If you are a principal or instructional coach looking for AI that helps with classroom observations, start with StaffDraft, free for 10 drafts a month and £6 a month after that, because it checks your text for a child's name before it writes a word. Then add Otter.ai, free for 300 minutes a month, so the feedback conversation is one you can look up from. MagicSchool drafts the next steps on a tier that stays free. SLT AI handles the appraisal paperwork, TeachScribe captures the observation by voice while you are still in the room, Gemini Notebook reads a term of your own write-ups back to you, and Alder Branch coaches the observer rather than the teacher. None of the seven decides a rating, and the research below is nowhere near supporting one that does.
The Seven at a Glance
| Tool | Best for | Strengths | Limitations | Price / free tier | Privacy considerations |
|---|---|---|---|---|---|
| StaffDraft | Turning rough notes into a written record | Checks for personal identifiers and safeguarding issues before it drafts; states drafts are not retained after processing | Writes staff communication, not framework-aligned feedback | Free tier of 10 drafts a month; £6 per month individual; £5 per user per month for 20 or more users | Directory record: 18+ accounts, states it does not collect student data, UK GDPR stated, hosting UK |
| SLT AI | The appraisal paperwork and the report upward | Built for school leadership writing, with a personnel tool set | Leadership writing only, and written for UK school structures | £10 per month, or £95 per year for one leader; £85 per user per year for 2 to 5 | Directory record: 18+ accounts, states it does not collect student data, UK GDPR stated, hosting UK |
| MagicSchool | Drafting the feedback and the next steps | 80+ staff tools behind one login on a free forever tier; strongest review evidence here at 4.74 from 19 directory reviews | General teacher platform, so nothing in it knows your evaluation rubric | Free forever tier; Plus $8.33 per user per month billed annually, $12.99 monthly | Directory record leaves FERPA, COPPA, minimum age and hosting unstated; custom data privacy agreements sit on the Enterprise tier |
| Otter.ai | The post-observation conversation | Transcribes and summarizes the debrief so you can listen instead of write; carries an independent 9ine security assessment | Transcribes talk, understands no rubric | Free Basic: 300 minutes a month, 30 minutes per transcription; Pro $8.33 per user per month billed annually; 20% off Pro on a .edu address | Directory record leaves student data, FERPA, COPPA, age and hosting unstated |
| TeachScribe | Capturing the observation by voice as it happens | Push-to-talk microphone, maps what you say to the framework; part of the DfE AI Tools for Education program | Built for early years frameworks, not a US evaluation rubric | No public pricing | Directory record: no minimum age, states it collects student data, UK GDPR stated, hosting UK |
| Gemini Notebook | Reading a year of observations for a pattern | Answers only from documents you upload, with a citation back to the source; 4.84 from 90 directory reviews | Finds patterns, writes no feedback | Free tier available | Directory record leaves student data, FERPA, COPPA, age and hosting unknown |
| Alder Branch | Coaching the observer, not the teacher | Aimed at the adult doing the observing, as reflective professional learning | One directory review, so the rating is one person's view | No public pricing | Directory record: 13+ accounts, states it does not collect student data, FERPA and COPPA recorded as yes, hosting United States |
Why This Is Suddenly a Question Worth Asking
The job of writing up an observation is where the vendors arrived. Edthena launched Observation Copilot in late 2025, a free app that takes a principal's unstructured notes and "creates the write-up by aligning the principal's comments to the district instructional framework", then suggests next steps (The Learning Counsel). In January 2026 it partnered with the Danielson Group so the output maps to the Framework for Teaching, and the announcement quotes a principal saying a write-up that took "two to three hours" now takes "about 20 to 30 minutes" (The Learning Counsel, January 23, 2026). That figure is one principal quoted in a vendor announcement, not a measured result, and it is the claim this whole category is being sold on. In June 2026 Education Advanced added AI-assisted feedback to its own Evaluation product, generating draft feedback mapped to whichever rubric the district uses, Danielson, Thoughtful Classroom, a state framework or a custom one, for the principal to edit and approve before any teacher sees it. Its chief executive, Krista Endsley, described the point of it exactly: the tool "removes the drafting burden so they can put their energy into the conversation with the teacher, not the document" (Education Advanced, June 30, 2026).
