INCI Ingredient List Check With AI: Names, Not Safety Sign-Off

AI can check the paperwork side of an INCI ingredient list: spelling and consistency of names, whether the order matches your formula's percentages, and which fragrance components may need declaring as allergens. It cannot decide that a product is safe. In the EU, that is a qualified safety assessor's signed report, and AI output should only prepare the ground for them.

The useful way to think about it is two separate questions. "Is the list written correctly?" is a clerical check, and AI does it quickly, with a person confirming. "Is this formula safe for the people who will use it?" is a professional judgement, and no chatbot output stands in for it. Keeping the two apart lets an indie brand use AI where it saves hours without drifting into using it where it creates liability.

Follow me on Instagram@sagnikteaches

What the list has to get right before anyone checks safety

An INCI list (International Nomenclature of Cosmetic Ingredients) is the ingredient declaration printed on the pack and shown online. For the EU market the main rules are these, and each one is something a prompt can test:

Connect on LinkedInSagnik Bhattacharya

• Every ingredient appears under its INCI name, not a trade name or a friendly name. Water is "Aqua", not "purified water"; a botanical extract has a Latin name followed by the part used, such as "Camellia Sinensis Leaf Extract".
• Order follows weight. Ingredients above 1% are listed in descending order of weight at the time they are added. Ingredients below 1% may follow them in any order, and colourants (CI numbers, other than hair dyes) may be listed in any order after all the other ingredients.
• Nanomaterials are flagged. Any ingredient present as a nanomaterial is followed by the word "nano" in brackets, so the label depends on which grade of a pigment or UV filter you actually buy.
• Fragrance components that count as allergens are named individually, once they exceed the thresholds in the regulation: 0.001% in a leave-on product and 0.01% in a rinse-off one.
• "Parfum" stands for the fragrance blend, with named allergens added after it where they exceed those thresholds.
• Good practice: one list everywhere. Pack, outer carton, online listing and the paperwork your assessor holds should all carry the same list, even where the rules only name the packaging.

The allergen list has changed. The amending regulation, Regulation (EU) 2023/1545, adds 56 fragrance allergens to the 24 that its recitals say already had to be named individually, so about 80 in all; count the entries in the current annex rather than relying on a round figure. Under its transition rules, products that do not follow the new labelling could be placed on the EU market only until 31 July 2026, and those already placed by then may be made available until 31 July 2028. If you sell or ship to customers in the EU, that date matters for every label you reprint.

Subscribe on YouTube@sagnikteaches

Rules differ by market, and other markets have their own ingredient-naming and label rules; the cosmetic label requirements checklist compares them. What follows concentrates on the EU rules above, because they are the strictest common baseline for indie brands.

A worked formula: one leave-on face serum

Here is an illustrative formula for a small brand's leave-on face serum. The figures are invented to demonstrate the checks and are not a real product.

The rows above are in the brand's own wording and in no particular order, which is how many indie spreadsheets look. Before anything else, check that the shares add up to 100% (here they do); a formula that totals 98% means a row is missing or a figure is wrong, and every later step inherits the error. The check is to turn them into a correct label list, and every step below shows what AI does and what you confirm.

Step 1: Convert names to INCI, and flag every doubt

Give the AI the formula table with the supplier's own specification names beside it, and ask for a conversion with a confidence flag. The supplier column is the anchor: the AI is checking that your wording matches the supplier's, not inventing names from memory.

I run a small skincare brand. Below is a formula table: my working name, the supplier's name from their specification sheet, and the percentage.
Task:
1. For each row, give the INCI name you believe applies, using the supplier column where it already looks like INCI.
2. Mark each row CONFIRMED (supplier column and your answer agree) or CHECK (they differ, or you are not sure the name is exact).
3. Do not guess spellings. If unsure, say "CHECK" and say why.
4. Do not comment on safety, concentration limits or claims.

[table pasted here]

A reasonable response (illustrative) confirms Aqua, Glycerin, Niacinamide, Sodium Hyaluronate, Phenoxyethanol, Ethylhexylglycerin, Citric Acid, Xanthan Gum and Tocopherol, and flags two items for checking: the extract name, because the exact INCI wording for an extract depends on the part of the plant and the extraction used, and "Parfum", because the AI cannot know which allergens the fragrance contains.

