AI Label Review Mistakes: What It Missed in FDA Regulations
I Ran a Label Through AI
AI is everywhere right now, and I get it. It’s fast, it’s easy, it’s cheap and it feels like you’re getting an answer without having to hire an expensive consultant or attorney. So I wanted to see what would actually happen if I ran a label through AI and asked it to review it for compliance.
Some of it was helpful. A lot of it was not. But the biggest takeaway, was that what it missed could actually create risk if a brand relied on it alone. What is scariest is that I ran this on a AI platform that market’s itself on label compliance. As you read through, think about what would happen if I ran it through a more popular, but not as specialized AI agent? Disaster.
Disclaimer: Please know that I sent these notes to the company as soon as I saw the results.
It Missed Basic Regulatory Nuance
One of the first things that stood out was how it handled net quantity. It flagged net weight on a dietary supplement as an issue, even though supplements sold by count do not require weight in US customary and metric, count alone is sufficient. AI did not understand that.
If you don’t understand when something applies and when it doesn’t, you end up overcorrecting or missing something more important.
It Didn’t Properly Review Claims
This was the biggest gap and the scariest.
AI did not flag several claims that should have at least been questioned. There were statements that implied disease treatment and some statements that were not claims that were labeled as claims, both of which could be flagged by the FDA. What does this mean? Well, you could get a warning letter, have to destroy product, pay some fees, or worse.
Claims are not just about wording. Claims are about interpretation, and AI misses the mark.
It Didn’t Understand Panel Differences
In the AI report created, it referred to a Supplement Facts panel as a Nutrition Facts panel. That is a basic distinction. If it cannot consistently identify the panel it is reviewing, it raises questions about how deep the review actually goes.
On top of that, there was no real evaluation of units or % Daily Value, which is a core part of label compliance, not something optional. Once I pointed out the %DV, AI could review against actual regulations (that I gave it), but it took me pointing it out and pointing it in the right direction to fix the issue.
It Missed Formatting Details
Latin names in supplement panels must be italicized. That was not flagged.
Not the biggest issue on its own, but it shows that the review is not getting into the level of detail that matters when you are finalizing a label.
It Gave Weak Guidance on Allergens
The label included a “may contain” statement, and AI did not push back on it or explain the limitation. “May contain” does not remove liability. It does not replace proper allergen controls. Without cGMPs and validated processes, that statement does not protect you.
This is something brands rely on more than they should.
It also marked the location as acceptable, when it wasn’t. The allergen statement must be immediately after the ingredient statement without any intervening material. The specific label I sent had an allergen statement with many different graphic and text statements in between it, yet AI said it was acceptable. Seems small, but again, it is against a very straightforward regulation.
It Flagged Something That Was Actually Compliant
AI marked country of origin as non-compliant, even though it was 100% compliant. This is a perfect example how AI can create confusion, because it does not always understand context or how labels are actually finalized.
It Missed Serving Size and Basic Rules
For the specific label being reviewed, the serving size was listed in ounces where it should have been listed in pieces to align with RACC, and that was not flagged. It also missed the use of “about” in servings per container, which is only allowed when there are between 2 and 5 servings (truly one of my biggest pet peeves).
So What’s the Takeaway?
AI is a tool. It can help you get started. It can help you ask better questions. But it is not reviewing your label the way a regulatory professional does and is not a suitable replacement.
It does not understand nuance, claims, risk, and it missed a lot of simple regulations.
I am not anti-AI, but I am pro-accuracy. If you are using AI to review your label, use it as a starting point, not the final decision, because at the end of the day your name is on that package. If something is off, whether it is claims, allergens, or something small that adds up, that responsibility comes back to you.
If you want a second set of eyes on your label, I’d love to help you understand any type of nuance or compliance flag.
Happy labeling,
Lauren
