Using AI to Find the 'Invisible Threads' as an Instructional Coach
Jan 05, 2026by Matthaeus Huelse
When Coaching Volume Becomes a Challenge
When I first started coaching at an elementary/middle school, my first and primary concern was not having the subject matter expertise needed to be an impactful support for my teachers. As a former German teacher (and also a native speaker), I felt like a fraud when talking to any educator, whose subject I still held fear for from my own time as a student. Sure, I could support most second language acquisition teachers, and I could muddle through subjects I enjoyed in high school and college, but you could chase me out of the room with any math or physics unit. Turns out, I was wrong. Because the toolkit I had to rely on wasn’t my subject matter related one, but my pedagogical intuition and experience. Once I realized that, my challenge changed. With my toolbelt secured and my sights on the right aspect of coaching, I encountered a new challenge: Volume.
As coaches we find ourselves going in and out of classrooms with different student dynamics, varying teacher strengths, and a multitude of factors and variables that will affect real learning outcomes. We tackle an enormous variety of qualitative data: formal surveys, informal hallway chats, classroom observations, and random sticky notes we write ourselves. Not only did I have to learn an efficient way to keep all that information neatly organized, categorized, and up-to-date, I then had to actually sit down and make sense of it. If you are working with one or two teachers that might be manageable, but with coaching success, comes a growing caseload. How often do you find yourself looking at hastily taken notes in your notebook, unnamed google docs with scattered bullet points, or trying to remember whether it was this teacher that needed support with student collaboration or the other one? Start throwing coaching cycles into the mix, and your brain will wave the white flag, trying to keep it all organized and accurate.
The Trap of Recency Bias
Even after committing to a standardized system to keep my notes straight, I noticed that after a while, every “problem” began to look the same. Why was every teacher receiving the same tips and supports, even when I was honestly trying to personalize my feedback? Recency bias had taken hold. Suddenly, I was pulling the same tool from my belt over and over - like a plumber trying to use a hammer to fix every pipe. Was I truly finding trends across the faculty? Or was I just sticking to advice that worked once and generalizing when I should have been narrowing down? I don’t claim to own a silver bullet for these bias traps. However, I do want to share how I started using AI to synthesize solutions, rather than just summarize issues.
From Summary to Synthesis: Re-framing the Role of AI
Let’s break that down, because the distinction matters. I frequently tell coaches and teachers that AI can help save time, but most of us approach it with the sole goal of receiving a final product - whether that is a lesson plan, differentiated readings, or a quickly generated report. Certainly, I can ask AI to catch me up on a TLDR email chain or pull my scattered notes into a coherent summary. But the real magic lies in AI becoming more than just a task-master: it should be a brainstormer and a thought partner.
The "Human-in-the-Loop" Principle
However, before I tell you more, I need to make one thing abundantly clear. I am not at all suggesting that we outsource our own expertise, experience, and knowledge to an algorithm. I sincerely believe that is the worst possible outcome. Rather, I want us to realize that AI is only as useful as our ability to recognize a good output from a bad one. That requires us to use our own background and acknowledge that we are - and should always remain - the most valuable part of the equation when it comes to working with AI. We call that principle “Human-in-the-loop”.
This is exactly why I have shifted my focus toward tools that prioritize "grounding," specifically Google’s NotebookLM. Unlike a standard chatbot that pulls from the entire internet (and occasionally hallucinates facts to fill in the gaps), NotebookLM works within a "walled garden." It only knows what you feed it. If I upload ten PDFs of anonymous observation notes and survey results, it answers my questions using only that information. But the feature that truly keeps the "human in the loop" is the citation system. When I ask it to identify a trend, it doesn’t just give me a paragraph of text; it gives me footnotes. I can click a citation and immediately see the exact source - the specific sticky note or survey response - that generated that insight. It shows its work. This allows me to validate the AI’s findings against my own intuition, ensuring that I am not just trusting a machine, but verifying a pattern.
Asking Better Questions: Finding the Invisible Threads
So, how does this move us from summary to synthesis? It starts with how we talk to the data. Once I have my sanitized notes uploaded (more on safety in a moment), I stop asking for efficient summaries and start asking for connections. Instead of asking, "Summarize the feedback from the 3rd-grade team," which simply creates a shorter version of what I already read, I prompt for the invisible threads: "Based on these observation notes, what are the conflicting values between what teachers say they want in the survey and what is actually happening during instruction?" or "Identify the emotional through-line connecting the frustrations of the math department with the literacy team." Suddenly, the AI isn't just organizing my files; it is acting as a spotlight, illuminating trends that were buried under the sheer volume of the noise, and giving me a hypothesis that I can then go out and test.
Protecting Trust: The Non-Negotiable of Data Sanitization
Of course, none of this matters if we break trust in the process. As coaches, our currency is relationships, and nothing bankrupts a relationship faster than a teacher feeling like their vulnerability has been exposed to a machine - or worse, leaked. This is why "sanitization" is non-negotiable. Before I ever upload a PDF or a set of notes, I scrub the data. Names become "Teacher A," specific room numbers are removed, and any identifying student details are stripped away. I am also radically transparent with my staff. I tell them, "I use AI tools to help me find the big themes in our collective feedback so I don't miss anything, but I never feed it your names or personal details." And finally, I practice strict data hygiene: once I have synthesized the insights and verified them, I delete the files from the notebook. We must treat this data with the same reverence we treat a confidential conversation in our office.
Conclusion: Reaching for the Right Tool
Ultimately, using AI in this way isn't about looking for a shortcut; it's about clearing the deck so we can do the actual work of coaching. By offloading the cognitive load of sorting, categorizing, and remembering the sheer volume of data we encounter, we free up our mental energy for what really counts: the human connection. We stop guessing what our teachers need based on who we spoke to last, and start supporting them based on what they are actually telling us collectively. It allows us to put down the "hammer" I mentioned earlier and finally reach for the right tool. So, I challenge you: take that stack of sticky notes or that folder of exit tickets sitting on your desk, sanitize them, and see what "invisible threads" you can find. You might just hear exactly what your teachers have been trying to tell you all along.
💾 Want more?
- 📚 Read
- 🔊 Listen to the
- 🎬 Watch (EDU Coach Network All-Access Members only)
Stay connected with news and updates!
Join our mailing list to receive the latest news and updates from our team.
Don't worry, your information will not be shared.
We hate SPAM. We will never sell your information, for any reason.
