The AI that helps instructional coaches most does four jobs. Gemini Notebook turns de-identified observation notes and district policy into coaching prep, AILitKit builds AI literacy PD, Brisk Teaching and ScreenPal model feedback, and Otter.ai transcribes adult-only debriefs. SchoolAI shows staff what monitored student AI looks like. Each has a free tier, and none should feed a teacher's evaluation.
Prices, free tiers and privacy pages were last checked at each vendor's own site on November 12, 2026; what changed is listed at the foot of the page. Gemini Notebook is the tool Google renamed from NotebookLM in July 2026.
Which AI Tools Do Instructional Coaches Use, and What Does Each Cost?
Six tools, each matched to one coaching job. The privacy column is what each vendor declares on its own pages, not an audit; the 9ine assessment is the independent one.
| Tool | Best use | Strengths | Limitations | Price / free tier (checked November 12, 2026) | Privacy considerations |
|---|---|---|---|---|---|
| Gemini Notebook | Coaching prep from your own notes, research and policy | Answers from the sources you upload, with in-line citations; audio overviews for PD | Free tier has no published numbers; personal accounts differ from Education accounts | Free; Education Fundamentals and Standard 100 notebooks, 50 sources each, 50 chats a day | School accounts: Google says uploads, queries and responses are not human-reviewed or used to train models; FERPA unknown in directory |
| AILitKit | AI literacy PD and lesson guides for non-specialists | Coaching notes on every activity; unplugged activities; mapped to 12 frameworks | Teacher-only; priced in pounds; one review | 3 free guides then 1 a month; £59 a year per teacher; £249 a year per department (5 staff); £999 a year per school (up to 50) | Directory record declares FERPA, COPPA, GDPR; EU-hosted; no pupil accounts; 18+ |
| SchoolAI | Showing teachers what monitored student AI looks like | PD Hub stays on the free trial; teacher sees every student chat | Free trial is 5 Space sessions a year; paid plans only via school or district purchase | Free trial (5 Space sessions a year); Pro and Scale: no public pricing | Vendor declares FERPA, COPPA, SOC 2 Type 2; US-hosted; collects student data; no minimum age |
| Brisk Teaching | Modeling AI feedback inside the Docs teachers already use | Works in Google and Microsoft; $0 plan with 20+ tools; 9ine-assessed | Feedback tools are built for student work; free batch feedback has an unstated upload limit; school prices not shown | Free ($0, 20+ tools, Standard model); Premium and Intelligence: no public pricing | Says it acts as a FERPA "School Official"; may use de-identified data to improve its service; 9ine-assessed |
| ScreenPal | Recording model lessons and video feedback | Free plan listed with no watermark; quizzes; low education pricing | 15-minute cap on free recordings; 10 hosted videos; 5 free video quizzes; automated captions need a paid plan | Free; Deluxe Edu $3 a month annual; Max Edu $6 a month annual; Team Education $3 a user a month (10+) | FERPA "School Official" clause; iKeepSafe certified; US-hosted; 60-day deletion for schools |
| Otter.ai | Transcribing adult-only debriefs and PLC meetings | 300 free minutes a month; summaries and action items; 9ine-assessed | Trains its AI on de-identified recordings; no FERPA clause; 13+ | Free (300 min a month, 30 min a conversation); Pro $16.99 a month or $8.33 a month annual; 20% education discount | Not for classrooms with students; check your state's recording-consent rules |
Why Does the Line Between Coaching and Evaluation Matter This Year?
None of these tools coaches a teacher. That is still your job, and this year you will be asked to defend where it stops. On September 1, 2026, Washington DC's Office of the State Superintendent of Education published a model AI policy that puts "evaluations of teacher performance" in its red, prohibited category and "providing supplemental coaching" in yellow, allowed with safeguards (OSSE). That line, coaching yes, evaluation no, is the one every coach will be asked to hold this year. Meanwhile an NPR/Ipsos poll of 545 K-12 teachers found that only 33% say their school has a formal policy on student AI use, and 52% say their school has offered no guidance or they are not sure what it is (Ipsos, June 2026). Who fills that gap in most buildings? The coach.
