Generative AI’s a Good Writer Now, But its Real Value for PR is Much Bigger

[Editor's Note: The essay below was submitted days prior to Microsoft's announcement today (Feb. 7) detailing its latest collaboration with OpenAI, the company behind chatGPT.]

Full Disclosure: A machine wrote the following paragraph:

After spending the past two years immersed in generative AI, I can confidently say that AI will reinvent the practice of PR and transform our industry, unlike any other force since the Internet. Much like how ERP and CRM systems revolutionized manufacturing and customer relations, generative AI will soon automate our workflows, streamline content creation and reduce stubborn friction points that prevent us from best servicing our clients.

In the paragraph above, the machine won, overcoming skepticism lobbed at it. The machine captured my voice, offered an appropriate level of domain knowledge and, most importantly, spoke to my audience.

How? An engineer at Carnegie Mellon University and I used reverse engineering on the writing process. The resulting custom programming and UI lets one aim the machine at a precise vector of knowledge and produce edit-ready content.

Write This Way

The greatest breakthroughs came when we applied these prompts to ChatGPT. We then worked to overcome its limitations by:

  1. Eliminating hallucinations. Left alone, machines hallucinate, filling in missing information with whole-cloth fiction. ChatGPT’s hallucinations can be so assertive it can make fiction sound true. Instead, we supplied the machine with facts upfront, which helped eliminate hallucinations.
  2. Avoiding the college essay style. Without direction, generative AI writes in an academic style. Again, we created prompts that overcome this, resulting in more thoughtful content that resonates with audiences.
  3. Gaining an accurate voice. For this, we tuned the prompts, pointing the machine to internal archives of data and information so it could understand the writer’s preferred voice. The machine learned (for lack of a better word) to write like me by first drawing from writers I aspire to emulate.
  4. Co-writing with the machine. To gain more thoughtful results, we asked the machine to generate headlines before writing a paragraph. We then edited the best headline and fed it back into the machine. After that, we asked for a first paragraph.

The above insights helped us customize the user interface (UI), pointing the machine to our desired end.

Ultimate Win for PR

Now, here’s the crushing part. All this work will be antiquated as soon as Microsoft embeds OpenAI into its software. No doubt, Microsoft and OpenAI have the money and might to execute better than I can imagine. And that’s good news for communicators, because AI-assisted content creation is only the beginning.

The ultimate win for PR pros is how AI will transform our workflow.

For example, in crisis communication, generative AI will let us consider all audiences and offer a selection of messages individualized for consumers, the media and shareholders. And not just all shareholders but discrete audiences of long-term holders versus shorts, who just entered the fray on the dip. It will create messages that are appropriately voiced and nuanced.

In addition, AI will help eliminate some of PR’s great frictions, such as when executive you represent responds too slowly to a media inquiry and loses the interview. AI can help accelerate the time it takes executives to provide thoughtful perspective to reporters.

Moreover, AI can overcome the friction of a brainstorming session drying up. Here, AI will offer new ideas.

Transforming PR pros’ workflow is AI’s real promise. That promise eventually will relegate my opening AI trick to nothing more than the feeling I had making my first music purchase on an e-commerce site, in 1998.  Awkward and off-putting, but oh, such a harbinger of better things to come.

Greg Matusky is CEO and founder of Gregory FCA

[Editor's Note: The writer’s views do not necessarily reflect those of PRNEWS. We invite opposing essays from readers.]