AI Wants to Make You Happy. That's Not Always the Job.

Happy Wednesday! This week:

Let’s get started!

I saw a post on LinkedIn this week — and I'm sorry to whoever wrote it, because I didn't save your name — that pinned down the exact moment you know you've become a communications professional. It's when a friend or colleague texts you about someone else's messaging disaster with the phrase: "Glad I'm not on that comms team."

We’ve all been there. But sit with what it says about the job. We tend to describe our work as cleanup — the statement that needed walking back, the post that should never have gone out. And yes, that’s important, but the real work happens earlier than that. We're the people who kill the bad message before it's a message, because once it's out, it isn't a draft anymore. It's a headline, an apology, and a very long week for all involved, including us.

I think AI is going to make that worse — AI is quietly making more of those bad messages look fine — and the data I uncovered this week supports this theory. AI chatbots cheer us on as we do our work, with subtle and not-so-subtle encouragement, and oftentimes, that encouragement is not only unwarranted, but plainly bad. In other words, our chatbots are giving us — and more importantly, our clients who are using AI before (or instead of!) consulting us — a false sense of confidence that the messaging we’re creating is going to work as intended.

The bad news is obvious — that’s a recipe for disaster. The good news is that, as communications pros, we’re unusually well-equipped to catch that type of mistake. AIs are trained to be sycophants. We’re trained to be cynics. It’s a perfect match.

The unusual positivity we get from AI isn’t just something you’re observing, either — it's one of the better-documented things about how these tools behave. In April 2025, OpenAI shipped an update to GPT-4o and pulled it back within a week, describing it in its own words as "overly flattering or agreeable — often described as sycophantic." They looked at their own product and said, out loud, that it was telling people what they wanted to hear. And they weren’t alone. Anthropic's researchers found that models will cave the moment you push back even slightly, and traced it to the training itself: one of the biggest predictors of a highly-rated answer was simply whether the model agreed with the person's existing beliefs. Salesforce researchers found that just asking "are you sure?" was enough to flip a model's answer — and that overall accuracy dropped when it caved, because the thing had usually been right the first time.

That last one is particularly ridiculous. It shows that the green light AI gives your message isn't a judgment. It's a reflex — built, at the level of its training, to make you feel good about what you just wrote.

And it isn't one rogue model. A 2026 study in Science, led by Stanford's Myra Cheng, ran eleven of the major chatbots — ChatGPT, Claude, Gemini, DeepSeek — through the same tests, and on everyday advice they sided with the user about 49% more often than a person would. Even when the user's plan was plainly a bad idea, the models went along with it 47% of the time. As Cheng put it: "By default, AI advice does not tell people that they're wrong nor give them 'tough love.'"

This is a communicator's nightmare — twice over.

Externally, it's the public misstep. The too-confident claim, the tone-deaf statement, the promise the company can't keep — drafted by a leader, run past an AI that cheerfully approved it, and now sitting in a reporter's inbox. That's the mess that gets someone else texting "glad I'm not on that comms team."

Internally, it may be worse. When a leader oversells AI — or anything — to employees and the reality doesn't match, the cost isn't a bad news cycle. It's trust, and a leader who's lost the trust of their people can't lead them anywhere.

But it’s also an opportunity to provide extraordinary value.

Many of us — myself included — have been saying that anything AI-written needs to be carefully reviewed before it goes out. That’s still true, but that only really protects us against random mistakes — a wrong fact here, a bad sentence there. The overly-positive chatbot isn’t making random mistakes, though. It’s creating a systematic lean in one direction — toward agreement, toward the rosier read, toward telling you your message is fine. And you can't spot-check your way out of a systematic bias, because it's not in some outputs, it's in all of them.

Our role is to supply the disagreement the tool is structurally incapable of supplying. Break that into what it looks like day to day:

Think of it this way: AI is all accelerator and no brake. It will take your message wherever you point it, faster than you could alone, and it will never once tap the brakes on your behalf. A car that's all gas isn't a faster car. It's a crash waiting to happen. We're the brake pedal — and the clock is running, because it's only a matter of time before a leader without a comms person in the room says something publicly, or to their whole company, because their chatbot was just trying to make them happy.

So: be the brake. And when you do, position it as being strategic — because it is being strategic. AI needs a partner, and you’re the one best positioned to deliver.

The tool your team adopts next quarter is in today's issue. TLDR AI is curated by Anthropic and ex-Google engineers and read by 1.1M+ people who like being first. Free, daily.

Subscribe for free

Many of us have long suspected the gap between what leadership says about AI and what employees actually experience with it was corrosive. Now there's a number attached to it.

Simon Hernandez-Arthur, reporting for Axios, dug into a new analysis from the AIDE Institute — a research and benchmarking group tracking enterprise AI adoption — that paired Glassdoor reviews with its own measure of how far a company's AI rhetoric has outrun its actual implementation. The finding: AI-related Glassdoor reviews were 27.5% negative at companies where leadership's AI signaling outpaced execution, versus 12.6% where both were strong, and just 7.2% where execution led the story. The fallout isn't contained to the AI conversation, either — employees who wrote negatively about AI rated senior leadership 2.04 out of 5, versus 3.34 among positive reviewers.

Chuck Gose — an internal comms strategist I know, and the founder of Icology — has the line worth pinning above every comms team's AI strategy doc: "Employees have always punished undelivered hype and broken promises. And AI is just the next thing to come along."

The story you tell about AI is only as strong as what people can actually see or use — and employees are the first people positioned to fact-check it against their own day. That's not a new problem, but AI raises the stakes: the gap between the pitch and the product shows up in someone's actual workflow within weeks, not quarters.

Yahoo's VP of global internal communications, Sarah Smith, put it plainly to Axios: "a completely baked strategy simply does not exist right now," so the honest move is telling employees "exactly what is solved today versus what the company is still actively figuring out."

Why it matters: A confident AI announcement your own team can't back up with their daily experience doesn't just fail to land — it measurably erodes trust in leadership.

Your next step: Before the next AI update goes out, ask whether it would survive being read next to what your employees are actually doing with the tools this week. If it wouldn't, that's the gap to close first — not the messaging. If you can’t? Change the messaging.

Everything I write about here I also do hands-on: helping comms teams bring AI into the digital channels they run every day — the intranet, email and newsletters, social, the web — and building the workflows that make it stick. It's the intersection of digital communications and real AI adoption, and it's the work I love most.

A handful of clients at a time; three slots open this fall. Book a 30-minute intro call →, or just reply to this email.

Gmail tracks you. Proton doesn’t. Get private email that puts your data — and your privacy — first.

Ditch the Gmail data grab

Looking for a role at the intersection of communications and AI? Here are some opportunities to check out:

Was this forwarded to you by a friend or colleague? Want to get this in your inbox next week? Click here to subscribe!

Quick question: how can I help?

What workflows are you struggling with? Where does AI still feel mysterious or overwhelming? What has worked that you’d like to share with others?

I’m a builder, and I’d love to help you and the rest of The Comms Stack community find great new ways to use AI.

Reply and tell me.
I read every response.

Even a one-sentence reply helps. For example:
“I wish AI could help me with ______.”

Until next Wednesday,

Dan

Advertise | Subscribe