Three big issues with using LLMs to create marketing content.

Large language models (LLMs) like ChatGPT should be great for creating marketing content. So why do they rarely hit the mark?

Jacquard recently established a set of best practices for creating your own AI solution for producing high-quality marketing content at scale – and discovered significant limitations of LLMs through our research. 

Let’s explore the limitations of LLMs and see why marketers should be looking for purpose-built architecture for brand messaging: 

1. Brand Safety

One of the most pressing concerns with LLMs is the risk of outputting biased or inappropriate language. While things have come on leaps and bounds since the early days of LLMs, including techniques like “reinforcement learning from human feedback” (RLHF), these solutions are still imperfect. 

Some open-source models may even provide raw, unfiltered output. In today’s climate, a single viral misstep can severely damage a brand’s reputation. Ensuring that your content aligns with your brand’s values and guidelines is crucial, but not always guaranteed with LLMs.

2. Factual Errors

One word: hallucinations. Another major issue is the tendency of LLMs to produce content that ranges from misleading to outright false, a phenomenon known as ‘hallucinations’. This happens because LLMs don’t operate from an underlying knowledge base but from probability distributions. They generate content based on what fits contextually rather than factual correctness. 

As a result, marketers may find themselves dealing with inaccuracies that could misinform their audience and damage their credibility.

3. Tone Errors

The nature of LLMs means that their output won’t always be appropriate, correct, or on-brand. Factual errors, tone mismatches, and other issues are common. Unlike traditional software development, where bugs can be fixed with code changes, addressing problems in LLMs requires modifying prompts or retraining the models. 

But these changes can affect the output across all inputs, potentially introducing new issues while solving existing ones. This unpredictability makes it challenging to ensure consistent quality and reliability in marketing content.

So while LLMs offer truly exciting possibilities for content generation, marketers need to be aware of the pitfalls.