What has not arrived is any rule about it. GovTech reported in March 2026 on an assistant principal at Pelham Lab High School in the Bronx feeding her observation notes into ChatGPT and getting Danielson-aligned language back, and found that most schools have no formal policy governing the practice at all. In the same piece, Rob Weil, who leads the American Federation of Teachers' National Academy for AI Instruction, put the case against plainly: "It doesn't understand the relationships, it doesn't understand the context ... I just don't believe the tools are good enough right now." He added the sentence that should decide your own practice: "Using it without transparency is a real problem" (GovTech, March 21, 2026).
So this is a window where the tooling is ahead of the policy, and the gap is yours to close. In England the same complaint is being made from the other direction: school leaders told TES that Department for Education guidance has been "quite laissez-faire" and that they want "much clearer rules of the road" (TES, September 9, 2026). Nobody is coming with the rule. Write it yourself, before the first formal observation of the semester, not after.
The Line That Decides Everything Else
An observation produces two kinds of record, and AI belongs near only one of them.
The first is a personnel record about an adult: your notes, the write-up, the rating, the improvement plan. The second is information about the children who happened to be in the room. The US Department of Education states that a photo or video becomes an education record under the Family Educational Rights and Privacy Act, the federal law governing student records, when it is "(1) directly related to a student; and (2) maintained by an educational agency or institution or by a party acting for the agency or institution". A student captured incidentally in the background does not make the file an education record; a recording where "the person or entity taking the photo or video intends to make a specific student the focus" does (US Department of Education, Student Privacy Policy Office).
Read those two conditions together and a practical rule falls out. The moment you record a lesson and upload it to a vendor for analysis, that vendor is a party acting for your school, and whether the file is a protected student record turns on how the camera behaved. A wide shot of a teacher at the board is one thing. Fifty minutes centered on the child you were worried about is another, and it needs the agreement your district signs, not the free tier you clicked through.
That is why six of the seven tools below take text and not video. It is the cheapest possible way to stay on the right side of the line: observe with your eyes, write with your own hands, and let the software work on the words you wrote about an adult.
How I Picked These Seven
The pool was all 138 tools in the aieducator.tools directory, counted today, searched for anything touching observation, lesson feedback, appraisal, walkthroughs or coaching. Selection was on fit for the observer's job rather than the teacher's, and no vendor paid to appear. Prices, free tiers and privacy terms were checked against each vendor's own published page on October 3, 2026. Where a vendor publishes no figure, this post says so instead of estimating.
Here is the finding that shaped the whole list, and it is not flattering to anyone selling into this market. Of those 138 tools, exactly one names classroom observation as the job it does: TeachScribe, and it maps to the Early Years Foundation Stage, the English framework for children up to five, rather than to any American evaluation rubric. Exactly one more, SLT AI, names appraisals among its features, and it is the only record in the directory that names appraisal at all. The purpose-built American option that a principal would actually reach for, Edthena's Observation Copilot, is not in the directory at all, which is why it is named here and not recommended on a record we hold. For a job that every instructional leader in the country does several times a term, the independent shelf is close to bare.
Two smaller numbers from the same count. Of the 27 tools we list under school administration and safety, four publish no price whatsoever, so the category where a superintendent or chief academic officer is most likely to need a budget figure is the one most likely to withhold it. And five of those 27 carry an independent 9ine security assessment, against 36 across the whole directory, which you can check on the independent security assessments page.
