Crumb vs. Meat: Proving AI wrong

Artificial Intelligence (AI) tools have quickly become popular sources of information, offering instant responses across a wide range of topics. While these platforms can be useful for general knowledge, they should not be relied on as the sole source of dependable information.

As experts in the South African regulations, we have seen first-hand how inaccuracies, oversimplifications and outdated references can slip through in AI-generated content. This raises an important point: when dealing with complex regulatory queries that directly impact consumer safety and compliance, relying solely on AI should never be the first port of call.

We put this point to the test.

Putting AI to the Test

To trial AI’s ability, one of our regulatory consultants asked a well-known AI tool this question:
“I am making a schnitzel to sell at a local retailer. Is there a limit to how much crumb vs meat I have to include in my schnitzel?”

The response was:

In South Africa, there isn’t a specific legal percentage that says a schnitzel must contain X% meat and Y% crumb – unless you make a specific claim about the meat content.

However, there are important regulatory principles under:

  • Department of Health food labelling regulations (R146 / R429 updates)
  • Department of Agriculture, Land Reform and Rural Development (DALRRD) for meat product standards.

The Actual Regulation

Firstly, DALRRD’s name has changed to ‘Department of Agriculture’ only; R. 146 is not applicable at all for compositional requirements; and R. 429 is a draft that should certainly not be followed.

The closest AI gets to the correct answer is the reference to ‘meat product standards’, although it does not specify which particular regulations from the Department of Agriculture are applicable. We might be lured into taking this initial information as final; however, if we look at the Regulations regarding the classification, packing, and marking of processed meat products intended for sale in the Republic of South Africa (R. 1283/2019), they actually state:

Coated processed meat products shall –

  1. consist of a minimum core of 50% of any of the products falling within the categories of whole muscle processed meat products, comminuted processed meat products, reformed processed meat products, or unspecified processed meat products, of which the surface has been covered with pre-dust, batter, or breading (or any combination thereof); and
  2. have a maximum of 50% coating, provided that the coating shall consist of pre-dust, batter, or breading (or any combination thereof).

This means that in fact there is a very specific compositional standard: a minimum meat content of 50% and a maximum 50% crumb coating for schnitzel products – in this case, showing the AI’s response to be completely false.

While AI tools can be helpful companions for brainstorming, drafting, or exploring broad concepts, this example proves that we should not use them to replace the depth, precision, and accuracy that comes from consulting regulatory experts and official documentation. This is especially important in the regulatory space, where misinformation about food composition, safety and labelling can have serious consequences. Professional expertise remains the gold standard for accurate guidance.

Skip the AI admin and grow your confidence in the regulations relating to the meat industry by joining our Meat Masterclass and learning directly from our regulatory experts!