Custom Observation System: From Raw Notes to Framework-Aligned Feedback in 10 Minutes

What This Builds

A custom observation write-up system that goes beyond the basic Level 3 template: a Custom GPT (or a saved prompt system) pre-loaded with your specific teaching framework, your school's evaluation rubric, teacher-specific context, and your writing preferences. When you open it and paste your observation notes, you get a complete write-up that references your framework's specific components, is calibrated to that teacher's development stage, and sounds like you wrote it rather than a generic observation form.

The Level 3 guide gets you from 60 minutes to 15 minutes per write-up. This gets you to 10 minutes and eliminates most of the editing you still do.

Prerequisites

• ChatGPT Plus subscription (required for Custom GPTs)

  • Cost: {{tool:ChatGPT.price}}/month
  • Sign up at {{tool:ChatGPT.url}}
• Comfortable using ChatGPT for observation write-ups (Level 3)
• A completed observation cycle or two using the Level 3 system
• Your school's teaching framework (Danielson, Marzano, or your own rubric)

The Concept

A Custom GPT is like a new assistant who has memorized your observation rubric, knows how you give feedback, and remembers the context of every teacher on your staff. You give it a name, upload your framework, write instructions for how it should behave, and then every conversation starts from that trained baseline.

The key advancement over the Level 3 system: instead of pasting a prompt template each time, your Custom GPT already knows the template. You open the GPT, paste your notes, and it produces the write-up. And because it can reference teacher-specific context you've provided, the feedback can build on prior observations rather than treating every write-up as a blank slate.

Build It Step by Step

Part 1: Create your Custom GPT

1. Log into ChatGPT at {{tool:ChatGPT.url}} with your Plus account
2. Click your account icon (top right) → My GPTs → Create a GPT
3. You'll see two tabs: Create (conversational builder) and Configure (direct editing). Click Configure. It's faster to write your own instructions than to use the conversation builder.
4. Name your GPT: [School Name] Observation Writer
5. Write a short description: Converts raw classroom observation notes into framework-aligned teacher feedback for [School Name] staff.

Part 2: Write the system instructions

In the Instructions box, paste this template (customize the bracketed sections):

You are an observation write-up assistant for charter school administrators at [School Name]. Your role is to convert raw classroom observation notes into formal, framework-aligned teacher feedback.

Teaching Framework: [Danielson Framework / Marzano Framework / describe your school-specific rubric]

Output format for every observation write-up:
1. Context line: Grade, subject, lesson phase, approximate timing
2. Domain-by-domain observations: For each domain you observed evidence for, cite specific evidence from the notes. Reference specific component numbers (e.g., Danielson 3b, 3c) where applicable.
3. Strengths: 2-3 specific strengths with evidence. Begin with what the teacher did well.
4. Areas for Growth: 1-2 growth areas, evidence-based, framed as "continue developing" — never punitive
5. Next Steps: 2 specific, actionable steps the teacher can take before the next observation
6. Word count: 300-400 words (unless I specify otherwise)

Tone: Coaching-oriented, not evaluative. Assume the teacher is a capable professional working to grow. Write in second person to the teacher ("Your students...").

Evidence discipline: Use only what's in my notes. If my notes are sparse for a domain, say "Insufficient evidence observed to rate [domain]" rather than inventing evidence.

What NOT to do:
- Do not invent classroom events or student behaviors not mentioned in my notes
- Do not use vague praise ("Great job with engagement") — every claim needs evidence
- Do not flag minor issues if they don't rise to the level of a meaningful growth area
- Do not include both a formal rating AND a developmental next step for the same domain — pick the appropriate depth for the observation type

When I paste observation notes, immediately produce the write-up without asking clarifying questions first (unless notes are under 5 bullet points). I can ask for revisions after.

Part 3: Upload your framework document

In the Knowledge section of your GPT configuration, click Upload files. Upload:

• Your teaching framework document (Danielson handbook PDF, Marzano manual, or your rubric document)
• Any sample observation write-ups you've written that represent your ideal output (2 to 3 strong examples)

This lets the GPT reference specific framework language and match your writing style.