One discrepancy worth recording, because it will cost somebody a planning assumption. Our directory records SLT AI as freemium. Its own pricing page lists four paid tiers and no free plan at all (SLT AI pricing). Trust the vendor page, and expect the directory record to be corrected.
What I left out, and why. ClassMind, the research system described below, is a published prototype and not a product you can buy. Any tool whose pitch is that it scores the lesson for you is out on principle, for the reason the AFT gave above. And the big consumer names are dealt with in their own section, because a search assistant will offer them to you whether they fit or not.
Best for Turning Notes Into a Written Record: StaffDraft
StaffDraft is the best fit for getting a defensible written record out of scrappy notes because it runs a check for personal identifiers and safeguarding content before it drafts anything, and it states that "drafts are not retained after processing" (StaffDraft). Safeguarding here means the school's duty-of-care process for a child at risk, and the point is that the tool looks for that material in your text and flags it rather than cheerfully rewriting it.
That ordering matters more than it sounds. Most drafting tools accept whatever you paste and worry about nothing. This one is built on the assumption that a member of staff in a hurry will paste a child's name into a box, which is exactly what happens in the twenty minutes between a lesson and a meeting.
Watch out for: its document types are parent messages, email replies, report comments, references, newsletters and announcements. None of those is a framework-aligned observation write-up, so you are using a good general school drafting tool on a job it was not shaped for.
Pricing: a free tier of 10 drafts a month with the identifier and safeguarding checks included; £6 per month for unlimited drafts individually; £5 per user per month for organizations of 20 or more, which adds policy auto-referencing and an admin dashboard; and one-off credit packs from £2.99 for 20 drafts that do not expire (StaffDraft). Checked October 3, 2026.
Safety: the directory record sets a minimum account age of 18, so no student ever logs in, states that it does not collect student data, and records UK GDPR compliance with UK hosting. FERPA and COPPA are marked not applicable, which is accurate for a UK product and is not the same as clearance for a US district. The vendor offers a formal data protection agreement, which is the document your district counsel will want. Ask for it before the first write-up, not after.
Best for the Appraisal Paperwork: SLT AI
SLT AI is the best fit for the writing that surrounds an observation rather than the observation itself, because it is the only record in the directory that names appraisals among its features, alongside governor reports, school improvement planning and a set for what UK schools call SEND, meaning special educational needs and disabilities. It holds no directory reviews yet, so there is no rating to lean on.
The useful distinction is between the feedback you give a teacher and the account you give of your own quality assurance. The second is a genre: it has an audience that was not in the room, it gets quoted back at you, and it eats an afternoon. That is the part to hand over.
Watch out for: it is written for UK school structures, so an American assistant principal will meet vocabulary that does not match the job. And fluent leadership prose is the easiest thing in education to generate and the easiest to see through. A learning walk summary that reads better than the teaching it describes is a problem you have made worse. For the writing either side of the observation, our guide for assistant principals covers the rest of that load.
Pricing: £10 per month, or £95 per year, for one leader; £85 per user per year for 2 to 5; £75 per user per year for 6 to 10; custom quotes above that. The pricing page lists no free plan, despite our directory record saying freemium (SLT AI pricing). Checked October 3, 2026.
Safety: the directory record sets a minimum account age of 18, states that it does not collect student data, and records UK GDPR compliance with UK hosting. FERPA is marked not applicable, correctly for a UK product, and the record lists Cyber Essentials certification, the UK government-backed baseline scheme. The real risk here is habit rather than platform: leadership writing pulls named examples toward it, and a governing board report does not need a child's name to make its point.
Best for Drafting the Feedback: MagicSchool
MagicSchool is the best fit for producing the feedback itself because it puts more than 80 staff tools behind one login on a tier that is free forever rather than a trial, and it carries the strongest review evidence of anything here at 4.74 from 19 directory reviews. For an observer, the use is not the write-up. It is the three next steps, phrased three different ways, so the conversation on Thursday has somewhere to go.
