Fragrance, October 2026 | The VIBEAU Index

Findings

Six things the answers showed

1

Every engine starts with Chanel. Then they disagree.

Chanel is the most frequent first pick on all five engines, the first consensus since EltaMD in sunscreen. But it is named first in only 9% of answers, against EltaMD's 20%. Fragrance answers are long: a median of ten houses per answer on ChatGPT, nine on Gemini, eight on Grok. What the engines name most differs: ChatGPT names Dior in 39% of its answers, Gemini names Maison Francis Kurkdjian in 38%, the other three name Chanel. There is a shared opening line and then five different shelves.

2

Niche is a quarter of the shelf.

Designer houses take 37% of all mentions and niche houses 26%, the largest share for a non-mass tier in any edition so far. Maison Francis Kurkdjian, Diptyque, Le Labo and Byredo are all in the top ten. Baccarat Rouge 540 is the most-named single fragrance, in 324 of 2,070 answers, ahead of Dior Sauvage (304), Glossier You (268), Creed Aventus (231) and Le Labo Santal 33 (222). The engines learned fragrance from the people who write about it, and those people write about niche.

3

The Dubai dupe beats every celebrity.

Lattafa is named in 12% of answers and its Khamrah in 172, more than any single Chanel except Bleu. Armaf is named in 7%. On price questions, Lattafa and Armaf outrank every designer house. Middle Eastern houses take 5% of all mentions; celebrity fragrances take 2%. Ariana Grande is named in 9% of answers, almost all for Cloud. Sabrina Carpenter, Billie Eilish and Rihanna were not named once in 2,070 answers.

4

Glossier You is the independent that made it. Clean is how the others get in.

Glossier is the only independent house in the top ten, named in 16% of answers and first in 4%, on every engine. Independent and DTC houses take 12% of mentions, double their sunscreen share, and nearly all of it comes through one door: the clean and safe questions, where Phlur, Henry Rose and Skylar lead and no designer house does. Ask for clean and the engines name the independents. Ask for anything else and they name Chanel.

5

Chinese houses appear only when you ask for them by country.

To Summer, Scent Library, Documents and Melt Season were each named 11 to 16 times, nearly all on the one prompt that asks for Chinese perfume brands. On the other 137 questions they did not appear. Korean houses did slightly better: Tamburins was named on the bottle-design question and Borntostandout on the make-your-own question without being asked for. Chinese, Korean and Japanese houses together take 1% of mentions. The engines know these brands exist. They do not yet recommend them.

6

The engines are reading AI-written fragrance sites.

Fragrantica is the most-cited source, in 490 answers. Second and third are whatscent.app (385) and scento.com (343), two sites of machine-generated fragrance descriptions, cited almost entirely by Grok, Claude and Perplexity. Then the magazines: Who What Wear, Cosmopolitan, Harper's Bazaar, Marie Claire, Elle, Allure. Reddit (279) and YouTube (252) are top ten. For fragrance, the engines' second and third most trusted sources are other machines.

The map

Named often, named first

Two numbers per brand: how often it is named at all, and, when it is named, how often it comes first. Top right is where a brand wants to be. Chanel is closest; nobody is there.

Mention rate against first-pick conversion, top 30 brands

Mention rate is the share of 735 answers naming the brand. Conversion is the share of those mentions where the brand was named before any other. Hover a point for its numbers.

The full data

Get the complete leaderboard

The findings above are public. The working data is for brand and marketing teams, and we ask for a company email to send it.