What a person should do next is open the supplier's specification and the official CosIng database entry and compare the spellings character by character. Typical slips are the spelling "Glycerine" versus INCI "Glycerin", "Vitamin E" instead of "Tocopherol", and "Hyaluronic Acid" versus "Sodium Hyaluronate" (the sodium salt, which carries its own name). If the AI and the supplier disagree, the supplier's current sheet plus CosIng is your reference, not the chatbot; the Commission describes CosIng as an information database with no legal value, so the regulation's annexes and your assessor still have the last word. While you are there, ask the colourant supplier whether the titanium dioxide grade is a nanomaterial, because if it is, the label must say so.

Step 2: Test the order against the percentages

Ordering is arithmetic, and it is where AI is most helpful as a second pair of eyes, provided you give it the percentages and ask it to show the working. Sort every ingredient above 1% from highest to lowest; place everything below 1% after them.

For the serum above, the list that follows from the table (illustrative) is:

Aqua, Glycerin, Niacinamide, Camellia Sinensis Leaf Extract, Phenoxyethanol, Ethylhexylglycerin, Sodium Hyaluronate, Citric Acid, Xanthan Gum, Tocopherol, Parfum, Linalool, Limonene, CI 77891

Notice what is going on. Aqua (86.9%), Glycerin (5.0%), Niacinamide (4.0%) and the green tea extract (1.5%) are all above 1%, so their order is fixed. The preservative blend is 1.0% of the formula in total, but the label names its two components, and each of those is below 1% on its own (for example 0.9% and 0.1%; take the real figures from the supplier's sheet). So Phenoxyethanol and Ethylhexylglycerin belong to the sub-1% group, and listing them first within it is a choice, not a requirement. This is the trap with any blend: split it into its components before you sort, or a 1.5% blend of three ingredients can push a 0.5% component above an ingredient genuinely at 1.2%. Sodium Hyaluronate (0.5%), Citric Acid (0.3%), Xanthan Gum (0.2%), Tocopherol (0.2%) and Parfum (0.1%) may then appear in any order. The colourant is last.

The prompt to use:

Using the percentages given, sort the ingredients into the order an EU INCI label requires: descending by percentage for anything above 1%, then anything below 1% in any order, with colourants (CI numbers) last. Split every blend into its named components first, using the supplier's percentages. Show your working as a two-column list: ingredient, percentage. Point out any ingredient that sits exactly at 1% and say which side of the line it could go. Do not change my percentages.

The "exactly at 1%" instruction is the detail a generic prompt misses: the regulation lets ingredients below 1% go in any order after those above 1%, and says nothing that covers exactly 1%, so the safe course is to keep such an ingredient in weight order at the boundary. Brands often round a supplier's range (0.8-1.2%) to one figure, and the order may then be wrong in practice. If your percentage is a rounding, return to the batch record, not to the chatbot.

Step 3: Find the allergens that must be named

This is the check where AI is most tempting and least reliable, so use it for flagging and never as the final word. The fragrance supplier provides an allergen declaration (usually sent alongside its IFRA certificate), giving the percentage of each named allergen in the fragrance concentrate. Multiply by the fragrance's share of your formula to get the level in the finished product, then compare it with the threshold.

Illustrative sum for the serum (fragrance at 0.1% of the formula, a leave-on product, threshold 0.001%):

The working: 12% of 0.1% is 0.012%; 6% of 0.1% is 0.006%; 0.5% of 0.1% is 0.0005%. So the label names Linalool and Limonene after Parfum, and Citral stays off. The AI is a good calculator-with-commentary here if you paste the table and ask it to show each multiplication, but a person must check the supplier's sheet against the current allergen list, because it is the new, longer list that applies to products placed on the market after 31 July 2026.

Below is a fragrance supplier's allergen declaration (share of the concentrate) and my usage level. The product is leave-on. For each allergen:
1. Calculate the share of the finished product, showing the multiplication.
2. Say whether it exceeds 0.001%.
3. List only the allergens on the declaration; do not add others from memory.
4. Finish with a line: "Allergens I could not check against the current official list: ..." and name any you are not certain are on it.

The last line of that prompt makes the AI tell you what it is unsure of, which is the information you need to take to a person. A concrete example of why: a brand that switches fragrance supplier often finds the new declaration lists a substance the old one did not. The AI cannot know that unless the new sheet is pasted, so the allergen check belongs in your change-control routine every time the fragrance, supplier or usage level changes.

Step 4: Run a consistency sweep across every place the list appears

Many label errors come not from wrong names but from different versions of the list living in different places. A brand with a bottle label, a box, a website product page, a marketplace listing and a safety report may have five versions after a few reformulations. Paste each version and ask the AI to compare them.

