PR Daily recently wrote a story comparing four AI tools: ChatGPT, Copilot, Gemini, and Claude. They gave each the same “assignment” and had them face off in an epic battle. And they asked…who wins? 

They prompted each tool to write a news release, brainstorm ideas, suggest a list of journalists for a pitch, and parse some data.

Depending on the task and how it was prompted, there were winners for each category, but there was no clear overall winner. And, if you took the test at face value, it’d be easy for you to dismiss AI, particularly if you’re not super fond of it. “See! Even PR Daily thinks it’s not useful!”

And they’d be correct because of the way they prompted it. It’s not useful that way. Asking the AI to write a release or brainstorm ideas or do anything else without sophisticated prompting is going to get you crappy results, which is precisely what PR Daily found.

The results are significantly better if you prompt it differently and even have spent time training it yourself. Can it do your job? Not yet! But it can give you better results for those four assignments than meets the eye. 

I’m going to show you how…

General vs. Sophisticated Prompts

Can you tell I’m obsessed? Anything else happening in the world of marketing and communications? Nope. Just AI. 😂

I keep talking about it for three reasons: 1) People are still really fearful of it and you don’t need to be—and I’m on a quest to show you why; 2) Things change so quickly that I easily have new fodder every week; and 3) Talking about it helps me align my own thinking and use of it. 

The first reason—the fear—is where I’d like to start today. I mentioned a couple of weeks ago that when we fear something, we look for reasons it won’t work, which is how I read the article from PR Daily. I’m not saying they’re wrong, but it scratched the surface of what AI can do. And because we look for reasons to support our beliefs, I am here to tell you that it will do the job it was asked to do with better prompting.

Let’s use the same prompt they used to show the difference between general and sophisticated ones.

Revising the Original AI Prompt

They prompted, “I am the editor of PR Daily. Brainstorm a list of 10 stories about AI that would be of interest to PR professionals. Suggest two sources I could talk to for each.” 

The results, as they found, were not good. Some tools provided some interesting sources to talk to, but the topics were bland.

What if we changed the prompt to, “I am the editor of PR Daily. I am an experienced editor, writer, and content strategist who has done everything from digital content production and social media reporting to copywriting and account management.

“Because PR Daily is one of the industry’s leading continuing education sites, it’s important we talk about AI in a way that isn’t repetitive to our competitors. I would like to write a series of stories about AI in the PR industry that no one else has covered yet, including how our jobs will change or even disappear in the next five years. Please brainstorm a list of topic ideas and suggest two sources for each idea.

“Do not include things such as crisis communications, sentiment analysis, generative search, content marketing, chatbots, influencer marketing, copywriting, or creative development.

“The ideas should be fresh, interesting, and new.”

When that prompt is used, the results are MUCH better. 

And the AI Winner Is…

CoPilot and Gemini didn’t return anything useful, so I’d put them at the bottom of the list.

ChatGPT was a little better, but overall, it was meh, which doesn’t surprise me. It’s not super creative nor provides great research or data points. It’s great for building structure and outlines, but the actual output is better in Claude (my favorite) and Perplexity (which I added, even though PR Daily didn’t use it). ChatGPT is also the only tool that didn’t suggest me as a source. As if!

But Claude and Perplexity? They both rocked the assignment. In fact, they gave a few different answers, so I can combine the two lists and have content ideas for months!

Perplexity suggested using AI to analyze physiological responses to PR messages so you can create more effective and emotionally resonant campaigns. You’ll probably see that topic come back up in the future! I like it a lot. 

Claude suggested using AI to enhance your stakeholder mapping, particularly when the groups are complex, and the messaging is robust. This is another great idea, and it provided some unknown sources for more information. 

With a little better prompting, I had two clear winners, whereas PR Daily had none. 

You have to train it and provide a recipe for success. Without it, just like with humans, it will fail the test.

AI Is Your 24/7 Intern

How many of you hire or work with interns? How many of you were interns? Can an intern create something brilliant their first time out of the gate? Maybe so, but most likely not.

Some interns are decent enough writers and can use social media personally, but don’t know how to apply their skills to the business world. You have to spend time with them, coaching, teaching, and mentoring. Over time, their writing and judgment improve, and they can predict what comes next.

After a few years, the intern turned mid-level professional, learns how to develop strategy, increases their skillset, learns new skills, and leads others. 

