Level 9 · Modern and Experimental Perfumery

AI in Perfume Research and Formulation: Uses and Limits

AI can summarize, draft and check arithmetic, and it can invent references, limits and untested formulas. Learn where it helps and how to verify every output.

In short

General AI tools can help a perfumer summarize a topic, draft a record template, check arithmetic and suggest questions, but they can also state wrong restriction limits, invent citations and generate formulas nobody has made. So treat each output as an unverified draft. RUŌOD Lab, the app, contains no AI and does not generate formulas. This lesson concerns outside tools that you may decide to use.

What you will learn

  • Separate useful from risky uses of general AI tools
  • Verify an AI output against primary sources
  • Protect a confidential formula when using online tools

Before this lesson: Contemporary Creative Techniques

Where it can help, and where it is dangerous

Our working view of general-purpose AI tools. It reflects their typical behavior and not a formal evaluation.
TaskReasonable useRequired check
Outlining a topic to studyGood starting listConfirm with a real source
Writing a spreadsheet or templateOften usefulTest with known values
Explaining a calculationUseful if you can follow itRework it by hand
Brainstorming conceptsHarmless and sometimes usefulYour judgment
Describing a material's odorWeak; it has never smelled anythingYour own nose
Suggesting a formulaOnly as an untested hypothesisAll normal testing and restriction checks
Stating a regulatory limitDo not rely on itThe current primary source
CitationsTreat as leadsFind the actual paper
Safety conclusionsDo not use it for thisA qualified assessor

What the research shows

Separate general chat tools from specific research models. Olfactory research has shown that structure can predict some perceptual ratings to a useful degree. The DREAM challenge, reported in 2017 by Keller and colleagues, predicted some sensory attributes from molecular features using data from 49 people and 476 molecules, and Snitz and colleagues studied odor similarity. Neither of these means a general tool can compose a perfume. You will go deeper in the Molecular Perfumery track's prediction lesson.

Some large fragrance companies have also announced AI tools for composition. Little about how well those perform has been published independently, so be careful with any claim that a specific fragrance is better because of such a tool.

A verification habit

  1. Mark every factual claim. Numbers, names, dates, limits and references.
  2. Find a primary source for each. For restrictions, that is the current standard or law. For chemistry, a database or a paper you have read.
  3. Rework the arithmetic. Use your calculator, and check that a formula totals 100%.
  4. Test any citation. Search the title and authors. If you cannot find it, treat it as invented.
  5. Write the output down as your own only after this. You are responsible for what you use.

Formulas from a machine

An AI formula is a list of plausible words and numbers. It may name materials that are hard to buy, or use shares that exceed a restriction, and it may not produce a coherent smell. If you want to learn from one, treat it like a stranger's unverified formula: check every line, work out the restricted materials, make a small study batch on strips, and evaluate it. Never test such a formula on skin on the basis of its having been generated.

Confidentiality and authorship

  • A formula may be your main commercial asset. Think about what happens to anything you paste into an online service. See protecting confidential formulas.
  • Do not paste a client's formula unless your agreement and the tool's terms allow it.
  • If AI substantially shaped something you publish, describing it honestly is wise. Check what your market, clients and publishers require.

Practical exercise: Audit one AI answer

About 45 minutes · You will have: a marked-up AI answer in which every factual claim is verified, corrected or struck out

  1. Ask a general AI tool, without sharing any confidential information: "What are the main allergens and restrictions to consider in a jasmine-style perfume? Include references."
  2. Save the answer without editing it.
  3. Mark up every factual claim as verified, wrong, or unverifiable, each with your source and date.
  4. Count them, and note the claim that would have been the most damaging had you trusted it.
  5. Write your own short rules for what you will and will not use AI for.

You have done this well if each claim has a verdict and source, and your rules come from what you found.

Common mistakes

  • Trusting a confident tone. Tone gives no evidence.
  • Treating a citation as proof it exists. Look it up.
  • Testing a generated formula on skin. It has not been assessed.

Knowledge check

1. An AI tool gives you an IFRA limit for a material. What should you do?
Show the answer

Check the current standard yourself, and do not rely on the answer. General tools can state wrong or outdated limits with confidence.

2. An AI writes a formula that looks plausible. What is it?
Show the answer

An untested hypothesis, which still needs every normal check. Only making, evaluating and assessing it can tell you more.

3. What should you consider before pasting a client's formula into an online tool?
Show the answer

Confidentiality and what your agreement and the tool's terms allow. You could disclose something you have promised to protect.

Written and reviewed by the RUŌOD Lab team. This is general education, not legal, regulatory, medical or safety advice. Example dilutions and exercises are for learning, are not validated products, and have not been safety-assessed. Completing a lesson confers no qualification. How we check these lessons.