Watch out for: nothing in it knows your district's rubric. It will produce feedback that sounds like good practice in general and attaches to no indicator in particular, which is precisely the failure mode a teacher will spot in the meeting. The wider staff tool set is covered in our guide for instructional coaches, whose job overlaps with this one at the coaching end. Write the rubric language yourself and let the tool work on the explanation.
Pricing: a free forever tier including the 80+ teacher tools and basic privacy and security features; Plus at $8.33 per user per month billed annually, or $12.99 billed monthly; Enterprise at custom pricing, which is where the "Custom Data Privacy Agreement", single sign-on and district controls sit (MagicSchool pricing). Checked October 3, 2026.
Safety: the directory record leaves FERPA, COPPA, minimum age and hosting unstated, so treat those as open questions rather than failures. The vendor's own pricing page tells you something useful: the custom data privacy agreement is an Enterprise feature, which means the free tier is not where identifiable detail about a named adult or child belongs. Write about the lesson, not the people in it.
Best for the Conversation Afterward: Otter.ai
Otter.ai is the best fit for the post-observation debrief because it transcribes and summarizes the conversation, which lets you look at the teacher instead of your notepad, and it is one of 36 tools in the directory carrying an independent 9ine security assessment. It holds 4.20 from 5 directory reviews.
Our view is that the feedback conversation is the part of this cycle that actually changes teaching, and that spending it with your head down writing is the one clearly wasteful thing an observer does. This is the tool that stops it, and it is the same division of labor Education Advanced names above: energy into the conversation, not the document.
Watch out for: a transcript of a coaching conversation is a record of an employee's professional difficulties, held by their employer. Decide who can see it and how long it is kept before you record the first one, and tell the teacher. Weil's point about transparency applies with more force here than anywhere else on this list.
Pricing: free Basic gives 300 transcription minutes a month, 30 minutes per conversation and 3 lifetime file imports; Pro is $16.99 per user monthly or $8.33 billed annually; Business is $30 monthly or $19.99 annually; students and teachers with a .edu address get 20% off Pro, bringing the annual plan to $6.67 a month (Otter.ai pricing). Checked October 3, 2026.
Safety: the directory record leaves student data, FERPA, COPPA, age and hosting unstated, and the 9ine assessment is the one piece of independent scrutiny attached to it. Because this tool records adults talking about children, the question to settle is retention, not compliance: ask your technology lead how long audio sits in the account and who else in the district can open it.
Best for Capturing the Observation in the Room: TeachScribe
TeachScribe is the best fit for capturing what you see while you are still seeing it, because the practitioner wears a push-to-talk microphone and speaks the observation aloud, and the system transcribes it and maps the learning to the framework rather than making you type between activities (TeachScribe). It is part of the Department for Education's AI Tools for Education program, funded by Innovate UK, which is a more meaningful credential than most of what gets printed on a vendor page.
It is on this list for the mechanism, not the market. Voice capture in the moment is the right answer to the oldest problem in observation, which is that notes written from memory at four o'clock are a reconstruction, and everyone involved knows it.
Watch out for: it maps to the Early Years Foundation Stage and to custom school frameworks, with the vendor saying it also adapts to IB PYP, Montessori, Reggio and SEND targets. A US evaluation rubric is not on that list. If you lead a kindergarten or pre-K setting, this is the closest thing to purpose-built that exists; a high school principal should read this section as a preview of a product category rather than a purchase.
Pricing: no public pricing. The site publishes no figure and directs you to book a demo (TeachScribe). Checked October 3, 2026.
Safety: this is the one tool here that states it collects student data, which follows from what it does: it identifies the child in the observation. The directory record applies no minimum age, states UK GDPR compliance and records UK hosting, and the vendor's site repeats GDPR compliance and UK-hosted data. Whether uploaded audio is used to train models is not stated on the page, so ask in writing, and sign a data processing agreement naming sub-processors and retention periods before a single child is recorded.
