AI-powered search tools can identify published perfume notes but may embellish fragrance descriptions with ingredients absent from official listings, according to a study by fragrance discovery platform Scento. The findings highlight potential accuracy issues as consumers increasingly turn to AI for perfume research and recommendations.
The study tested ChatGPT, Google Gemini and Perplexity on 100 perfumes from 45 fragrance houses on October 20, 2026, comparing their answers against published note pyramids in Scento’s Perfume API database. Each platform received the same question for every fragrance, generating 300 responses.
Gemini was the most likely to introduce unlisted notes. It added at least one note absent from the official listing in 49% of its answers, compared with 24% for ChatGPT and 25% for Perplexity. ChatGPT performed best at returning every officially listed note, doing so in 82% of answers, followed by Perplexity at 78% and Gemini at 69%.
ChatGPT and Perplexity each reproduced the official note list exactly in 70% of answers, while Gemini did so in 43%. Across the 99 perfumes answered by all three platforms, all matched the official note list exactly for just 39 fragrances.
The study also identified a notable difference between designer and niche fragrances. Depending on the AI platform, answers about designer perfumes were 22 to 32 percentage points more likely to include notes absent from official listings. Scento said the study could not establish the cause of this disparity.
Scento has launched a free Perfume API designed to help AI assistants deliver more reliable fragrance information. The database contains note pyramids for more than 1,400 designer and niche perfumes from over 140 fragrance houses, alongside information on more than 400 raw materials and a proprietary classification system covering 14 scent families. The read-only API is available through a connector compatible with AI clients supporting the Model Context Protocol (MCP), as well as a REST API.
However, Scento cautioned that an unlisted note is not necessarily an incorrect one. The study was limited to a single round of English-language testing in the United States, and the results do not establish whether additional notes reflect inaccurate information, alternative fragrance versions or details omitted from official note pyramids. The company said the API stores source pages for its entries to support verification.
