The real plus is the customer reviews. Under each product, the agent sums up thousands of reviews in a few tags, with the strengths (✓) and the weaknesses (⊖). For the Tatcha, for example, you see “Hydration” and “Comfort” on one side, “Breakouts” and “Price” on the other.
And these tags are clickable. A click on “Breakouts” sends the question straight to the chat. It replies that 30 reviews mention breakouts, explains this by the formula (a rich, fragranced cream) and recommends a lighter texture for acne-prone skin. A click on “Hydration” confirms the strength and offers to compare it with two other Tatcha creams. At a glance, then in one click, you know what customers love, what disappoints them and why. It's the best feature of the test.
As soon as you add the cream to the basket, a “Works well with…” block appears in the chat. It suggests the cleanser from the same brand, in the 120 ml size at $40.
If you click the suggestion “Show me what pairs with this”, it recommends the same cleanser, but in the 240 ml size at $59 (shown as “$76 value”). It then suggests adding a sunscreen.
To reach the $50 that unlocks free shipping, it suggests lip balms between $10 and $14. You do, however, have to phrase the request in terms of products: the question “do I qualify for free shipping?” gets nowhere.
On the moisturisers tested, the agent never recommended Sephora Collection. I then asked 3 neutral questions, and the own brand came up every time: in 1st position for a mascara under $20 (“best budget pick”), in all 3 recommendations for a brush set, and in 3rd position for a micellar water. For a gift, it also highlighted the Sephora Favorites set. So the brand mainly comes up on entry-level products and accessories, where it is already strong in store.
Test your agent on your 10 best-sellers, written the way your customers write them (“No 5”, not “N°5”). The reason: a product missing from the agent's index doesn't create any visible error in your dashboards. It simply disappears from the recommendations.
Feel free to tell me which brand you'd like me to test next time: just reply to the newsletter.
Speak soon!
Gautier
CMO @ Agentic Commerce Club