Best for Reading a Year of Observations: Gemini Notebook
Gemini Notebook is the best fit for finding the pattern across a term of observations because it answers only from documents you give it and cites the passage each answer came from, and it holds 4.84 from 90 directory reviews, the largest review base of any tool here. It is the tool Google renamed from NotebookLM, and notebooklm.google now redirects to notebook.google.
Here is the job nobody does because it takes a weekend. Load twenty observation write-ups from across the department, and ask which indicator comes up most often and which teacher has had the same next step three times in a row. That is a question about your own practice as an observer, and the answer is usually uncomfortable in a useful way.
Watch out for: it answers from what you put in, so twenty vague write-ups produce a confident summary of vagueness. And a folder containing twenty staff members' observation records is one of the more sensitive collections in the building. Strip names before uploading, or use the account your district contracted rather than a personal one.
Pricing: a free tier is available; paid capacity comes through Google's own subscription plans rather than a product-specific price (Gemini Notebook). Checked October 3, 2026.
Safety: the directory record leaves student data, FERPA, COPPA, age and hosting unknown. Google publishes its own privacy terms for the product, and the one thing to establish with your technology lead is whether your Google Workspace for Education agreement covers this product for staff use, because the consumer version and the contracted version are not the same arrangement.
Best for Coaching the Observer: Alder Branch
Alder Branch is the best fit for developing the person doing the observing, because it is aimed at the adult's own professional learning rather than at producing a document, and its directory record states that it does not collect student data while recording FERPA and COPPA as yes with United States hosting. It holds 5.00 from a single directory review, so read that number as one person's opinion and not a trend.
Why this belongs on a list about paperwork: every write-up above is downstream of the observer's own judgment, and that is the one input no tool here improves. A February 2026 synthesis in The Learning Professional reports that a 2023 study found early versions of ChatGPT were poor instructional coaches, that a pilot of nine coaches running 41 AI-enhanced coaching conversations saw over half say the platform saved them time, and that in automated feedback trials "many teachers were unable to find time to read the feedback" (University of Maryland Education Systems Improvement, February 2026). The same synthesis records an average teacher talk ratio of about 81% of classroom discourse, which is the kind of number an observer can act on and a rating cannot capture.
Watch out for: no public pricing and a login wall, so you cannot evaluate it without talking to someone. One review is thin evidence. Treat this as a pilot with two assistant principals rather than a district purchase.
Pricing: no public pricing. The directory records it as paid and the application sits behind a sign-in (Alder Branch). Checked October 3, 2026.
Safety: the directory record sets a minimum account age of 13, states that it does not collect student data, and records FERPA and COPPA compliance with hosting in the United States. Because the record claims FERPA and COPPA alignment rather than leaving them blank, that claim is the thing to put in front of your district's privacy officer and have confirmed in a signed agreement.
What the Research Actually Supports
The research does not yet support letting AI judge a lesson, and it does support letting AI prepare the human who will.
The strongest recent work on automated observation is ClassMind, described in a September 2025 paper by Ao Qu and colleagues, which analyzes classroom video and generates "temporally precise, best-practice-aligned feedback to support teacher reflection and improvement". The authors report that teachers in field testing found it useful, easy to use and novel, and that the same teachers raised concerns about privacy and about the right balance between automated analysis and human judgment in a coaching relationship. The paper reports qualitative findings, not effect sizes (ClassMind, arXiv, September 2025).
That is the current state of the evidence: promising enough to build on, nowhere near strong enough to put a number on a career. Our view is that the division of labor it implies is the right one for schools to adopt now. Outsource the recording, the drafting and the pattern-finding. Keep the watching, the judgment and the conversation.
The Big Names Search Assistants Keep Suggesting
Ask a chatbot for AI tools for classroom observations and it will offer four things that are not.
