Lessons Learned From My Six-Month Dive into AI

Happy Wednesday! This week, I’m celebrating half a year writing this newsletter, and looking back on what I’ve learned to date. I’ll be back next week with my more traditional approach — the news roundup, tools you should try, and more. For today, one ask: reply back to this email and let me know how you’re use of AI has changed since January.

Let’s get started, and thanks!

Dan

I started The Comms Stack six months ago — that’s about 25 newsletters sent, and a lot of things learned along the way. I’m going to use today to look back at what I’ve learned — and where I think communications is going. Almost everything sorts into two paths: the AI tools we use, and the way AI impacts our brand whether we use it or not. I’m going to talk about those two paths today — and then share some things I was wrong about.

As above, there are two aspects to AI that communicators need to be thinking about. Path A is how we use AI; Path B is how AI acts upon the work we do. In most of my conversations with comms pros, there’s a lot of passion around one of those paths, but the other gets mostly ignored. We need to go down both simultaneously.

Getting started on either can be overwhelming, as there’s no outlined maturity path for either. So, let me propose a pair today. Here’s a way to self-assess where you are on each of these journeys. Think of these steps as a guideline, not a bible; the constant under both paths is human judgment. (Always remember: our value comes from strategy, not stenography.)

My challenge to you: Read through these pathways, figure out what step you’re on, and make a concrete effort to move to the next step over the upcoming weeks.

Step 1: Learn the tricks of prompting. We’ve all (I hope!) booted up ChatGPT or Perplexity, etc., and tossed a few prompts in. But AI chatbots don’t have the context for what’s behind that prompt — we need to give it to them. Back in January, I proposed a prompting framework that outlines our thinking, and that’s a good start. Also, you should practice asking AI to write prompts for you — that will help surface your thought process to the chatbots.

Step 2: Understand how LLMs actually work. You don’t need to master the tech side, but you do need to understand where the pitfalls lie so you can avoid them. When your chats get too long, hallucinations increase. If you’re using the wrong model, you may get answers that don’t make sense. In February, I shared the difference between “non-thinking” and “thinking” models and why that matters to you. That distinction will go away at some point, but it’s still relevant today. Case in point: I ran a webinar for an agency yesterday and someone asked about the difference between Claude’s Haiku and Sonnet models, and if it mattered. I did the Car Wash test (click the last link, you’ll see!) and Sonnet passed the test, but Haiku failed.

Step 3: Go beyond one-off usage — integrate AI into your workflows (and build guardrails). This was a theme in many newsletters. A lot of employers and solo practitioners are still dabbling here and there. Real change happens when you integrate AI use into the work you’re already doing. It really doesn’t matter, at first, how you integrate it — any first step is a good one. Asking Copilot to give you a rundown of what’s in your inbox and on your calendar each morning. Running copy through ChatGPT before hitting send, even if you’re only using it as a glorified spellcheck. That habit will spark curiosity; you’ll find yourself seeing other ways to enhance your work using AI. But — again — make sure you do so with safeguards in place.

Step 4: Go from a consumer of tools to a builder of them. AI is transformative not because the existing tools feel magical, but because we can create solutions on our own. I’ve been advocating loudly for learning how to build your own, hosting webinars where I walk through, prompt by prompt, how to use Claude Code. I’ve shared how leading companies have cracked the nut on AI adoption — and again, the through line is clear: when we shift from users to builders, we unlock the real value AI can provide.

Step 1: Acknowledge the change. Our audiences are learning about us through AI-powered information gateways — AI chatbots, AI-powered search results, and more. This changes our work significantly. When I spoke with Noah Greenberg, CEO of Stacker, back in March, he made the point clearly: in an AI-mediated world, discovery isn’t just about publishing good content — it’s about making sure the right ideas surface when AI systems synthesize answers. Increasingly, that practice is called Generative Engine Optimization (GEO) or AI Search Optimization — and we need to understand it.

Step 2: Take ownership of this change. Actually, we need to do more than understand GEO. We need to own it. AI Search works differently than regular search engine optimization (SEO). As I outlined in April, it runs on earned media and narrative consistency, two things that communicators have already mastered. Many companies are treating GEO as SEO with a new coat of paint. This is a mistake, and — unfortunately — it’s one I keep seeing happening, over and over, without the comms team speaking up. My biggest concern about the communications in the AI era is that marketing teams are going to run forward with GEO while we choose to say on the sidelines.

