Ann Handley posted something on LinkedIn last week that stopped me mid-scroll. She’s a Wall Street Journal bestselling author and one of the most respected voices in marketing, and she wrote:
“AI literacy is not prompt literacy. It’s judgment literacy.“
Her post went on to ask a question that nobody in the AI training industry seems to be asking: “Why do we keep teaching people how to use AI – without ever teaching them when not to?”
I messaged her. I had to know where someone would go to learn that.
Her honest answer: “I don’t know of a course that teaches exclusively this. At MarketingProfs, our sessions about AI typically include a few slides that touch on when not to use AI, or how to protect against hallucinations, but I don’t know of a whole session or series.”
She added, “I think that’s actually the story, and why I wrote what I wrote. We have an entire industry built around AI skills training – prompt engineering bootcamps, certification programs, tools tutorials, a million LinkedIn posts about the perfect prompts you need to do this or that or else you’re falling behind. What we don’t have is anything that asks: when should you put the tool down? When does using it cost you something you didn’t mean to give up?”
That gap is real, and it matters more than the AI training industry currently acknowledges.
Prompt Literacy Takes An Afternoon. Judgment Literacy Takes Years
The distinction Ann draws is not subtle once you see it. Prompt literacy is teachable in an afternoon. You learn the syntax, the structure, the iterative refinement loop. You learn to be specific, to add constraints, to tell the model what not to do as well as what to do. This is genuinely useful and genuinely learnable quickly.
Judgment literacy is something else entirely. It is knowing when the speed of AI output is actually eroding something you needed to build slowly. It is recognizing when the struggle itself is the point, when the friction of not knowing the answer yet is what produces the expertise that will matter later. It is understanding, as Ann put it, “when AI helps and when it shortcuts the very struggle that teaches us something.”
One commenter on her post put it precisely:
“Prompt literacy is teachable in an afternoon and judgment literacy takes years, because judgment is mostly knowing the value of the struggle you’d be skipping.”
I’ve been teaching an online course on AI content that audiences actually trust for several years. And I’ve spent recent months analyzing what the AI training landscape actually offers practitioners. The pattern is consistent. The courses that exist (and there are now many of them) teach you what tools can do. The better ones teach you how to deploy them strategically. Almost none of them teach you when to put them down.
This is not a minor gap in the curriculum. It is the central question of the current moment.
Why The Gap Exists
The AI training industry has a structural incentive problem. Courses that teach you to use tools generate demand for more tools, more courses, more certifications. There is no business model for teaching restraint. Nobody is building a prompt engineering bootcamp whose primary lesson is “sometimes don’t.”
But the cost of skipping the judgment question is real and measurable. Anthropic’s own research found that junior engineers who leaned heavily on AI coding agents demonstrated weaker understanding of their work when tested afterward. When the tool produced output, their struggle that would have built expertise did not happen. The output and the expertise are not the same thing.
For SEO professionals and content marketers specifically, the exposure is direct. MIT’s AI Labor Exposure Map, which I wrote about last week, found that nearly three-quarters of the time a marketing specialist spends at work goes to tasks that AI can already handle. The question is not whether to use AI for those tasks. For many of them, you should. The question is which tasks in that 74% are actually the ones where the doing is the learning, where outsourcing the execution also outsources the understanding you needed to build.
That question requires judgment. It cannot be answered by a prompt.
Culture, Not Coursework
When I asked Ann where practitioners should go to develop this judgment, her second message reframed the question entirely.
“Do we actually need a course? What we need instead is permission and better modeling. Leaders who visibly choose the long road. Managers who say out loud when they are not going to use AI for certain things, and here’s why. Individuals who see the value. Said another way: culture not coursework.”
That reframe is worth sitting with. The judgment about when not to use AI is not a skill that gets transmitted through a certificate program. It is a professional norm that gets transmitted through observation, through watching someone you respect make a deliberate choice to do something the slow, human-fumbling-in-the-dark way, and then explaining why.
Ann has a book coming out in February 2027 from Penguin Random House called “ASAP (As Slow As Possible): When to Take the Long Road in a Shortcut World.” The title captures the tension precisely. In a professional culture that has made speed the primary virtue, choosing slowness requires not just judgment but courage: the willingness to be seen taking longer when everyone around you is accelerating.
What Practitioners Can Actually Try Right Now
Ann’s point about culture rather than coursework is correct in the long run. But while that culture is still forming, practitioners need something concrete. Here is a workflow worth replicating, drawn from an experiment I ran with the editorial team at The Acton Exchange, a nonprofit community newspaper in Acton, Massachusetts, in November 2025.
The team faced a deadline problem. A steering committee had just held a three-hour working session on a critical school district reorganization question, reviewing 156 pages of materials. The meeting wasn’t recorded, which meant no transcript was available. But the 101 pages of supplemental information and 55 pages of public comments the committee had received ahead of time were accessible.
So, the team tried something new. We crafted a detailed prompt specifying what the article needed to accomplish: accurate and trustworthy information, a compelling story, relevant to residents. We uploaded all 156 pages to four AI engines simultaneously: ChatGPT, Gemini, Perplexity, and NotebookLM. Each engine took a different route from the same prompt and the same source material. ChatGPT produced 748 words focused on data and process. Gemini produced 712 words focused on why the status quo was no longer viable. Perplexity produced 1,232 words focused on what the options meant for residents. NotebookLM produced 1,506 words organized around five surprising truths.
We reviewed all four drafts together at an all-hands editorial meeting. Perplexity’s draft was the most accurate and the most useful as a foundation. We chose it as our starting point. Then we did what no AI engine could do: We added direct quotes from people who were in the room, reflecting the community voices that the Acton Exchange exists to represent.
The key lesson from this experiment is not which engine performed best. It is what the process revealed about judgment. Town Manager John Mangiaratti had observed a few weeks earlier that the tools were helpful for the first 75% of content, but that “the remaining 25% of details, nuance, and context are either missing or incorrect.” Superintendent Peter Light agreed, adding that quality improves with better input prompts.
That 75/25 split is a practical frame for any content workflow. Use AI to get 75% of the way there quickly. Then apply human expertise, primary source verification, and direct observation to close the gap. The 25% that requires a human is not a bug in the workflow. It is where the judgment lives.
Before adopting any AI tool in your content process, have an explicit conversation with your editor or team about which tasks the AI will handle and which require human oversight. Document your prompt. Run the same prompt through more than one engine when the stakes are high. Verify outputs against primary sources before publishing. And disclose your process to your audience, as the Acton Exchange did at the foot of this published article.
Ann Handley is right that the real skill is judgment: knowing when speed is useful and when it actually erodes something you needed to build. The Acton Exchange experiment didn’t resolve that question. It made the question visible in a way that a prompt engineering course never would.
Prompt literacy gets you to 75%. Judgment literacy is what closes the rest.
More Resources
Featured Image: Yuriy2012/Shutterstock