This post is for instructional coaches, and for the principals and directors of curriculum (chief academic officers, in larger districts) who decide what coaches are allowed to use. It is not a list of tools for teachers; we have that post already. If you are the one writing up the observation, the tools differ, and we cover them in AI tools for classroom observations and AI tools for principals. The second half of this page answers the question coaches ask more often than any other right now: how do you actually coach teachers into using AI, when most of them have had no training and some of them do not want any.
How This List Was Built
On November 12, 2026 the pool was the 19 tools in the aieducator.tools directory that list professional development as a use case, plus the 33 in the collaboration and communication category. Each was judged on four things: whether it does a job a coach actually has (prep, PD design, feedback, documentation), whether the vendor documents what happens to recordings and notes, whether it can be used without ever touching student data, and whether the pricing is honest enough to put in a budget request. Nobody pays to be included, and vendors cannot buy placement.
Two tools were left out on purpose. TecAdRise sells services and training on a quote basis rather than a product that can be assessed from its public pages. Alder Branch, which the first version of this article excluded for its thin privacy documentation, no longer appears in the directory. Two well-known coaching platforms, Edthena's AI Coach and TeachFX, are not in the directory, so they get a verdict line below rather than a full section.
Pricing, free tiers and privacy pages were first checked on September 7, 2026, read again on September 20, and read a third time on November 12, 2026. The third reading moved six facts, and each correction is in the section it belongs to and in the change note at the foot.
What the Directory Numbers Say About PD Tools
Of the 19 directory tools that list professional development as a use case, 8 declare FERPA compliance, 8 state that they collect no student data, and only 2 carry the independent 9ine security assessment. Six name the United States as where data is hosted, three the European Union, two the United Kingdom and two Singapore; the other six give no verified location. Across the whole directory of 138 tools, 44 leave their FERPA status unknown. An unknown is not a failing grade; it means the record has not been filled in, and it is a prompt to read the vendor's privacy policy yourself before you upload a single observation note. (Counts taken from the directory's public API on November 12, 2026, by filtering on the professional development use case.)
Best for coaching prep from your own evidence: Gemini Notebook
Gemini Notebook is the best fit for turning a week of observation notes, a district policy and a research paper into a coaching conversation, because it is built to answer from the sources you give it and cite them in line, which is the behavior a coach needs when the stakes are a colleague's practice (Google). The directory rating is 4.84 from 90 reviews, more reviews than any other tool in this post.
Here is the workflow. Upload your typed, de-identified notes from three visits to the same classroom, the coaching goal the teacher set, and the relevant pages of your district's AI guidance. Ask: "Across these three visits, where did the lesson's cognitive demand drop, and what did the teacher do just before it dropped?" You get an answer with a citation to your own notes, which you then check against your memory of the room. The doing (sorting three pages of notes) is outsourced. The thinking (what to say to the teacher, and how) stays with you.
For PD, the audio overview feature turns a policy document into something staff can listen to in the car; a coach can hand it out before a session so the meeting starts at the questions rather than the reading.
Watch out for: The free tier has no number you can plan around. Google's limits page says Gemini Notebook "has compute-based usage limits," lists a personal account without a paid plan only as "Standard limits," and puts its paid AI Plus and AI Pro plans at two and four times that, with the quota refreshing every five hours until a weekly limit is reached (Google). The school editions still publish figures.
Pricing: Free with a personal Google account. Through Google Workspace for Education, the Fundamentals and Standard editions allow 100 notebooks per user, 50 sources per notebook, 50 chats a day and 3 audio overviews a day. Education Plus sits one tier up at 200 notebooks, 100 sources, 200 chats and 6 audio overviews; the 500, 300, 500 and 20 figures this article used to give for Education Plus now belong to a higher tier that lists Google AI Pro for Education (Google). Checked November 12, 2026; Google heads the table "Subject to Change."