Compare these five versions of the same ingredient list. Report every difference in: ingredient names, spelling, order, punctuation, capitalisation and allergen additions. Present the result as a table with one row per difference and a column for each version. Do not decide which is correct; just list the differences.

BOTTLE LABEL: ...
BOX: ...
WEBSITE: ...
MARKETPLACE: ...
SAFETY REPORT FORMULA (as supplied): ...

A typical, believable difference report: the website says "Aqua (Water)" while the bottle says "Aqua"; the marketplace listing was last edited before the allergen update and misses Limonene; the box lists "Tocopheryl Acetate" because an earlier batch used a different Vitamin E source. None of those is a safety finding. Each is a label error, and each is the sort of thing a regulator, marketplace reviewer or customer can notice. Fixing them is quick once they are visible.

A short example: a different business, same check

The same sweep applies to a soap-and-balm maker with a market stall and an online shop. Her stall cards carry the ingredient list from two years ago, before she swapped one of her essential oils; the online shop has the new one. The AI comparison lists the difference in under a minute, where checking by eye across a printed card and a web page would have missed it. The lesson is not about AI: it is that a list kept in one master document, with each place referencing it, produces fewer mistakes than five lists edited by hand.

What stays with the safety assessor

This is the part that must not be blurred. In the EU a cosmetic product needs a safety report (the CPSR, the cosmetic product safety report) prepared by a qualified safety assessor before it goes on sale, and the assessor takes professional responsibility for it. AI can help you assemble what they need, but it cannot replace these decisions:

• Whether each ingredient is acceptable at its concentration for this product type, body area and user group, including restrictions and prohibited substances.
• The margin of safety calculation for ingredients that need one, using exposure estimates and toxicology data.
• Impurities, traces and packaging contact, including what could migrate from a pump or bottle into the product.
• Preservation and stability, including whether the preservative system is adequate and whether the tests were done.
• Special populations, such as products for children or for use near the eyes.
• Claims support for statements such as "hypoallergenic" or "dermatologically tested".

What AI can do is shorten the assessor's first hour and so your bill, if they charge by time: a clean formula table, correct INCI names, a consistency sheet showing every place the list appears, and the allergen calculation shown. See what an assessor signs and what AI can draft for the longer version, and AI and the formulating chemist for the upstream side.

A hand-over note for your assessor

Before sending a formula to an assessor, ask the AI to turn your notes into a one-page brief. It saves them guessing and you waiting. A filled-in example for the serum (illustrative):

PRODUCT: Leave-on face serum, 30 ml, dropper bottle (glass with rubber teat)
FORMULA VERSION: v4, dated for the batch record
USE: daily, face, adults; not for use around eyes; not tested for under-16s
INCI LIST AS PROPOSED FOR LABEL: Aqua, Glycerin, Niacinamide, Camellia Sinensis Leaf Extract, Phenoxyethanol, Ethylhexylglycerin, Sodium Hyaluronate, Citric Acid, Xanthan Gum, Tocopherol, Parfum, Linalool, Limonene, CI 77891
SUPPLIER DOCUMENTS ATTACHED: specifications for all 11 raw materials; fragrance allergen declaration; certificate for the colourant
TESTS DONE: preservative challenge test (attached); stability at 4 weeks, 3 months pending
OPEN QUESTIONS FOR THE ASSESSOR: niacinamide level for a daily-use leave-on; adequacy of preservative at pH 5.5; whether rubber teat needs a migration check
MARKETS: EU; labels printed in two languages

Notice that the brief states the open questions plainly. That is more useful to an assessor than a confident list that hides the doubts.

A checklist you can reuse for every label change

Print this and tick it before any label goes to the printer. Each line names what to confirm and who confirms it.

1. Names match the supplier's current specification sheet. Person compares; AI listed the differences.
2. Each name verified in an official source (the EU's CosIng database is the usual one). Person.
3. Order follows percentages above 1%, with the 1% boundary documented. AI shows working; person checks the percentages against the batch record.
4. Fragrance allergen sums done against the current declaration and the current allergen list. AI shows multiplications; person confirms the list and thresholds.
5. Colourants listed with their CI numbers, in the permitted position. Person.
6. The same list appears in every place. AI comparison; person fixes the master and every copy.
7. The assessor has seen the exact list that will be printed. Person sends; assessor confirms it matches their report.
8. Change log updated. Date, what changed, who approved.

