Professional learning · Coaches, Leaders · 50 minutes
Two Drafts of Feedback
Coaches score a human-written and an AI-drafted feedback note on the same lesson transcript, then set the ground rules for when AI belongs in observation, if it belongs at all.
Generative AI can turn a lesson transcript into a tidy feedback note in seconds, and some coaching platforms now offer to do it automatically. Coaches in this session read a short fictional transcript, score two unlabeled feedback drafts against a quality checklist, and guess which one a chatbot wrote. The reveal sets up the harder conversation: consent, student privacy, transparency with the teacher, and the line between an AI draft and a coach's judgment.
In this packet
- Handout A: Lesson Transcript and Two Feedback Drafts (reading)
- Handout B: Feedback Quality Checklist (checklist)
- Handout C: AI in Observation Sort (card sort)
- Facilitator key (last page)
TCEA ELE indicators
- C1.2 I can provide constructive feedback to improve teaching strategies.
- AI2.3 I am aware of the importance of transparency and accountability in AI systems.
- C1.1 I know how to mentor teachers in effective instructional practices.
- AI2.1 I know how to apply ethical principles when using or developing AI for education.
Full facilitator guide: mglearn.github.io/eles/activities/two-drafts-of-feedback.html
Handout A for Two Drafts of Feedback
Lesson Transcript and Two Feedback Drafts
Fictional. Mr. Delgado teaches 5th grade math at Bluebonnet Elementary. His goal for this observation: "More students explain their thinking out loud, not just give answers." Highlight each time a student explains their thinking.
1Transcript excerpt (minutes 6–11). Mr. Delgado: "Which is bigger, three-fourths or five-eighths? Hands up." (Four hands.) "Marcus?" Marcus: "Three-fourths." Mr. Delgado: "Good. Why?" (Two seconds.) "Because three-fourths is the same as six-eighths, right? And six is more than five. Great."
2Mr. Delgado: "Next one. Two-thirds or three-fifths? Ana, you look like you've got it." Ana: "Two-thirds, because I drew them both on the number line and two-thirds was farther from zero. It's a little farther." Mr. Delgado: "Nice, Ana used a number line. Everyone, write that down."
3Mr. Delgado: "Last one before the worksheet. Four-sixths or two-thirds? Anyone?" (One second.) "They're equal! Four-sixths simplifies to two-thirds. Okay, worksheets out, pages 12 and 13." A student at the back table: "Wait, how did you..." (Students begin the worksheet; the question is not picked up.)
4Draft X. "Mr. Delgado, thank you for having me. Your goal was more students explaining their thinking. Ana's number-line explanation was a real example of that, and you named her strategy for the class. I also noticed that after you asked Marcus 'Why?' about two seconds passed before you gave the reasoning yourself, and on the last problem the question at the back table went unanswered. Three things to try: wait at least five seconds after 'Why?', have partners explain to each other before you call on anyone, and ask Ana to teach her number line to a small group."
5Draft Y. "Mr. Delgado demonstrated strong questioning techniques throughout the lesson. The turn-and-talk at the start of the lesson was highly effective in engaging all learners, and students were clearly comfortable sharing their reasoning. One area for growth is wait time; research suggests extending wait time improves the quality of student responses. Consider implementing think-pair-share, exit tickets, and differentiated grouping to further increase student discourse. Ana struggled initially but showed growth. Overall, this was an excellent lesson aligned to grade-level standards."
6Questions to consider. Which sentences can you prove from the transcript? Which sentences could have been written about almost any lesson? Which draft respects Mr. Delgado's goal? Is there anything in either draft about a student that shouldn't be in a document others might see?