ChatGPT is what the assistant principal in the GovTech story actually used, and it is the most capable drafting tool on this page. It is also an account whose boundary between work and personal is invisible from inside the tab. If you use it, use it on an anonymized account of a lesson, and tell the teacher you used it.
Gemini Notebook earns its place above, and it is the one general tool genuinely suited to this work, because it will not answer beyond the documents you gave it.
MagicSchool and Diffit are staff platforms built around the teacher's job rather than the observer's. Diffit re-levels a text for reading age and has nothing to say about an observation; it belongs in the follow-up, when a teacher needs the adapted material your feedback asked for.
Khanmigo and Quizlet are student-facing study tools. Neither has any view of a lesson, a rubric or an adult, and a search assistant offering them here is pattern-matching on the word education.
Four Questions That Decide It
Work through these in order and the shortlist chooses itself.
1. Is the AI touching an adult's record or a child's? If the material is about a teacher, you are in personnel territory and the tools above apply. If a child is identifiable, read the Department of Education's two conditions above and assume you need the agreement your district signs.
2. Does a camera point at anyone? Text-only workflows keep you clear of the hardest part of this. Video brings consent, retention and the education-record test with it. The default should be text, and video should be a decision somebody senior made on purpose.
3. Would you be comfortable telling the teacher? This is Weil's transparency test and it is the fastest one available. If the honest sentence is "I put your lesson into ChatGPT", say it before you do it, in a staff meeting, as practice rather than confession.
4. Who holds the account? A district-contracted account with a data processing agreement, or none. Anything else makes the write-up on a personal subscription an employment record in a private silo, which is a problem waiting for a grievance.
One Observation Cycle, Written Up
Here is how the tools fit a single real task: one formal observation of a second-year high school math teacher, from the pre-conference to the signed record, with no new system going in.
Monday, pre-conference. Fifteen minutes, no software. Agree what you will look for and which indicator it sits under. Write that one line yourself. Everything downstream inherits it, and a tool cannot invent it.
Tuesday, the lesson. Observe with paper. Capture what people did and said, with times, and no judgments yet. If you lead an early years setting, this is where voice capture earns its keep and where TeachScribe fits; in a high school, your own shorthand is still the best instrument available.
Tuesday afternoon, the write-up. Type your notes up yourself, with the teacher's name removed and the children described by what they were doing rather than who they were. Run that through StaffDraft for the prose, and keep its identifier check switched on. Then put the rubric language back in by hand, because that is the part you are accountable for.
Wednesday, the next steps. Use MagicSchool on the anonymized account of the lesson to generate three possible next steps, phrased for a second-year teacher rather than a trainee. Choose one. Discard the other two. One next step delivered beats three recorded.
Thursday, the conversation. Record the debrief with Otter.ai, having told the teacher it is recording and why. Your eyes stay on them for the full twenty minutes. Afterward, the summary becomes the agreed record, which removes the argument about what was said.
Friday, your own practice. Put this write-up with the term's others in Gemini Notebook and ask one question: how many of my last ten next steps were about questioning. If the answer is nine, that is a fact about you and not about your teachers.
What you measure. One thing: whether the single next step shows up in that teacher's room three weeks later. If it does not, the observation was paperwork.
The Bottom Line
Start with StaffDraft if your problem is the hours the write-up eats, and with Otter.ai if your problem is that you spend the feedback conversation writing instead of listening. Add MagicSchool for the next steps, free, and keep it on anonymized text. Hold the rubric language, the judgment and the rating in your own hands, because the research is not close to supporting anything else and the union representative in the room will know it. If you lead early years, TeachScribe is the one genuinely purpose-built option on the shelf; everyone else is waiting for this category to be built, and it is worth watching the school administration and safety category to see who gets there first.
One tool is not the point. The write-up was never the job. The job is the twenty minutes on Thursday when a teacher decides whether to try something different, and no software has ever been in that room.
Dan Fitzpatrick is The AI Educator. More on Dan.