Mistakes that look like AI errors but start earlier

Four patterns come up often and are worth recognising so you fix the cause.

The percentage was a plan, not the batch. A formula sheet records the recipe, but the batch may differ by a tolerance. If the order of two ingredients depends on a 0.2% difference, record the actual amounts and decide the order by the recipe the assessor approved.

The supplier changed the grade. A supplier's "same" ingredient comes with a different INCI name after a change in process. AI will not know unless you paste the new sheet, so a standing rule helps: any new specification sheet triggers Step 1 again.

A trade name slipped in. "Hydrolyzed Oat Complex" is not an INCI name; its components have names, and the specification sheet lists them. Ask the AI to flag any entry that looks like a brand or blend name, then take the component list from the supplier.

The list was copied from a competitor. It looks tempting to ask AI to "write an INCI list for a vitamin C serum" and use it. This produces a plausible list for a formula you do not have. An INCI list describes your product, and one invented for a different formula is wrong however tidy it looks.

Keeping your formula private

A formula with percentages is the most valuable thing an indie brand owns. Before pasting it anywhere, decide what the AI needs. For Step 1 (names) it needs no percentages. For Step 2 (order) it needs only rank, which you can supply as numbers 1 to 11 instead of percentages. For Step 3 (allergens) it needs the fragrance level and the supplier's allergen percentages, but not the rest of the formula. Business plans of the main chat assistants do not train on your content by default, and consumer plans let you turn training off in the privacy settings, but a smaller sheet is safer on any plan. Ask your assessor and your manufacturer what they require too, since some contracts forbid sending formulas to third-party tools.

If you also write the copy that accompanies the product, keep that separate: the skincare claims checking routine covers what product descriptions and ads may and may not say. A correct list under a misleading headline is still a problem.

Check it worked

After you run these steps on one live product, test the result. Pick three people: someone who did not write the label, your assessor, and a friendly stockist. Ask the first to read the pack list aloud against the website list, ask the assessor to confirm the list matches their report exactly, and ask the stockist to compare the carton on their shelf with the list on your website and on the latest batch you shipped them. If all three match without comments, your process holds. If any differs, trace it to the step above that should have caught it and add a line to your checklist. Over a few products the checklist becomes the whole system, and AI becomes a fast, inexpensive proofreader for the clerical half while the professional half stays where it belongs.

Questions indie brands ask about INCI checking

Can AI tell me whether my formula is safe to sell?

No. AI can compare names and spot inconsistencies on paper, but safety depends on concentrations, how the product is used and who uses it. In the EU that judgement sits in a safety report signed by a qualified safety assessor. Treat AI output as preparation for that person, never a replacement.

Does AI know the official INCI name for every ingredient?

Not reliably. It knows common names well and gets rarer botanical extracts, blends and recent additions wrong. Always confirm each name against your supplier's specification sheet and an official source such as the EU's CosIng database before printing.

Do I list ingredients in order of strength?

Down to 1%, yes: ingredients above 1% go in descending order of weight at the time they are added. Those below 1% may follow in any order, and colourants may be listed in any order after all the other ingredients. That is why your supplier percentages must be correct before the label is checked.

Should I paste my full formula into ChatGPT or Claude?

Think about it first. A formula with percentages is your core trade secret. Use a business plan that does not train on your content by default, or paste only the label order and supplier names, not the percentages, and keep exact quantities for your assessor.

Further reads

• Eco-Friendly Claims on Skincare Packaging: Check AI Copy — Packaging wording that needs the same careful review.
• AI Product Photos vs a Photographer: Cost and Quality Compared — Product images for the listing once the label is right.
• AI Influencer Outreach for Skincare Brands With Ad Disclosure — Disclosure rules when others promote the product.
• Skincare Replenishment Emails: Time Refill Reminders From Orders — Reorder emails timed from order history.
• Supplier Scorecard Template: Score Delivery and Quality With AI — Track which suppliers send accurate specification sheets.
• AI for Beauty Brands: Product Copy, Photos and Customer Questions — A worked product page for a small beauty brand: AI drafts the copy, photos and replies, and a claims check keeps every line provable.
• AI for Skincare Products: Six Jobs a Small Brand Can Delegate — Six brand jobs for a small skincare label to hand to AI, each with a prompt, an example and the check that keeps claims safe.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Regulation (EC) No 1223/2009 (Articles 10 and 19), Commission Regulation (EU) 2023/1545 on fragrance allergen labelling, the Commission's CosIng database page (all read November 2026).