Why Sebastian Dobrincu wants online shops to let AI assistants in

Sebastian Dobrincu, Founder & CEO at Scento
The Unpopular Opinion

Why Sebastian Dobrincu wants online shops to let AI assistants in

Sebastian Dobrincu is Founder & CEO of Scento, a Paris-based online perfume retailer that hand-pours decants from original bottles so people can try a fragrance before buying a full bottle. The conversation explores why Scento opens its catalogue and checkout to AI assistants, what a test of 300 perfume questions found, and why publishing accurate product data may not be enough to shape their answers.

Sebastian DobrincuFounder & CEO at Scento
Published November 22, 2026
The takeaway

Dobrincu argues that retailers should publish accurate, machine-readable product data rather than block AI assistants by default. His own test suggests that assistants can still add details not found in official fragrance notes, and he says he cannot yet show that publishing better data changes their answers.

01

What’s one opinion about your industry that most of your peers would disagree with?

Online shops should let AI assistants in, not keep them out. Many retailers treat ChatGPT and other assistants as a threat to their traffic and either block them or ignore them. I think that when a shopper asks an assistant what a perfume smells like, or where to try it, the answer should come from accurate data the shop publishes itself, not from a guess. So at Scento we open the shop to assistants on purpose, checkout included.

02

What convinced you? Tell us about the moment, client or result that changed your mind.

A test we ran in October. We asked ChatGPT, Gemini and Perplexity the same plain question about the notes of 100 perfumes, 300 questions in total, and compared every answer with the official note list. All three were usually close on the notes themselves. But Gemini added at least one note that isn’t on the official list in 49% of its answers, ChatGPT in 24% and Perplexity in 25%. For only 39 of the 99 perfumes all three engines answered did all three match the official list exactly. That settled it for me. The assistants answer these questions whether we take part or not. The only choice a shop has is whether the answer can draw on its own accurate data.

03

What’s the usual way people in your field handle this, and what do you do instead?

The usual way is to treat product data as something to protect: block the AI crawlers, keep the catalogue behind the website and hope the customer comes to the site to search. We do the opposite. Our robots.txt lets the main AI crawlers in. We publish a plain guide for assistants at scento.com/llms.txt: what we sell, our sizes and where each answer lives on our site. Our Scento Perfume API is free, with no key and no sign-up: more than 1,400 perfumes, more than 140 houses and more than 400 raw materials, with each note pyramid traced to the house’s or an authorised retailer’s published list. It also runs as an MCP server that assistants connect to, and it is listed in ChatGPT’s plugin directory. And an assistant can start a checkout on a real cart, but the customer always finishes and pays on our own site.

04

Where has this view cost you, or where have you had to admit you were partly wrong?

It costs engineering time that doesn’t show up in next week’s sales report, and anyone, competitors included, can query the same data for free. I accept that trade: I’d rather be the source the assistants quote than guard data they will describe anyway. Where I was partly wrong: I expected that publishing accurate data would on its own fix the answers. Our own study doesn’t show that. It measured what the engines say, not why they say it, and a note that isn’t on the official list isn’t always wrong, because some houses publish only a few key notes. Publishing good data is necessary, but I can’t yet prove it changes what an assistant says.

05

What should someone in your field stop doing, and what should they start doing?

Stop blocking assistants by reflex and treating every AI answer as lost traffic. Start by asking three assistants about your ten best-selling products and comparing their answers with your own data. Then publish the facts in a form machines can read, such as an llms.txt file and plain-text copies of your key pages, with a link back to the page where the customer buys.

Thank you to Sebastian Dobrincu and the team at Scento for sharing an honest view with Leaders Perception readers.

Share this interview

Have a view your industry needs to hear?

Leaders Perception publishes interviews with business owners and leaders who see their industry differently. Five questions, about ten minutes, no calls. It is free and every completed interview gets published.

Get featured on Leaders Perception

Explore additional categories

Explore Other Interviews