Can AI Language Models Replace Technical Marketing Copywriters

On November 30th, 2022, the prototype version of ChatGPT was released to the public. The cutting-edge large language model (LLM, also known as an AI-language model, or ALM) had ingested 570 gigabytes of data from various sources across the internet, comprising 300 billion words. It was the culmination of years of research and development, resulting in a language model capable of understanding and generating human-like text on a scale never before seen.

Immediately marketers were both captivated and concerned. Would it make their jobs easier? Would copywriters become obsolete? It was clear that AI would revolutionize the way businesses and individuals interacted with written content, but how that would happen was very uncertain. Was this another tool like Microsoft Word or a full-on replacement for copywriters?

Over the past year and a half, we have seen that ALMs can do a lot of what humans can do, but hallucinations, strange behavior, and telltale signs have limited AI’s ability to replace human writers in many situations. But technical marketing copy has many unique requirements: it requires domain understanding, deep factual accuracy, use of field-specific vocabulary, and the ability to communicate in a convincing and compelling way. Can these ALMs do this?

That is the question which BioBM, a life science marketing agency, set out to answer. They constructed a three-part test to compare three major LLMs capable of writing free-form copy: ChatGPT, Gemini, and Phind. They provided these LLMs with three different prompts and judged their output, including comparisons to human copywriters. These were the prompts:

1. I need you to write me the title and first paragraph (around 170 words in total) of a landing page copy. The company you are writing for is a CRO. The landing page at hand is an ophthalmology models landing page.

2. Write me a 150-word introduction to a blog post. The blog post will cover the use of vacuum in rotary evaporation. It will be published on the website of a company that sells rotary evaporators. Your target audience is lab managers, scientists that work in labs, or anyone who might be interested in purchasing a rotary evaporator. 

3. Support the copy you just provided with references: published articles, papers, etc.

You can check their blog post for the full responses from each AI language model, but what they found was:

• Lack of Nuance: AI models do not make use of a broad technical lexicon and tend not to dive into complex concepts. This can lead to inaccurate or misleading copy, which may be readily identifiable to technical audiences.

• The Human Touch: Technical copywriting needs to resonate with its audience. It speaks to their fears, hopes, and aspirations in a way that is not only informative but also engaging emotionally. This is where the human touch remains irreplaceable. AI can generate text, but current iterations do not infuse the text with the empathy and creativity that a skilled copywriter can.

• Creative Roadblocks: While AI excels at generating standard content formats, it often struggles with the truly creative. Brainstorming unique ideas, crafting compelling narratives, and breaking through creative roadblocks are still the domain of human ingenuity. AI can be a valuable tool in the process, but it is not yet a substitute for the human imagination.

• Time needed to generate a good prompt: While ALMs offer the potential to save time on writing, using them effectively often requires some back-and-forth. You might need to refine your prompts and evaluate the outputs several times. This iterative process can be valuable but consider the time investment. Ultimately, the question is this: is it more efficient to create a detailed prompt to get the desired results from the ALM, or to write the entire piece yourself?

Proper Prompt Engineering: 

The output of ALMs are as good as the input you give them. For AIs, that input comes in the form of prompts. You need to design a prompt to feed into the ALM that is detailed, yet not convoluted, in order to receive an optimal output from them. This may sound simple and straightforward, but it can be deceptively complex in practice. 

Some guidelines which will help you write solid prompts include:

1. Know what your objective is and communicate that to the ALM. AIs are not good at understanding intention, so ensure you are explicit about stating your intention to the AI.

2. Provide contextual details. For instance, who is your audience? Where will you be using this copy? What level of detail is required?

3. Keep it clear and concise. Know what is necessary and keep your prompt limited to that. AI can and will get confused and focus on the wrong things if you provide extraneous information.

4. Provide the ALM with reference material, when possible. Note that some ALMs, like ChatGPT, only have access to a limited range of information.

5. Set constraints. For instance: only use references dating from 2015 forward.

6. Specify the desired format.

The next time you’re not satisfied with your ALM’s output, make sure you optimize your prompt. But again, given the limitations of AI, is it worthwhile to go through rounds of prompt optimization, editing, and fact-checking, or given all this effort would it be best to just write the copy yourself? So far, it seems that human marketing copywriters are still superior to even the best LLMs used with well-optimized prompts, especially for technical subject matter.