Happy Wednesday! This week:
Why almost nobody in comms is in the middle on AI — and why the gap widens even if you do nothing
Claude is marking everything it touches. That's not the problem.
Quick Hits: the heaviest AI users in your company aren't the people you'd guess
Let's get started!
THE LEDE
💡 Everyone Thinks They're in the Middle of the AI Adoption Curve
As a job seeker, fractional comms/marketing leader, consultant, and “guy with a lot of side projects,” I am constantly networking. In any given week, I end up talking with a lot of communicators, at different companies, at very different points in their careers. That's not a research method, it's a side effect of being between things and wearing too many hats. Regardless, in my position, you start hearing patterns, and there's one I can't stop noticing.
When it comes to AI adoption, almost everyone seems to land in one of two groups. The first is using the chatbots as search engines and maybe for some research, and as a writer/editor — and that's roughly where it stops. The second has built something — a website, a workflow, a custom GPT — but it’s rarely only one thing. Once someone has built one thing they've usually built four.
What I almost never meet is somebody in the middle — the person who tinkered once and stopped. There’s a divide occurring, and I think most people don’t see it. Most people think they’re in the middle of the adoption curve… but I increasingly am seeing that there’s actually no one in the middle.
I’ve been a digital channels strategist for more than fifteen years, and this is the first time I’ve really seen this. I had the privilege of being Sesame Street’s first dedicated social media lead and ultimately built out a small digital marketing and communications team; we were the digital adoption team, and my colleagues were focused on more traditional ways of reaching audiences. No one really “dabbled” in social or websites or newsletters.
We’re seeing a similar adoption curve today, but because people who aren’t focused on AI are still using AI to some degree, they’re not seeing themselves as falling behind. They’re using AI more than they used, say, Twitter back in the day — and therefore, they think they’re in the middle of the AI adoption chart. But I don’t think that’s the case. The data suggests I’m right: A few weeks back I cited the Axios HQ and Off the Record numbers here, showing that while 96% of comms leaders use AI at least weekly, only about 40% have built a custom GPT or an AI-powered workflow, or anything else of that sort. That’s evidence of a bifurcation, not a smooth adoption curve.
That divide is dangerous if you’re hoping to stay current. AI isn’t like social media or SEO or spinning up a Substack. AI isn’t a digital channel. It’s a tool. There’s an emerging gap between those who use it to build things and those who, well, don’t.
Here's the catch: the gap widens even if you do nothing.
Think about a stocked pantry versus ordering takeout. Takeout is the same price every single time. A pantry costs more up front, and then every meal after that gets a little easier. Building with AI is like stocking that pantry — the context files, the voice profile, the notes you've already fed the thing are all still sitting there when you start the next one. Using is like ordering takeout. Every session starts cold, and there's a ceiling on how good you can get at ordering. Every day you wait to stock up that pantry, you reinforce the DoorDash-a-burrito habit, while the people who have a full spice cabinet become better and better cooks.
And the numbers here are worse than I expected. Per Semafor, OpenAI looked at its own enterprise customers: in January, firms in the top 10% of usage burned 2.6 times as many tokens per user as those in the median company. By June it was 8.3 times. Five months in, and the distance between the top and the middle roughly tripled. The middle has continued to evaporate.
And yes, there are some issues with that data point. Most notably, this is OpenAI counting OpenAI's customers and handing the study to a reporter, so read it the way you'd read any research a company commissions about itself. But if it’s right — and I’m confident it is directionally correct — the people on the builder side of the divide are lapping everyone else.
Why this matters for us: comms gets handed the change story after the change is decided. If the distance is growing on its own — and I think it is — then the team that assumes it's mid-pack isn't holding steady, it's sliding while feeling fine about it. You find out the day leadership asks comms to explain an AI strategy nobody on the team has the muscle to execute.
Your next steps:
Answer this first — and be honest. When did you last end a week with something that didn't exist on Monday? Not a document — a thing, one that does work for you. If nothing comes to mind, you have your answer, and it isn't the middle. (A year ago I'd have had nothing to point at either.)
Take one thing you did twice last month — a recurring report, a briefing doc, a media list refresh. Give it fifteen minutes with Claude or ChatGPT, and don't ask it to do the task. Ask it to help you build the thing that does the task. That's the difference between using and building, and I think you’ll find it easier than you’d think.
Save whatever comes out of that process somewhere that you'll find it again. Drop it in a Word doc — that’s fine. There are better ways to do it (e.g., a GitHub repo) but I don’t want to scare you away. Start with what you know — build your pantry, and grow it at your own pace.
SPONSORED BY
AI news from people who build AI
TLDR AI is the free daily brief curated by Anthropic and ex-Google engineers. The stories, models, and research they'd send a colleague, summarized for 1.1M+ readers.
THIS WEEK IN AI
🌎 Claude Is Marking Everything It Touches. That's Not a Problem.
Anthropic announced last week that Claude now watermarks the text it produces. Not a visible tag — the mark is woven into the word choices themselves, so it allegedly survives a copy/paste, a retype, a move into your CMS. Every model released after August 1, 2026 does it automatically, at no cost and no extra tokens, along with signed provenance data on image files.
