Anna Otto, a computer science and online learning coordinator, went right to the best expert she could think of when she was designing a short course about artificial intelligence for middle schoolers: AI itself.
That was two years ago, when generative AI, the technology behind large language models like ChatGPT and Google’s Gemini was just emerging in the public eye. Even computer science experts like Otto were unfamiliar with it, she said.
“So, AI helped us understand itself, if we want to personify it,” said Otto, who works in the Adams 12 Five Star school district in Thornton, Colo., and was a 2025 Education Week Leader to Learn From.
A common danger for teachers who are planning with AI is that they might not realize or might not be familiar enough with the standards to say, ‘Oh, wait, that's not exactly the standard, right?'
The AI tool’s suggestions were helpful even though researchers have found that AI has a spotty track record when it comes to creating lesson plans and student assignments that get at deeper critical-thinking skills and present a range of perspectives.
“When you just say ‘create a lesson connected to this standard for this grade level,’ it tends to produce very teacher-centric, sage-on-the-stage type of lessons,” Otto said.
In designing the AI course, Otto got around that problem, in part, by being as specific as possible about the kind of lessons she wanted to see.
That meant prompting the tool multiple times or creating customized bots—which ChatGPT calls “GPTs” and Gemini calls “gems"—to guide her work. For instance, the district emphasizes formative practices, which aim to continually assess how well students grasp a particular topic in real time.
So, Otto put information on the strategy right into her custom bot, directing it to incorporate the approach into all the lessons in the AI course.
“The common challenge is just getting your prompt right, to get the information that you want formatted in the way you want, and to make sure that you’re creating student-centered learning opportunities as you develop lessons and activities for kids,” Otto said.
Even with those safeguards, AI made mistakes, Otto said. For instance, she asked the tool to embed standards developed by the Computer Science Teachers’ Association in its lesson plans.
The tool, however, sometimes put a new twist on a particular standard or even changed the standard so much it wasn’t recognizable. That happened most often when Otto simply referred to the standards by number rather than prompting the tech with the full text of a standard.
“A common danger for teachers who are planning with AI is that they might not realize or might not be familiar enough with the standards to say, ‘Oh, wait, that’s not exactly the standard, right?’” she said. “And then you’re suddenly off-course in the learning outcomes for your students.”
Otto recommends using AI’s suggested lesson plans as simply a starting point, not a finished product. “It’s never perfect, right?” she said. “That human in the loop thing is always critical. … You still need a base level of expertise to be able to evaluate [suggestions] and prompt effectively, right?”