Safety: For school and work accounts, Google states that uploads, queries and the model's responses "will not be reviewed by human reviewers," even when you give thumbs-up or thumbs-down feedback, "and will not be used to train AI models"; on a personal account, content you send as feedback can be read by Google's review teams and kept for up to three years (Google). The directory record leaves FERPA, hosting and minimum age unknown. Use the district account, not your Gmail, and keep student names out of anything you upload regardless.
Best for AI literacy PD that teachers can deliver: AILitKit
AILitKit is the best fit for a coach who has been told to "do something about AI literacy" this semester, because every activity in a lesson guide comes with coaching, in the vendor's words "so you know exactly what to say" (AILitKit), and the guides include unplugged activities that need no devices. That matters where student-facing AI has been paused. New York City has prohibited student-facing AI tools from its 2-K program through grade 8 for the 2026-27 school year, and K-12 Dive reports that Los Angeles Unified is weighing classroom AI bans alongside new screen-time limits (K-12 Dive, September 2, 2026). Unplugged AI literacy is the version of the mandate those schools can still deliver.
The workflow: upload the teacher's own lesson, any subject, and in about a minute, by the vendor's account, you get four AI literacy activities with coaching notes and support and challenge versions, mapped to 12 AI literacy frameworks "including UNESCO and PISA 2029" (AILitKit). The sample guides on the vendor's site show one or two of the four activities marked unplugged. A coach can run one guide with a department, then leave the teacher with the coaching language rather than a slide deck.
Compared with writing AI literacy sessions from scratch in ChatGPT, the difference is the alignment map and the framework coverage heatmap, which lets a curriculum director see which grades and subjects have had nothing yet.
Watch out for: It launched in 2026 and has one review in the directory, so the evidence base is the product itself, not a track record. It is priced in pounds, which will irritate a US business office.
Pricing: Three free lesson guides to start, then one a month, no card required. Teacher plan £59 a year; Department plan £249 a year for a lead and up to four teachers; School plan £999 a year for up to 50 teachers (AILitKit). No dollar pricing is published. The £7.99 monthly teacher price this article gave in September is unverified: the pricing page showed annual prices only on November 12, 2026.
Safety: Teachers only; the vendor's pricing page says "No pupil accounts," and its site states that the original file is deleted after the guide is made and that the AI provider retains no data (AILitKit). The directory record declares FERPA, COPPA and GDPR, EU hosting, no student data collected, and an 18+ minimum; those are the vendor's declarations, not an audit. Because no student ever logs in, there is no student login for your district to cover in a data agreement, which is the simplest safety case in this post.
Best for showing teachers what monitored student AI looks like: SchoolAI
SchoolAI is the best fit for a coach who needs to demonstrate, in a staff meeting, what a student-facing AI session looks like when the teacher can see every message, because the free trial keeps chat with Dot (the teacher assistant), the Tools and Assistants, and the PD Hub available (SchoolAI). Directory rating 4.92 from 13 reviews.
The job for a coach is not tutoring students. It is calibration, and this is my view rather than the vendor's. Teachers who have only seen ChatGPT assume student AI means an open chatbot; ten minutes inside a Space, watching the Mission Control view populate with student messages, changes the conversation about what a policy should permit. The PD Hub, with certifications and live sessions, is free, so a coach can assign it as pre-work.
Compared with MagicSchool, SchoolAI's edge for this job is the live monitoring view; compared with Khanmigo, it is teacher-built rather than pre-built.
Watch out for: SchoolAI's July 15, 2026 announcement says free trial accounts "now include five Space sessions per year," and that the paid plans "are only available through a school or district purchase" (SchoolAI). Each run of a Space counts as one session. Five sessions is enough to demonstrate, not to run a pilot. You will need a district contract before teachers can use it properly.
