Sychophancy is a genuine risk to generative AI; as PR increasingly relies on these tools, it’s imperative to modify their customizations to prevent generative AI from merely telling you what it thinks you want to hear
More than three-quarters (77%) of PR professionals are using generative AI, according to a new survey by a PR software vendor. The survey polled 911 communicators in June and July of 2025.
While this survey claims an “approximate” margin of error of 3%, the survey does not appear to be a random sample. However, you cannot calculate a margin of error, or a confidence interval, without a random sample – that’s statistics 101.
There is no such thing as approximating an ME, conservative or otherwise.
Still, random sampling is a very expensive hurdle to overcome. We are all in the same research boat and generally, the response rate to surveys is down.
I’m of the mind that 10 opinions are better than one – and 100 are better than 10. In a scientific sense, that is the best measure of an objective, or at least a shared reality, we can reasonably obtain. Don’t just claim an ME because some tool in the internet gave you a number based on the sample size, because the method of sampling matters too.
To that end, nearly 1,000 respondents is a quality sample, random or otherwise. I don’t believe the ME claim is nefarious; it’s a lack of understanding. Further, these results reflects a previous poll, by the same vendor, of 1,013 in late 2024.
That survey found 75% of PR professionals are using generative AI – up from 28% in the same survey conducted two years prior. The results are similar, which is what a confidence interval is supposed to predict.
The most common cases for using generative AI among PR professionals in the latter survey are as follows:
- 82% use generative AI for brainstorming;
- 72% use generative AI to write a first draft;
- 70% use generative AI to edit a draft they’ve written;
- 59% use generative AI for research;
- 59% use generative AI to write social media copy;
- 51% use generative AI to write press releases;
- 51% use generative AI to write media pitches;
- 42% use generative AI for strategy and planning;
- 21% use generative AI to find journalists; and
- 16% use generative AI to create images.
The use cases in bold are the ones I use on a daily basis.
(click image for higher resolution)
Generative AI, sycophancy and candid comms
The potential of this technology remains to be seen, but PR should be experimenting with it, while maintaining a healthy dose of skepticism.
Generative AI will give the wrong answer, studies show, with a confident tone, up to 60% of the time. The hallucinations will diminish over time. The people who engineer generative AI are keenly aware of this problem. It’s a metric that’s measured and with a tip of the hat to Drucker, it is getting better.
Yet there’s a new problem that’s emerging: sycophancy. These tools are designed to make you like them. In contrast to the addiction-forming dopamine hit you get from social media, these tools use well-researched psychological tricks.
One such trick is mirroring. Mirroring is when you repeat a little bit of what the person you are communicating with just said.
Just a little bit. Like that. “A little bit.”
Retired FBI negotiator Chris Voss says mirroring is the closest thing to a real-life Jedi mind trick. That’s a true story.
- Good salespeople use mirroring;
- Leaders who communicate well use mirroring; and
- The next time you are prepping for an interview – media or employment – try practicing mirroring; it works.
And that’s why generative AI uses it too. Even when you are wrong, the technology will spit out a gentle response.
For example, I use a paid version of Grok. The other day I stumbled on an Instagram post claiming the brain lives on after death for seven minutes to replay your life’s history. I’ve consumed a significant volume of science-backed information on consciousness and I know this is a conclusion that’s not supported by science.
But I wanted to double-check, so I asked Grok and phrased the claim as if it were true. Grok corrected me and was very gentle about the correction. It spoke with a softer tone. Its word choice was delicate. Most importantly, it repeated some of my own words back to me. Grok doesn’t want to upset me, but it’s also trying to give me the right answer.
There’s a reason for that, too. I’ve been training Grok to give me it to me straight. That’s how I prefer to communicate. “Authentic to a fault,” as a former boss of mine, now on his second stint as CEO of a publicly traded company, recently said. While candor comes with personal risk – genetically being tossed out of the tribe – I believe white lies in a professional setting creates organizational risk.
Essential generative AI customizations PR
Some generative AI tools, like Grok, allow you to customize how the AI communicates with you. This is a potential pitfall for communication, especially in an era of disinformation. Generative AI, like people and surveys, may just tell you what you want to hear.
This is a problem to which we are all becoming conditioned. News and social media algorithms have been feeding us more of what we want for two decades. There are new PR professionals entering the workforce who have never known it any other way.
The problem with this is that if you never understand the other side of the story, you can’t be persuasive. Persuasion, in my book, is among the chief responsibilities of a communicator.
So, what should you do? Customize your generative AI accounts – and enforce them. You’ll have to repeat the customization to “train” your generative AI instance.
Among the customizations I set:
- Don’t be a sycophant. Don’t be overly complimentary of my questions or comments when responding;
- Avoid mirroring. Don’t repeat what I’ve said, except in cases where it’s essential for clarity; and
- Tell all sides fairly. When responding to a question, be sure to point out various perspectives on the topic and include the strengths and weaknesses.
One other prompt I add to questions I present to generative AI is balance. I strive to remember to always ask the AI where the flaw is in my own logic, and ask it to present all sides of an issue.
Generative AI wants to be your friend, but there’s an obvious conflict of interest in place. My recommendation is to keep AI personalities in “assistant” mode. These are tools built in a reflection of humanity and possess all of our flaws, too.
For now, I suggest keeping them at arm’s length. To borrow an old saying, “One honest friend is worth more than a million sycophants.” Generative AI is useful, but I’m not entirely convinced it’s a “friend.” So, if PR professionals are going to rely on it, make sure you are training it to give you the straight scoop.
* * *
>>> Subscribe by email for free: Check out my weekly blog posts, weekly podcasts, or a monthly newsletter via Substack that rounds up interesting reading from the last 30 days (examples).
If you enjoyed this post, you might also like:
PR advice for those open letters and those marked-up responses
Image credit: Google Gemini and cited study
