The GPT-3 marketing and public relations practitioner

There are clear applications for artificial intelligence (AI) content generation tools in marketing and public relations but make sure that you fact check. The shiny new toy is an ethical minefield.

The launch of ChatGPT and Playground by OpenAI has generated a huge amount of energy and discussion around its potential application in the professional sphere.

ChatGPT and Playground are among the first artificial intelligence (AI) applications using a text-based web interface that uses the GPT-3 language model. It has two main applications: generating text in response to a prompt and summarising a block of text.

GPT-3 is trained on 45 terabytes of text data from a variety of datasets. A terabyte is approximately 85 million pages of text. It’s a huge amount of data that includes books, websites, and Wikipedia.

Applications based on GPT-3 have no knowledge or intelligence, it is not sentient. It works by recognising patterns in text and predicting subsequent words. It can also use machine learning to summarise text.

Text at the click of a button

There’s a great deal of excitement about the technology as it seemingly is able to mimic human creativity and generate and edit content incredibly quickly. It generated a plausible definition of public relations from the 600 or more in research literature on the click of a button. 

A definition of public relations generated by artificial intelligence

A definition of public relations generated by artificial intelligence

However, the ability of GPT-3 and tools such as ChatGPT and Playground to generate accurate content depends on how much information it has been able to access as part of its training model, and the veracity of that content. It is equally likely to generate accurate information as it is bullshit.

A colleague forwarded a biography that Playground had generated about me. The first paragraph was factual and recognisable based on my Wikipedia entry. The remainder of the entry read as though it was legitimate but was fiction.

A biography of Stephen Waddington generated by artificial intelligence

A biography of Stephen Waddington generated by artificial intelligence

Public relations applications: analysis, editing, research and writing

The promise of GPT-3 is that it will automate and eliminate administrative tasks. There are already some very good and useful applications. 

GPT-3 is very good at summarising a block of text. This is already being used to generate a short precis of an article. It currently requires text to be cut and pasted into an application such as OpenAI Playground but it is inevitable that it will quickly be integrated into tools.

The combination of GPT-3 and audio and video transcription software such as Otter.ai provides the means to generate summaries of the key points of a conversation. This has application in a range of public relations situations from qualitative research to meeting summaries. 

More sophisticated text analysis is also possible. It can be used to provide an opinion on the tone and sentiment of an article or media coverage.

Nonsense and misinformation

OpenAI Playground is also good at generating text. It will produce blog posts, short articles, and undergraduate level essays in response to a series of questions or prompts but make sure that you fact check the output. It does not provide information about sources and will make up references.

Herein lies a limitation of the use of AI in public relations. If it generates nonsense at best, or misinformation at worst, it has no place in helping an organisation promote its reputation.

GPT-3 tools may not have any critical thinking capability, but it can do a reasonable job of summarising an argument. I asked it to summarise the main arguments for and against the Excellence Theory in public relations and got a reasonable answer. More nuanced topics such as the impact of public relations on misinformation or climate change are challenging.

These issues aside there are clear applications emerging for GPT-3 in content marketing, media relations, SEO, and social media. There is no doubt that language models will improve. GPT-4 already promises a 100x improvement on the current version.

My fear is that in the rush to adopt GPT-3 in marketing and public relations, we’ll generate content that is inaccurate that will subsequently be used to train future language models. We’ll perpetuate a feedback loop that will see the internet filled with even more rubbish.

Previous
Previous

Book review: Reputation in Business

Next
Next

End of term report