Pricing: Free trial with five Space launches a year; Pro and Scale plans are sold to schools and districts only, with no public pricing (SchoolAI). Checked November 12, 2026 and unchanged since September.
Safety: The vendor states it is FERPA and COPPA compliant, SOC 2 Type 2 compliant, COPPA Safe Harbor certified and a member of the Student Data Privacy Consortium (SchoolAI). The directory record shows US hosting, no minimum age, and that it collects student data and shares some with third parties. If your demonstration uses real students, that is a student-data question for your district's agreement, not something a coach signs off alone. Demonstrate with colleagues playing students instead.
Best for modeling feedback inside Google Docs: Brisk Teaching
Brisk Teaching is the best fit for a coach who wants to sit beside a teacher and model AI-assisted feedback on the documents that teacher already has open, because it runs as a Chrome or Edge extension inside Google Workspace and Microsoft 365, and its $0 plan includes 20+ tools (Brisk). Directory rating 4.77 from 35 reviews, and it carries the independent 9ine assessment.
The coaching move is to open a real (de-identified) student essay with the teacher, run Brisk's "glow and grow" feedback, and then spend the session on the part that matters: which of the machine's comments the teacher would keep, which they would cut, and why. In my view teachers learn more from editing AI feedback than from hearing a talk about it. The same session can cover the batch-feedback view, which shows classwide trends a teacher can plan from. For a wider field of feedback tools, see our comparison of AI writing feedback tools for teachers.
Compared with MagicSchool, Brisk's advantage is that the teacher never leaves the Doc; compared with ChatGPT, the feedback is structured for a classroom rather than a chat window.
Watch out for: Brisk's feedback tools are designed for student work. Running them on a lesson plan to give the teacher feedback is a workaround, not a feature, and the output reads that way. The free plan has an upload limit for batch feedback that the plans page does not quantify, and the paid tiers carry no published figure at all, so a coach cannot cost one from public information (no public pricing).
Pricing: Free plan at $0 with 20+ tools on the Standard model; Premium (schools and districts, 35+ tools) and Intelligence (districts) are both listed as custom pricing and require contacting sales (Brisk). Checked November 12, 2026. The plans page no longer supports the word "unlimited" for the free plan, which earlier versions of this article used: the free card does not say it, and the batch feedback row shows an upload limit for the free tier where the paid tiers show "No limit."
Safety: Brisk's privacy notice, effective September 7, 2026, says it operates as a "School Official" under FERPA, uses student data only to provide the service on the school's behalf, and stores data in the United States; it offers a Student Data Privacy Addendum to schools that contract with it. The notice also says Brisk may use de-identified information for "development, research and improvement of our Service," and does not say whether training AI models is excluded (Brisk). The directory record leaves FERPA, COPPA and hosting unknown and needs updating; the 9ine assessment is on file. A district agreement is still the thing that makes student work safe to upload.
Best for recording model lessons and video feedback: ScreenPal
ScreenPal is the best fit for a coach who records model lessons and short video feedback, because the free plan records without a watermark and the education plans cost $3 to $6 a month (ScreenPal). Directory rating 4.67 from 3 reviews.
The workflow that works for coaching is not filming a teacher. It is filming yourself: a ten-minute model of a number talk or a Socratic seminar opener, with a two-question quiz at the end so the teacher has to notice the move you wanted them to notice. Video feedback on a teacher's own plan, screen-recorded with your voice over their document, lands better than a comment thread. Automated captions are a paid-plan feature, so budget the $3 education plan if your district requires captioned video.
Compared with recording on a phone, ScreenPal gives you hosting, quizzes and, on paid plans, automated captions; compared with TeachFX, it analyzes nothing, which for a coaching relationship can be a virtue.