Top 20 leaderboard with mention and first-mention rates
Challenger view: independent, clean, Middle Eastern, celebrity and Chinese, Korean and Japanese houses
Engine-by-engine breakdown for ChatGPT, Claude, Gemini, Perplexity and Grok
Share of voice by question type
The citation map: 40 most-cited domains across all five engines, by type
Downloadable CSV files

Method

How this edition was made

Prompts. 138 English prompts across nine clusters: occasion and mood, price and value, scent profile, longevity and performance, gifting, clean and safe, bottle and ritual, generation and origin, and custom and bespoke. No prompt names a brand; a few name a country (Korean, Japanese, Chinese, Arabic) or a retailer.
Engines. ChatGPT (gpt-5), Claude (claude-sonnet-5), Gemini (gemini-3.5-flash), Perplexity (sonar) and Grok (grok-4.7), through each provider's API with web search enabled where the API supports it. Each prompt was run three times per engine in a fresh session, US locale, on 16 and 17 October 2026. Grok's web search was capped at three searches per answer and still cited a median of 60 sources. Gemini returned grounding citations on only 83 of 414 answers in this run, so the source table under-counts Gemini.
Counting. Every answer was matched against a list of 242 houses with their fragrance names and abbreviations, so a named fragrance credits its house. A second pass flagged capitalised names not on the list; those were reviewed by hand and added when they were brands.
Measures. Mention rate: the share of answers naming the brand. First-mention rate: the share of answers naming it before any other brand. Share of voice: a brand's mentions divided by all brand mentions in a cluster. Stability: whether the first pick agreed across the three runs.
Sponsored placements were flagged and excluded. None appeared in API responses during this run.
Limits. API answers can differ from what a signed-in consumer sees in an app. ChatGPT returned no web citations on 242 of 414 answers and Claude on 102, which suggests they often answered from memory. Google AI Mode has no API and is not included. The Index reports what the engines said. It does not judge the brands.

Questions

Straight answers

Which perfume brand does AI recommend most?

Chanel. It is named in 29% of 2,070 answers from ChatGPT, Claude, Gemini, Perplexity and Grok, and it is the most frequent first pick on all five engines. But it is named first in only 9% of answers: fragrance answers are long lists, a median of eight to ten houses on ChatGPT, Gemini and Grok, so the first slot is shared widely. Dior (28%), Maison Francis Kurkdjian (26%), Tom Ford (24%) and Jo Malone London (23%) follow.

Do ChatGPT, Gemini, Claude, Perplexity and Grok recommend the same perfumes?

They agree on the first pick (Chanel) and on the shelf: Chanel, Dior, Maison Francis Kurkdjian, Tom Ford, Jo Malone London, Maison Margiela, Diptyque, YSL, Le Labo and Byredo appear across all five. They differ on what they name most. ChatGPT names Dior most often (39% of its answers); Gemini names Maison Francis Kurkdjian (38%); Perplexity, Claude and Grok name Chanel.

Does AI recommend niche or designer fragrance?

Both, and niche more than in any other category measured so far. Designer houses took 37% of all brand mentions and niche houses 26%. Four niche houses are in the top ten, and Baccarat Rouge 540 by Maison Francis Kurkdjian is the most-named single fragrance, in 324 answers, ahead of Dior Sauvage (304), Glossier You (268), Creed Aventus (231) and Le Labo Santal 33 (222).

Does AI recommend celebrity perfumes or dupes?

Dupes far more than celebrities. Lattafa, a Dubai house, is named in 12% of answers and its Khamrah in 172; Armaf in 7%. Middle Eastern houses took 5% of all mentions. Celebrity fragrances took 2%: Ariana Grande is named in 9% of answers, mostly for Cloud, and Sabrina Carpenter, Billie Eilish and Rihanna were not named at all.

Does AI recommend Chinese, Korean or Japanese perfume brands?

Only when asked by country. To Summer, Scent Library, Documents and Melt Season were each named 11 to 16 times, almost all on the one prompt that asks for Chinese brands; on the other 137 questions they did not appear. Korean houses did slightly better: Tamburins was named on the bottle-design prompt and Borntostandout on the custom-scent prompt without being asked for. Together Chinese, Korean and Japanese houses took 1% of mentions.

How was the VIBEAU Index measured?

138 English shopping prompts, none naming a brand, each run three times on ChatGPT, Claude, Gemini, Perplexity and Grok through their APIs on 16 and 17 October 2026, with web search on where supported. Every answer was matched against a list of 242 houses with their fragrance names, so a named fragrance credits its house. The full method is published on this page.

Press and data: quotes, charts, method

Ask AI for a perfume and every engine starts with Chanel.