Your tool of choice is similar to an intern. Can you say to an intern, “Take this LinkedIn profile and write a news release?” Absolutely. Will it be any good? Probably not. Same thing with AI. The PR Daily prompts didn’t have clear winners for every assignment because they essentially asked an intern to do the work without any coaching or professional development.

When you take that mindset with you, opportunities suddenly open up. Instead of thinking, “See. AI is crap.”, you think, ‘What if I provided it some additional input to get something closer to what I need?”

It Makes You Far More Efficient

You might argue that this takes time, and you’d be right. But it takes FAR less time to improve your prompting and become more sophisticated at training it than creating it yourself. 

I mentioned last week that a colleague in another Slack community uses a combination of four tools plus himself to write a book. Because I’m in the throes of writing the second edition of Spin Sucks, I thought I’d test out his process.

In a matter of 45 minutes, I had a strong outline, data points, research, studies, and sources to consider for three chapters. It gave me enough to stop futzing with structure and actually start writing. It took me five weeks to do that with one chapter. Using the process outlined last week, I had two additional chapters within a few days. As of this recording, I have completed three chapters, two of which took four weeks and three days less to complete.

So yes, learning to prompt it takes some time, but it’s still far more efficient than doing the work yourself. 

Especially As You Train It

Here is how you can do that so your AI gets smarter and smarter over time. Just like developing a junior team member, consistency and intentionality are key.

First, create a “knowledge base.” This could be a document or series containing key information about your company, industry, writing style, and preferred sources. Because I have so much content on Spin Sucks, it’s been easy to feed all of that to the AI and ask it to use that information as a resource.

Even if you don’t have the same amount of content, you can use FAQs, articles, speeches, videos, and other external information to train it. You can also ask it not to use your content to train itself for the world, but I’m not entirely sure I would trust that. So, submit stuff that has already been shared externally. Don’t share town hall recordings, internal memos, or messaging.

To do this, you might prompt, “Before we begin, please review and incorporate the following information about our company and communication style.” This gives it context and helps it align with your specific needs.

I like Claude for this because you can create a project and provide all that information in the notes section. It will refer back to that when it’s creating for you. Also, if you keep one thread for a specific project, ChatGPT will refer to what’s already been created in that chat as it creates.

A Couple of Techniques to Try

Next, use a technique called “few-shot learning.” Instead of asking it to complete a task from scratch, provide it with examples of what you’re looking for. If you’re writing a news release, show it a couple of your best releases first. Then, ask it to create a new one in a similar style. This helps it understand your expectations and produce more tailored results.

Another effective method is “iterative refinement.” Don’t expect perfection on the first try. Instead, use the initial output as a starting point. Provide specific feedback on what you like and don’t like, then ask it to refine the work. 

For instance, you might say, “This is a good start, but can you make the tone more conversational and add more data points from reputable industry sources?” 

Each iteration helps it better understand your preferences.

Remember to save your most effective prompts and the best responses. You can refer to these as “gold standard” examples in future sessions. This creates a sort of institutional memory, even though it doesn’t remember previous conversations.

Humans Are Still Important

It’s also important to stay up-to-date with your AI tool’s capabilities. These systems constantly evolve, so check for updates or new features that could enhance your workflow.

Don’t forget the human element. While it can be incredibly helpful, it’s not meant to replace human creativity and expertise. Use it as a collaborative tool to augment your skills, not as a crutch. The most effective users can skillfully blend AI-generated content with their insights and experience.

By consistently applying these techniques, you’ll find that your tool of choice becomes more attuned to your needs over time. It’s like having an intern that better anticipates your requirements with each interaction. The key is patience and persistence. Just as you wouldn’t expect a new hire to master everything overnight, give yourself and your tool time to develop a productive working relationship.

Remember, the goal isn’t to make it do your job for you. It’s to free up your time and mental energy for the high-level strategic thinking and creative problem-solving that truly drive value in our industry. 

By mastering it as a tool, you’re not replacing yourself—you’re supercharging your capabilities and positioning yourself as a leader in the AI-enhanced future of communications. 

Gini Dietrich

Gini Dietrich is the founder, CEO, and author of Spin Sucks, host of the Spin Sucks podcast, and author of Spin Sucks (the book). She is the creator of the PESO Model and has crafted a certification for it in partnership with Syracuse University. She has run and grown an agency for the past 15 years. She is co-author of Marketing in the Round, co-host of Inside PR, and co-host of The Agency Leadership podcast.

View all posts by Gini Dietrich