BLOG: AI in public relations: the human element is key

By Jim Kelley

A recent survey from MuckRack, a media relations software provider, found that 61% of public relations pros either already use artificial intelligence tools or are interested in adding AI to their workflow. Only 15% said they have no interest.

PR applications actually started several years ago with writing and note-taking tools such as Grammerly, Jasper, and Otter that have AI elements built in.

The applications have exploded this spring with the free public availability of large language models, whether ChatGPT, Microsoft’s Bing, or Google’s Bard. People are using them to overcome writer’s block, suggest alternate wording, and arrange source material in plain language, in a sort of Google Search on steroids.

Every day now, PR pros rely on AI to help them craft pitches, write press releases and generate ideas for social media copy. But other types of predictive AI models have the potential to be equally, if not more impactful on the profession.

Yann LeCun, Meta’s Chief AI Scientist and one of three “Godfathers of AI,” recently wrote about “3 obstacles for a piece of content to have an impact on people,” which he lists as:

  1. Production

  2. Dissemination

  3. Attention

In LeCun’s words, “Computers and networks have made production and dissemination easy. The bottleneck is now capturing the audience's attention. Generative AI increases No. 1, but not 2 and 3.”

How can we use AI to produce and disseminate content that gets people’s attention? The human element is key.

Production

Most of the press coverage of AI has focused on ease of production. Who wouldn’t want to be able to quickly turn a short prompt into a 1,000-word article? But easy production comes with pitfalls, which we outlined in our first post in this series, “How language models go wrong.”

Raw responses tend to be formulaic, out of date, and often simply wrong. Because they are based on everything that has gone before, by definition they are unlikely to qualify as “thought leadership.”

So how can you use your human talents to create useful content with GPT? It’s important to think carefully about how you write your prompts. Instead of writing one overarching prompt, consider breaking it down into smaller elements.

You can start with a literature review, a task on which these models excel. Instead of manually Google searching on a topic, reading through and summarizing the results, you can use an up-to-date model like Microsoft’s Bing or Google’s Bard to complete this process for you. These search engine-based language models will search the web for relevant information and summarize it, providing links to source materials so that you can read more.

Often, the model will provide useful information that you may not have found on your own. For example, imagine you are looking for opposing voices on a topic. You may be well-versed on the arguments in favor, but you may not know as much about the arguments against. An AI tool searching the web can quickly summarize the most commonly found viewpoints, highlighting claims you may not be familiar with.

Dissemination

Media relationship management tools like MuckRack are also incorporating AI into their services. MuckRack now offers AI tools not just to draft press releases, but to recommend journalists with relevant beats who might be interested in the story.

The press release tool runs into many of the pitfalls mentioned in our previous post (“hallucinations,” outdated or misstated facts, etc.), but the journalist recommendation feature has the potential to become very useful if it’s based not just on a job description, but what the journalist has actually been writing about.

Artificial intelligence ultimately boils down to predictions. ChatGPT tries to predict what a human might say in response to your prompt. Google tries to predict which pages you are most likely to click on in response to your search. Social media algorithms try to predict which posts you are likely to engage and interact with.

Large, curated data sets allow AI models to make much more accurate and useful predictions. So when you evaluate an AI tool, it’s important to consider the size and quality of the data set it’s trained on.

MuckRack’sdatabase of over 250,000 journalists not only keeps their contact info up-to-date, but can track their clicks and replies to pitches. Training their journalist recommendation model based on outlet size, a journalist’s frequency of articles, and the full text of those articles allows their tool to make useful recommendations on which journalists are most likely to engage with a pitch.

Language models can also be used to transform or investigate existing content. Say you have a list of companies attending an upcoming agricultural conference, and you would like to identify and target the companies that manufacture fertilizer. You could do this manually by Google searching each company, but that can be incredibly time consuming. An AI model can perform this task in seconds.

Just be sure to watch out for “hallucinations” — the new list may include companies that did not appear on the original list of attendees you provided. And NEVER put information in a prompt unless you are comfortable sharing it publicly — these models are trained partly based on the information you input, so others may be able to access the information you include in a prompt.

Experience and common sense are still key. Should your content be in the form of a formal press release, a blog post, or a social media post? If a social media post is best, which platforms should it be shared on to reach your desired audience? AI can’t answer these questions on its own.

Attention

AI can help you to produce content and identify the best targets for dissemination, but can it help you to get people’s attention? In most cases, no. AI models have no concept of what is interesting to humans. Even worse, it has no concept of what is repugnant to humans: without careful content moderation, the early AI models have repeatedly given responses that are racist, misogynistic, or otherwise morally unacceptable. Responses are only as good as the existing information publicly available on a topic: garbage in, garbage out.

Before using AI to help with content development, you need to understand what purpose you would like to achieve. Who do you want to reach? How do you want this targeted audience to respond? What topics does your audience find interesting? The more precisely you can answer these questions, the more attention-grabbing and useful the resulting content will be.

While AI may not be able to make your posts attention-grabbing on their own, AI-powered tools like Google Analytics can help you dig deep into your existing content to determine what’s working. This, in turn, can help suggest new content that will resonate with your followers.

By paying attention to this feedback, you can ensure that your audience will pay attention to you.

Embed Block
Add an embed URL or code. Learn more
Previous
Previous

pv magazine: Building to last in the global South

Next
Next

BLOG: AI in public relations: how language models go wrong