Watch out for: Free recordings are capped at 15 minutes and free hosting at 10 videos, and the plans table excludes automated captions from the free plan. Video of a classroom that contains students is student data; the safety note below applies the moment a child's face or voice is in frame.
Pricing: Free (15-minute recordings, 10 hosted videos, 5 video quizzes, no watermark); Deluxe Edu $3 a month and Max Edu $6 a month, both billed annually; Team Education $3 a user a month, annual, minimum 10 users (ScreenPal). Checked November 12, 2026 and unchanged since September 20.
Safety: The privacy policy, last revised July 22, 2025, states ScreenPal processes student data as a "School Official" under FERPA, that student data is owned and controlled by the school, that it is hosted in the United States, that it deletes or de-identifies student data within 60 days of a school's request, and that it holds iKeepSafe FERPA and COPPA certifications (ScreenPal). The directory record is unknown on all of these and should be updated. Record yourself freely; record students only under your district's consent process.
Best for transcribing adult-only debriefs: Otter.ai
Otter.ai is the best fit for a coach who runs post-observation debriefs and PLC meetings and wants to be present in the conversation rather than taking minutes, because the free plan gives 300 transcription minutes a month with summaries and action items (Otter.ai). It is 9ine-assessed, with a directory rating of 4.20 from 5 reviews.
The value is specific: a debrief transcript lets you check, a week later, what the teacher actually committed to rather than what you remember them committing to. The action-item detection is good enough to seed the next coaching plan. If the meeting is a leadership or department meeting rather than a debrief, our guide to AI note takers for school meetings compares the wider field.
Compared with Gemini Notebook, Otter does the capture and Gemini Notebook does the synthesis; many coaches will use both, exporting the transcript into a notebook.
Watch out for: Otter's privacy policy, effective June 16, 2026, says it trains "our proprietary AI technology on de-identified audio recordings and on transcriptions (which may contain Personal Information)", and it has no FERPA clause at all (Otter.ai). That is why this section is titled adult-only. Do not run Otter in a classroom, and tell the teacher you are recording the debrief before you press the button; recording-consent law differs by state and your district will have a rule.
Pricing: Basic is free with 300 minutes a month, 30 minutes per conversation and 3 lifetime file imports; Pro is $16.99 per user per month, or $8.33 per user per month billed annually; teachers with a .edu address get 20% off Pro, which the pricing page shows as $6.67 a month billed at $79.99 a year (Otter.ai). Checked November 12, 2026: list prices unchanged since September, with lower promotional prices showing on the page that day.
Safety: Minimum age 13; the policy names Amazon Web Services, based in the United States, among the cloud providers it relies on for storage, and permits international transfers under standard contractual clauses rather than designating one location; deletion on request; no FERPA or COPPA statement and, as above, recordings are used for training (Otter.ai). The 9ine assessment covers the platform's security posture, not whether a coach should record a debrief; that judgment, and the consent, is yours.
What About Edthena, TeachFX, MagicSchool, ChatGPT, Khanmigo, Quizlet, Diffit and Perplexity?
Edthena's AI Coach is the closest thing to a purpose-built coaching platform, a conversation-based virtual coach with video analysis, priced from $1,980 a year for a building of up to 20 teachers, with district plans listed from $13,100 a year for five buildings and 200 teachers to $62,100 for 25 buildings and 1,000 teachers (Edthena, checked November 12, 2026); it is not in our directory, its evidence is case studies rather than published research, and its claim of full FERPA and COPPA compliance is the vendor's, so treat it as a serious candidate to pilot, not a recommendation. Edthena's separate Observation Copilot drafts feedback from observation notes, which makes it an observer's tool; it is one of the six in our guide for principals. TeachFX records a lesson and measures teacher talk against student talk; the teacher owns the recording and administrators cannot see it without permission, and pricing is by school or district quote (TeachFX); the same caveats apply. MagicSchool is a teacher tool with a strong free tier and Plus at $12.99 a month or $8.33 a month billed annually (MagicSchool), but its pricing page lists no observation or coaching tools, so it belongs in the teacher's kit, not the coach's. ChatGPT and Perplexity are fine for drafting a PD agenda; neither should ever see an observation note about a named colleague. Khanmigo and Quizlet are student tools and do not belong in this list. Diffit is a text leveler, useful for the teacher you are coaching, not for the coaching.