Handout B for Two Drafts of Feedback
Feedback Quality Checklist
Score one draft per checklist. For every "Yes," write the transcript line that proves it in the margin.
| Yes | Partly | No | |
|---|---|---|---|
| Names the teacher's own goal and stays focused on it. | |||
| Every strength it names actually happened in the transcript. | |||
| Quotes or describes a specific moment, not a general impression. | |||
| Notices what students did, not only what the teacher did. | |||
| Offers one next step the teacher could try in the very next lesson. | |||
| Avoids generic advice that could fit any lesson. | |||
| Describes students in a way that would be fine if the teacher or a family read it. | |||
| Sounds like it came from someone who was in the room. | |||
| Would help this teacher change something on Monday. | |||
| Our guess: written by a person / drafted by AI. What made us think so? |
Notes
Handout C for Two Drafts of Feedback
AI in Observation Sort
Cut apart. Place each strip under a heading. For every strip in the middle, name the guardrail on a sticky note.
Reasonable with consent and disclosure
Only with added guardrails
Don't do it
Handout C for Two Drafts of Feedback
AI in Observation Sort: cards to sort
Cut apart the strips. Sort each one onto the mat and be ready to explain why.
Using an approved chatbot to reword your own finished feedback so it is shorter, then checking every sentence before sending.
Asking an approved chatbot to suggest five questioning strategies for a fractions lesson, with no lesson data included.
Pasting a de-identified transcript (no teacher, student, or school names) into a district-approved tool to draft first-pass feedback.
Recording a lesson on video and running it through an AI tool that counts teacher talk time versus student talk time.
Letting a coaching platform's AI automatically send feedback to the teacher after an observation, with no coach review.
Pasting raw observation notes that name students and describe their behavior into a free consumer chatbot.
Using AI-drafted feedback in a teacher's formal evaluation file without telling the teacher AI was involved.
Asking an approved chatbot to turn your handwritten observation notes (initials only, no student names) into a clean summary you then edit.
Telling teachers at the start of the year which AI tools you might use, what goes into them, and that they can opt out.
Using an AI tool to compare one teacher's lesson transcript with another teacher's and rank them.
Facilitator only for Two Drafts of Feedback
Answer key and notes
Handout C: AI in Observation Sort
| Item | Sort | Why |
|---|---|---|
| Using an approved chatbot to reword your own finished feedback so it is shorter, then checking every sentence before sending. | Reasonable with consent and disclosure | The coach wrote and verified the content; tell the teacher you used a tool to edit. |
| Asking an approved chatbot to suggest five questioning strategies for a fractions lesson, with no lesson data included. | Reasonable with consent and disclosure | No personal data; the coach still judges which strategy fits. |
| Pasting a de-identified transcript (no teacher, student, or school names) into a district-approved tool to draft first-pass feedback. | Only with added guardrails | Needs the teacher's agreement, de-identification checked by a person, and the coach verifying every claim against the transcript. |
| Recording a lesson on video and running it through an AI tool that counts teacher talk time versus student talk time. | Only with added guardrails | Needs teacher consent, district policy on recording students, a clear retention and deletion plan, and a check of how accurate the counts are. |
| Letting a coaching platform's AI automatically send feedback to the teacher after an observation, with no coach review. | Don't do it | No human accountability; invented or generic claims reach the teacher unchecked. |
| Pasting raw observation notes that name students and describe their behavior into a free consumer chatbot. | Don't do it | Student information shared with an unapproved service; a likely privacy violation. |
| Using AI-drafted feedback in a teacher's formal evaluation file without telling the teacher AI was involved. | Don't do it | Fails transparency; evaluation decisions need human authorship the teacher can question. |
| Asking an approved chatbot to turn your handwritten observation notes (initials only, no student names) into a clean summary you then edit. | Only with added guardrails | Check that nothing identifying remains, verify the summary against your notes, and disclose the use. |
| Telling teachers at the start of the year which AI tools you might use, what goes into them, and that they can opt out. | Reasonable with consent and disclosure | This is the disclosure itself, and it builds trust. |
| Using an AI tool to compare one teacher's lesson transcript with another teacher's and rank them. | Don't do it | Turns coaching into ranking; undermines trust and uses colleagues' data for a purpose they did not agree to. |