How Writers Can Beat AI At Its Own Game

Today my company launched AI for Speechwriting and Executive Communication, a webinar that should signal to you that this stuff is in the leadership communication biz to stay.

The question is, what role it will play.

The lead instructor, Brent Kerrigan, is a deeply humane speechwriter who optimistically sees AI as a chance to separate real leaders from pretenders, to “start separating those who actually know what they’re talking about from those just reading AI-generated talking points.”

Okay: Since at least the advent of YouTube, we’ve known that speeches are anachronous rituals that must justify their inherent inefficiency and absurd inconvenience by giving an audience a true sense of communication occasion—an unmistakable communal feeling that only this speaker could have delivered only this message to only this group of people only at this moment in time.

Large language models can’t do that, by definition. They can only regurgitate combinations of noises other speakers have made to other groups of people at other moments in time.

And so using ChatGPT to build in a little front-end efficiency for someone composing a speech is actually a preposterously short-sighted thing to even try to do. When in fact what the speaker and speechwriter must do, in fact, is to show the audience that this speech could never in a million years have been written by artificial intelligence.

Matthew Brophy, a philosophy professor writing for Inside Higher Education, recently got AI to “meet the moment” by writing a passible commencement speech: “I prompted four of today’s most advanced large language models: ChatGPT-4, Claude 3.7 Sonnet, Gemini 2.5 and Llama 4 Maverick. Each generated a full address, filled with advice, encouragement and reflections on what it means to step into the unknown. I then asked ChatGPT-4 to synthesize them into a single speech that expressed a slightly more playful tone.”

Reading the result, one must acknowledge that it reproduces smooth commencement-speech patter with requisite mild humor and banal advice. And of course that’s what most commencement speakers do wind up delivering, in the end. So you might conclude this is as good as any.

No. Precisely because AI is available now, speakers and their speechwriters are going to have to prove that AI didn’t write their speeches. How? By making them so personal—so uniquely about themselves, about the audience, about the moment at hand—that people will know this speech couldn’t have been done by AI. And as AI gets better and better at aping authentic human communication, it will require real human ingenuity to demonstrate incontrovertible humanity at the lectern. But then, incontrovertible humanity is what the best speechwriters have brought to this work all along.

Even workaday speechwriters have been crafting such proofs since time out of mind: most typically, in the form of what we in industry circles call the “howdahell.” Wherein a speaker, usually early in the speech, mentions a local event or place that’s so obscure outside the speech location or the audience’s narrow focus as to force to the audience to say in wonder, “Howdahell did [insert speaker] know about [insert local watering hole, beloved local character, minor local controversy]?” Flattered that Mr. or Ms. Big (or his or her big speechwriter) bothered to learn something about us, the audience settles happily into a contented, receptive haze, the speaker’s work already half done.

In the age of AI, speechwriters are going to have to do a hell of a lot better than that to make audiences feel directly and genuinely addressed by another human being sufficiently to believe the gathering was worth their while, individually and as a collective. I mean, can you imagine taking a few days out of your life to fly to Seattle only to hear some big shot give a speech you suspect was written in two seconds by Claude?

So yes, register for this AI webinar, to learn how AI can help you in various aspects of speechwriting and leadership communication—research, audience analysis, argument-assessment, brainstorming, readability testing, scenario-planning and message planning.

And use the exercise to start to figure out how you’re doing to beat AI at its own game—by remembering with every word you write that for you and your speaker and your audience, this isn’t a game at all.