Step 3: Discovery isn’t enough. Credibility is king. The other big change caused by the shift from traditional search to AI-mediated results? The audience doesn’t have to ever visit our websites in order to evaluate our offerings. When it comes to SEO, all you need to do is appear on the search engine results page — the searcher will click to learn more, and you can make the case for yourself there. In a GEO world, that’s not the case. Therefore, merely appearing isn’t enough. I spoke with Burson’s Steve Rubel last month, and he makes a great point: we need to tell our story through these intermediaries — and those stories need to resonate and be credible. And how we do that is still something we’re discovering.

Starting this newsletter, I was rather convinced of a few things, which, in retrospect, I shouldn’t have been convinced of. Here’s where I went wrong — and why:

1) Agentic AI isn’t the future — yet. In January, I took a course via the Harvard Data Science Institute on agentic AI — building AI agents that work with other AI agents to do complicated work. I was convinced that this was going to be the path forward — humans as orchestrators of robot armies. I think that could still be the case, but we’re not there yet, and may never be. Today, I actively advise against this type of investment. People make mistakes — but so do the bots. But because we’re doing so much less and because we need to sleep and eat, there are fewer opportunities for mistakes. Plus, we have that little voice in our heads that can clue is in when something is amiss. AI bots have neither such limitation. In the time it takes us to make an uncaught, catastrophic error, the machines can make hundreds. Fully agentic AI simply isn’t worth the risk — and that’s an understatement. I’ve come to abhor the phrase “human in the loop” because of that pitfall. We should be the focal point of the loop, and we should introduce AI as needed.

2) Companies aren’t encouraging employees to be curious about AI. The best way to move down the AI adoption path is be curious — experiment, learn, repeat. But you can’t experiment without an environment that lets you do that. The biggest bottleneck to AI adoption, particularly at large organization, is still access to services like Claude and Perplexity. I figured that would be solved by now — but I was wrong. Just last week, I spoke with a leader at a larger agency who described their tooling as “the Fisher-Price version of AI” — but no one is complaining because that’s the norm across the industry. AI adoption is lagging way, way behind where I thought it’d be by now as a result. I still think all of us will be using Claude Code or similar platforms in the future. I no longer think that will happen in the near future.

3) AI writing is bad and everyone seems to know it. I didn’t think AI writing was going to replace communicators. I did think it was going to fundamentally reduce the amount of writing we’re doing. I don’t think that any more. AI-generated copy is obvious, and increasingly so — and I think we’re approaching the “uncanny valley” phase where not only is AI written copy obvious, but it’s also eerie to the point of being unsettling. I now advise people to generate first drafts to get past blank page paralysis, read what the chatbot wrote, and then throw the whole thing out and write it yourself. AI writing isn’t getting worse, but we’re getting better at recognizing it and seeing it for its faults.

That’s what I’ve learned so far — but if this experience has taught me anything, the next six months will make these lessons seem quaint. The AI world is moving quickly, with change happening faster than many can keep pace with. At some point, that will slow down, but I don’t think we’re there yet. So, I’ll keep learning.

Thanks for being here for the first 25 or so weeks — let’s see what the next 25 bring.

Posting everywhere means rewriting one idea six times, so you post to one, or none. SureThing turns one idea into native posts for every platform.

Put It on Autopilot

AI adoption is critical for communications teams — and I can help. Let’s work together to get your team moving — or moving faster.

📋 A prompt playbook. Need a kickstart? My communications-focused prompt playbook has more than 600 prompts to help you get started. Download it, free.

🎓 Self-paced learning. I’ve created six courses to get you up to speed on how AI impacts communications. Get started here.

📅 A strategy call. Whether you're trying to build systematic AI adoption across your team or focused on AI search visibility for your brand or organization, I’m here to help. Learn more and book an intro call.

#CommsJobs started with a problem nobody was solving.

Daniel Méndez Aróstica had just moved countries and was rebuilding his network from scratch. His feed began filling up with comms roles, so he started sharing them with his network. That network is now 25k strong.

Every Friday, the #CommsJobs newsletter drops the week's comms roles straight to your inbox. Hand-curated. No spam, no filler, and never a charge to the communicator looking for work. 

Your next role is hopefully in this week's list.

Subscribe and go find it.

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