How Can Instructional Coaches Use AI Without Turning Coaching Into Evaluation?
The answer is to decide, before you open any tool, which of four questions you are asking. Coaching stays coaching when AI touches the doing and never the judgment.
1. Whose data is going in? If the answer includes a student's name, face or voice, stop and find the district agreement. Gemini Notebook with de-identified notes, AILitKit and ScreenPal (filming yourself) never need one. SchoolAI with real students, Brisk with real essays and any classroom recording do.
2. Who can see the output? DC's model policy is explicit that AI-assisted evaluation of teacher performance is prohibited (OSSE). If a transcript, a talk-time ratio or an AI summary of a lesson could reach the person who writes the teacher's evaluation, it is no longer a coaching artifact. Keep coaching outputs in the coach's and teacher's hands only, and say so in writing.
3. Is the tool doing or thinking? Sorting notes, transcribing a debrief, generating four activity options: doing. Deciding what to say to a teacher about their questioning: thinking. Outsource the first; never the second. The early research agrees with the caution: in a 2024 pilot, nine coaches recorded 41 coaching conversations supported by an AI feedback platform, and over half of the coaches said it saved them time, while the same researchers' study of automated feedback without coaching "found many teachers were unable to find time to read the feedback" (Learning Forward, February 2026). Stanford's SCALE Initiative, reviewing the causal studies on AI in K-12 on November 10, 2026, found the matching result on the giving side: "Educators given AI-generated drafts or suggestions provided more and higher-quality feedback without spending more time on the task" (SCALE, November 2026). A draft helps the person giving the feedback. It does not get the feedback read. The coach is what made the feedback land.
4. Can the teacher see what you did? If you used AI to prepare, tell them, and show them. A coach who models transparent AI use teaches more about AI literacy in one debrief than a slide deck does in an hour. Microsoft's 2026 survey of 3,345 students and educators in K-12 and higher education across six countries found that 66% of educators want AI training monthly or quarterly and 53% say they have not received formal AI training (Microsoft, June 2026). Every coaching cycle is a training session in disguise.
AI can sort a coach's notes. It cannot sit across from a teacher and decide what to say next.
How Do You Coach Teachers to Use AI Tools?
Run it as a coaching cycle on one job the teacher already does, and run it more than once. The single workshop is the format most schools reach for and the one with the weakest evidence behind it. A study of 532 teachers in Germany and Denmark, surveyed in January and February 2025 and published in Smart Learning Environments, found that 59.3% had taken no AI professional development at all in the previous two years, 25.2% had taken it once, and 15.5% had taken it more than once. Teachers who had attended once reported higher levels than untrained teachers on the four measures of teaching quality the researchers used (cognitive activation, classroom management, individualization and supportive climate), and the associations were larger for those who had attended more than once (Smart Learning Environments, 2026). The study is a survey, so it shows association rather than cause. Our reading is that repetition adds to the effect, and repetition is what a coach has and a workshop does not.
The American picture is improving and still thin. EdWeek Research Center surveys put the share of teachers with no AI training at almost 60% in October 2024, 50% in fall 2025 and 42% by winter 2026, with 22% having had more than one session and 9% having anything ongoing (Education Week, May 18, 2026). Districts are unevenly placed to close that gap: RAND's district panel found 48% of districts had trained teachers on generative AI by fall 2024, up from 23% the year before, but 67% of low-poverty districts against 39% of high-poverty ones (RAND, April 8, 2025). In plenty of buildings the coach is the training budget.
