Can ChatGPT Create a Winning Perfume? Testing AI’s Fragrance Formulation Skills

As artificial intelligence becomes more capable across industries, a natural question emerges: can AI replace human expertise in specialized fields? Perfumery is one of those domains that seems to resist automation—it requires sensory perception, creative judgment, and years of experience. But what if we tested it?

Mark from Crazy Nose decided to find out. In a fascinating experiment, he challenged ChatGPT to create a fragrance formula for a Black Afgano clone and then actually blended it to see how close it could get to the real thing.

The Setup: A Detailed Prompt to ChatGPT

Mark didn’t just ask ChatGPT to “make a perfume.” He gave it a highly detailed prompt with specific constraints:

ChatGPT, please create a perfume formula for me to make a clone of Nasomatto’s Black Afgano. Present this in a scalable spreadsheet format, showing each material with its dose weight and relative concentration. Group materials into Top, Mid and Base notes and sort them within each group by relative concentration in descending order. You may use up to 30 of the materials in the attached list, however you should avoid using expensive materials like ambergris or oud essential oil. Assume all materials are supplied at a stock concentration of 100%. This perfume needs to not only emulate the smell of Black Afgano, but also its performance in terms of projection and longevity.

After thinking for 2 minutes and 57 seconds, ChatGPT delivered a complete spreadsheet with 30 materials, organized by note group and sorted by concentration. It even added helpful columns Mark didn’t ask for—a “role/function” column describing what each material contributes, and a cost breakdown.

What ChatGPT Chose

The formula revealed some interesting choices:

Top Notes: Bergamot, elemi oil, clary sage, black pepper, and pink pepper. Mark noted the citrus seemed odd since he doesn’t typically associate Black Afgano with citrus notes.

Mid Notes: Frankincense resinoid, roasted barley (for toasted grain notes), tobacco oil, guaiacol and birch tar (for smokiness), myrrh, and synthetic saffron for leather. A coffee note appeared too—unconventional but interesting.

Base Notes: Dominated by Iso E Super, supported by galaxolide musk, ambrofix (synthetic ambergris), patchouli, benzoin, black agar, oakmoss, and various other synthetic woods and musks. Mark was surprised at how little black agar was included (only 3.5%), especially given its importance in the original.

When Mark asked ChatGPT how close the clone would be, it confidently predicted 80-90% similarity, saying it would capture the main character and vibe even if the finest details might differ.

The Blending Process

Mark meticulously measured out 30 materials using micro pipettes and a precision scale—a painstaking process that required careful attention to each material’s dose weight. He then decanted a portion and set it aside to macerate for 3 weeks, allowing the blend to fully develop.

The Reveal: How Close Was It, Really?

Three weeks later, after diluting the concentrate to approximately 25% (similar to commercial perfume strength), Mark did a blind comparison with the real Black Afgano.

ChatGPT’s Version:

• Dense, resinous, and heavy
• Opening with noticeable spice and citrus
• A synthetic marzipan-like sweetness (common in Gualtieri’s work)
• Leather notes coming through
• Dominated by scratchy synthetic woods in the base (particularly norlimbanol)
• Very little projection—heavy on the paper strip
• Overall impression: dark, dense, and artificial-smelling

The Real Black Afgano:

• Much more diffusive and sweet in the opening
• A unique “sweetened cannabis” quality in the top notes
• The signature Gualtieri bitter-sweet almond/marzipan note
• Balanced synthetic woods that don’t overwhelm
• A wisp of incense smoke rather than burning wood smoke
• Better separation between the top, middle, and base notes
• Overall: refined, sophisticated, complex

Mark’s Verdict

Rather than ChatGPT’s predicted 80%, Mark rated the clone at about 5.5 out of 10—roughly 55% similarity. It captured some of Black Afgano’s DNA (the marzipan note, the black agar in the base), but it was a far cry from a convincing recreation.

“It’s much denser and it’s much more resinous,” Mark concluded. “It has that scratchy, artificial wood smell that the original doesn’t have.”

Can AI Actually Replace Perfumers? The Real Insight

Here’s the crucial limitation: AI has no sensory perception. It can’t smell, so it can’t evaluate what it’s creating. This is the fundamental gap that prevents AI from truly mastering perfumery.

However, Mark identified several areas where AI can legitimately help:

1. Material Cataloging — ChatGPT excelled at organizing materials into structured formats. Mark now uses the spreadsheet format ChatGPT created for his own work.

2. Material Suggestions — When stuck bridging two notes or looking for materials to achieve a specific effect, AI can offer useful suggestions to experiment with.

3. IFRA Compliance Assistance — ChatGPT can quickly research and provide compliance guidelines for material usage levels (though human verification is essential before commercial release).

4. Marketing Copy — Mark has already used ChatGPT to draft labels and marketing text for his own perfume launch, with good results.

The Bottom Line

AI won’t replace perfumers anytime soon. But as a laboratory assistant—helping with research, organization, compliance checking, and initial formula ideation—it has real value. The future of perfumery isn’t AI versus humans; it’s humans plus AI, with the perfumer’s nose making the final decisions that machines simply cannot make.

---

Discover Goodmood - Screen Companion

---

Disclaimer: This article is based on the YouTube video “Testing ChatGPT’s Perfumery Skills” by Mark from Crazy Nose. It documents his experimental evaluation of an AI-generated fragrance formula and represents his independent assessment of the results.