One of the best things I’ve read on this came from Gab Ferree, whose Comms Bestie Group Chat newsletter goes out free every Friday. Her headline was "Don't panic about Claude's watermark," and her reasoning is sharp: a mark doesn't prove Claude wrote something — it may attach to anything Claude merely processed, including a translation, a summary, or a proofread. And detection isn't even live yet; Anthropic says the technical details are still forthcoming. But then she makes the turn that I think is the actual story. She names the absurd dichotomy we're all living inside — companies bragging that they're AI-first and laying people off because they're so efficient, while simultaneously gasping "you used AI to write that?" — and argues the watermark is good news, because it forces the transparency that ends the double standard. Her prescription is two documents: a personal AI operating doc that says what you do and don't use AI for, and a company AI philosophy that treats your AI stance as a brand promise. She practices it, too — Off the Record's ethics page is public, and she updated her own disclosure this month to note she uses AI for some graphics built on her human-made brand guidelines. "Now you don't have to worry about getting 'caught' using Claude. You've been clear about how you do and don't use AI."
(Full disclosure, since that's rather the point of this whole piece: I'm a paying member of Off the Record, and I'm an affiliate too — if you join through that link, I get a cut. I'd recommend it either way, and I did for months before the affiliate part existed. It's 500+ comms leaders with real training, coaching and community behind it, and the Office Hours alone have earned back what I pay. And if paid communities aren't your thing, no worries — her Friday newsletter is free and public, which is how I found her in the first place.)
I agree with Gab, and I'd push it one step further. The reason not to panic isn't that the mark is unreliable — it's that being marked is going to be the normal condition of professional writing. Nobody has ever been shamed for using spellcheck or a word processor. An LLM is going to touch most published content at some point in its life, and a binary "AI was involved here" signal tells a reader almost nothing worth knowing. Which is not a reversal of what I argued a few weeks ago about AI-written copy — I still think shipping late-stage AI drafts is a mistake, and audiences still clock the uncanny valley instantly. But that's an argument about whether AI produces your voice. This is an argument about whether it touched your process. Those are different questions, and conflating them is how good teams end up hiding ordinary work. (I’m still not over the idea that you should not use em-dashes in hopes of avoiding the scarlet letters “AI.”)
Where I do have a real concern is depth. A yes-or-no mark is fine. Detailed usage metadata — how much was AI, which parts, how often you leaned on it — is not, and I don't think we've thought hard enough about who that lands on. Plenty of people use these tools differently by necessity, including neurodivergent colleagues for whom AI is an accessibility tool, not a shortcut. A disclosure regime that reports how much help someone needed isn't transparency. It's surveillance with better branding, and comms teams should say so now, while the standards are still being written.
Why it matters: The watermark doesn't create a disclosure problem for your organization. It reveals the one you already had — and the gap between what your company says about AI publicly and what it lets people admit internally is about to become legible to outsiders. As communicators, we should make sure our companies get ahead of this problem.
🎯 Quick Hits
I’m not quite sure if this is comms-adjacent or not, but it’s good to know if you care about GEO. Time Magazine, per Digiday, came up with a new ad format, one that was only visible to AI agents. It’s a neat strategy — but it’s not clear whether the LLM leaders are going to play nice. Two weeks after the news hit, Perplexity told Digiday that they were blocking those ads from appearing in their engine.
A new working paper — "How Organizations Use AI: Evidence from ChatGPT", from researchers at OpenAI, Columbia and Wharton — went through more than 17 million messages across 1,764 companies on ChatGPT Enterprise. They found that early-career workers and trainees send roughly eight to nine more messages a week than the average active user at their own company; executives, founders and partners send fewer. I argued in July that the answer to AI is to hire more juniors, not fewer, and I'll take the outside evidence. But here's the finding I didn't expect: the other group sending more than its share is marketing and communications. We're about 5% of the active users at the average firm and we out-message nearly everyone above us. (Note that this is OpenAI's own data about its own product again, so same caveat as above.)
AdWeek had NewtonX survey 500 marketers, and the two headline numbers only make sense sitting next to each other: 69% say their company mandates or strongly encourages using AI, and 43% say they've had no formal training on it. Of the ones who did get training, only 10% called it adequate. Telling people to use the thing and then not teaching them how is how you end up with half a department quietly guessing — and then quietly deciding it isn't for them. But the number that stopped me is further down: 81% say their organization overstates its AI proficiency, and they rate their C-suite below their tech teams, their middle managers and their junior staff. Or, as the report concludes, "the people making AI strategy decisions are the least equipped to evaluate them." That's the same gap the ChatGPT study above found from the other side — the people setting the strategy are the ones with their hands on it least. Fifteen minutes of somebody showing you what good looks like beats a year of being told to try harder.
Not really about communications, but fun: Fast Company covered ChatTJB, "the leading chat interface powered by AI" — but “AI” doesn’t stand for “artificial intelligence” for this one. It stands for “actual individual,” and specifically, Tucker Bryant, a former Google marketing manager. Bryant spent $6,000 on a downtown San Francisco billboard in July, and every prompt you type goes straight to Tucker, who answers it himself. He was fielding about 1,000 a day until the billboard went viral and the queue hit roughly 5,000 an hour, at which point he had to pull the site down. Too bad — but no, I won’t be launching ChatDNL.
I’M HERE TO HELP
🤝 Your Comms Team — Leveled Up
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.
SPONSORED BY
Master Claude AI (Free Guide)
The professionals pulling ahead aren't working more. They're using Claude.
Our free guide will show you how to:
Configure Claude to be the perfect assistant
Master AI-powered content creation
Transform complex data into actionable strategies
Harness Claude’s full potential
Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.
COOL AI TOOLS
🔨 Tools To Try This Week
*sponsored, but still awesome
Was this forwarded to you by a friend or colleague? Want to get this in your inbox next week? Click here to subscribe!
YOUR FEEDBACK WANTED
🔊 Help The Comms Stack Improve
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