Where the training lands matters as much as how often it happens. In that same study of 532 teachers, 86.5% of the ones who reported substantive AI use described "efficiency improvements to existing practices" rather than a change in what students do. Jessica Garner of ISTE+ASCD put the consequence to Education Week directly: "We can't stop with efficiencies. We still have a great need for educators knowing good use of AI for the purpose of teaching and learning" (Education Week, May 18, 2026). Our view is that the move past time-saving happens in a classroom with somebody watching, which is a description of coaching rather than of training.
Week one: name the job, not the tool
Ask what took the longest last week. If the answer is feedback on 90 essays, the cycle is about feedback and Brisk Teaching is incidental to it. If the answer is three parents who needed the same email, the tool barely matters. A cycle that starts with a tool has to keep justifying the tool; a cycle that starts with a job stops when the job is easier. Write the job down in the teacher's words and keep it where you both can see it.
Week two: model it live, including where it fails
Do the task in front of them, at their desk, on their material, and narrate what you are doing. Coaches in Digital Promise's January 2026 write-up worked this way: one opened a staff session with Google's Quick Draw to show predictive AI in real time, then introduced a ChatGPT assistant with an Irish accent "to demonstrate AI's capabilities and limitations in a disarming way"; another sat with teachers as a platform refined a weak prompt into a usable one, starting from a real request for an intervention activity for intermediate English language learners (Digital Promise, January 13, 2026). That account is qualitative and drawn from two schools, so treat it as practice worth borrowing rather than as evidence of effect. The part worth borrowing hardest is showing the failure: a first output that is generic, and what you type next.
Week three: co-plan one lesson, then watch it
Plan a single lesson together in which the tool does one defined thing, and agree before the visit what you will both look for. This is the step where coaching separates from training, because you are watching students rather than a product demonstration. Keep your own notes de-identified so they can go into Gemini Notebook afterwards without a district agreement.
Week four: debrief on what the teacher kept
Open the AI output from week three and ask which parts survived contact with the class, which were cut, and why. The answer is the teacher's professional judgment made visible, and it is worth more than any usage statistic. It also gives you the one sentence to carry into the next cycle.
The reason to run the cycle rather than send the resource is the strongest finding in this whole literature. In a working paper summarized by the National Council on Teacher Quality in May 2024, several hundred Utah teachers received automated feedback on their questioning through an app; the treatment group increased its use of focusing questions by about 20%, yet only a little over half regularly opened the emails carrying the insights and a third never opened one at all (NCTQ, May 23, 2024). The feedback worked on the teachers who read it. Getting it read is the job nobody has automated, and it is the same finding Learning Forward reports from the coaching side. If you are rolling this out across more than your own caseload, the person who can make the cycle stick across a faculty is usually the middle leader rather than the principal, which is the case made in our guide for department heads.
What Do You Say to a Teacher Who Does Not Want to Use AI?
Start with the job that is costing them time and never argue about AI in the abstract, because the abstract argument is one you cannot win and do not need. The reluctance is real and it is not marginal: in an EdWeek Research Center survey of 651 teachers in February and March 2026, 13% said they were "not at all" eager to learn about integrating AI into their teaching, and another 24% were only "slightly" eager; an earlier EdWeek survey of educators, reported in the same article, found 47% expecting AI to have a negative effect on teaching within five years against 43% expecting a positive one (Education Week, May 18, 2026). Coaching itself is not the sticking point for most: in December 2024, 57% of teachers told the EdWeek Research Center they would be somewhat or very comfortable with professional development or coaching that incorporated AI, and 20% said very uncomfortable (Education Week, February 14, 2025).
Three things help, and none of them is a persuasion technique.
Name one job and cap the commitment. Offer one task, once, with a defined end. A teacher who tries a tool on one task and decides against it has exercised exactly the professional judgment the policies ask for, and saying so out loud costs you nothing and buys you the next conversation.
Show the limits before the wins. A demonstration that only goes well teaches a teacher to distrust you the first time the tool invents a source. Show it failing on their material, then show what you type to recover, which is the part that transfers.
Put the boundaries in writing. A good deal of the resistance is not about the technology at all but about whether AI output will end up in an evaluation, which is our reading of why the red line in DC's model policy is the most useful sentence a coach can quote this year. DC's model policy answers that in the reader's favor, and your district's may too; if it does not, write the boundary yourself and give the teacher a copy; a version to adapt is in the next section. Where no policy exists at all, which is where the NPR/Ipsos poll puts most teachers, our guide to writing an AI policy for schools is the place to start, and a coach who drafts one has changed more than a single classroom.
What should not be on the list is an adoption target. Counting how many teachers used a tool this month measures compliance, and compliance is the thing the 86.5% efficiency finding above describes.
What Should a Coaching Agreement Say About AI?
It should say, in plain words, what goes into a tool, who sees what comes out, and that none of it reaches an evaluation. Here is a version to copy and adapt. It is a starting template, not legal advice, and your district's policy overrides any line of it.
Our coaching agreement on AI
- I may use AI to sort my own notes, draft options for our next session and transcribe our debriefs. I decide what to say to you; the tool does not.
- No student's name, face or voice goes into any AI tool unless the district has a signed agreement with that vendor.
- My notes about your classroom are de-identified before upload and go only into a district account.
- I record a debrief only after you say yes, each time, and you can ask me to delete the recording.
- Nothing an AI tool produces in our coaching, whether a transcript, a summary or a talk-time count, is shared with anyone who evaluates you.
- You can see any prompt I wrote and any output I used, whenever you ask.
Check lines 2 and 5 against policy first. DC's model policy backs line 5 (OSSE). Where a district has written nothing yet, this agreement is the only written boundary the two of you have, so sign it together and keep a copy each.
What I'd Do This Week
Pick one teacher already in a coaching cycle: say a sixth-grade math teacher whose goal is more student talk during problem-solving. Before the second visit, type up your first-visit notes with no student names, upload them to Gemini Notebook on your district account with the teacher's goal, and ask where in the lesson student talk fell away and what preceded it. Check the answer against your memory. Bring one finding, not five, to the debrief. Record the debrief on Otter only after the teacher agrees, then export the transcript into the same notebook. Measure one thing over the cycle, minutes of student-to-student talk per lesson, counted by you with a stopwatch, and let the tools carry everything else.
The Bottom Line
Instructional coaches need two tools, four questions and one written agreement. Gemini Notebook for prep, AILitKit for the AI literacy PD that policies now require, and SchoolAI, Brisk, ScreenPal and Otter for the single job each does well. Ask whose data goes in, who sees the output, whether the tool is doing or thinking, and whether the teacher can see what you did. Then compare the rest of the directory against the same four questions. The tools will change by January. The questions will not.
What Changed in This Update
Last checked November 12, 2026. Corrected at source: Gemini Notebook's Education Plus limits (200 notebooks, 100 sources, 200 chats and 6 audio overviews a day, where this page said 500, 300, 500 and 20); Brisk Teaching's free plan (no longer described as unlimited); ScreenPal's automated captions (paid plans only); Edthena's district prices; AILitKit's framework count (12) and new Department plan; and the directory counts (19 professional development tools among 138 listings). Three research claims were tightened to match their sources: the Smart Learning Environments study, the Education Week 47% figure and the K-12 Dive line on Los Angeles. Added: Stanford SCALE's November 2026 evidence review, Education Week's 2024 comfort figure and a coaching agreement to copy. Earlier updates: September 20, 2026 (two prices corrected, coaching-cycle sections added).
Dan Fitzpatrick is a Forbes contributor, three-time bestselling author and founder of The AI Educator, and has trained more than 150,000 educators across 30+ countries. Read more from Dan.