Part 4: Add teacher-specific context (the advanced piece)

This is what transforms a general observation tool into a system that knows your teachers. Create a plain-text document called teacher-context.txt with entries like this:

TEACHER: [First Name, Last Name]
Grade/Subject: [e.g., 3rd Grade Math]
Development Stage: [New teacher — Year 1 / Developing — Year 2-3 / Proficient — Veteran / On improvement plan]
Current Focus Area (from last observation): [e.g., "Improving wait time and cold call consistency in whole-group instruction"]
Background: [Brief note: "Former KIPP teacher, strong on data, needs work on student discourse"]
Notes format preference: [e.g., "Receives feedback well; can handle direct growth language"]

Upload this document to your GPT's Knowledge section. When you start an observation write-up, add one line: "Teacher: [First Name]". The GPT will calibrate the feedback depth and next steps to that teacher's developmental context.

Update this document at the start of each semester.

Part 5: Test it

Open your GPT and paste real observation notes from a recent visit. Check:

• Are framework components cited correctly (correct numbers, correct domain)?
• Is the evidence accurate to your actual notes?
• Are the next steps specific to what you observed, not generic?
• Does the tone match how you actually give feedback?

Refine your instructions if any of these are off. The most common fix: add a sample write-up to the Knowledge section ("Here's an example of exactly what I want the output to look like").

Real Example: Two Observations, 20 Minutes Total

Setup: Your Custom GPT is configured with Danielson Framework, your teacher-context.txt with 8 staff entries, and 2 example write-ups you've written.

Observation 1: Ms. Chen (Year 2 teacher, working on student discourse):

You type:

Teacher: Ms. Chen
Notes: Objective posted and referenced at start. Do Now was clear, students working independently. Transition to whole-group took 4 minutes — procedural but slow. During instruction, Ms. Chen called on 3 students who raised hands, no cold call. Students answering questions were 2-3 high responders. Class discussion lasted 8 min, low-responders not engaged. Exit ticket: 4 problems, 7 of 22 students completed all 4.

What the GPT produces: A 350-word write-up referencing Danielson 3b (Using Questioning and Discussion Techniques) with specific evidence of the participation imbalance, a strength on the Do Now and objective clarity (Danielson 1a/3c), and a next step directly tied to her current development focus: a specific cold-call protocol to try in tomorrow's lesson.

Observation 2: Mr. Torres (new teacher, Year 1, broader feedback needed):

You type:

Teacher: Mr. Torres
Notes: Room setup — desks in rows, can't see all students from front. Launch was 12 minutes of direct instruction, no student interaction. Students off task starting around min 8. IP: worksheet, 20 min. Circulated twice. No checks for understanding during IP. Students working quietly but some not engaging with work.

What the GPT produces: A write-up calibrated for a Year 1 teacher: strengths on classroom order and student compliance, growth area focused on the most critical pattern (no checks for understanding), and next steps that are doable for a new teacher (not overwhelming). Tone is supportive.

Total time: 8 minutes to review and make minor edits to two completed write-ups.

What to Do When It Breaks

• GPT references the wrong framework components: Add more examples to your Knowledge section that correctly use component numbers. Or specify "Always cite Danielson by its exact component number and name."
• GPT invents classroom events: This is the biggest failure mode. Add to your instructions: "Never describe anything not explicitly in my notes. If a note says 'good wait time,' describe wait time; do not elaborate on what the students were doing during that wait."
• Output is too generic for a specific teacher: Your teacher-context.txt may not be detailed enough. Add more about what that teacher is working on and what "good" looks like for them right now.
• GPT can't find the teacher in context doc: Use the exact same name format each time. If your doc says "Maria Chen," type "Teacher: Maria Chen" rather than just "Chen."
• Write-ups all sound the same: Add 2 to 3 varied examples to the Knowledge section (a strong write-up for a proficient teacher, a developmental one for a new teacher, one for a formal evaluation). Variety in examples produces variety in output.

Variations

• Simpler version: Skip the Custom GPT setup and just save a very detailed prompt template in a note app (Notes app, Notion, Google Keep). Paste it into a new ChatGPT conversation before each observation batch. Less setup, slightly more friction per use.
• Extended version: Build separate Custom GPTs for different observation types: one for informal walk-throughs (short, sticky-note format), one for formal evaluations (full rubric with ratings), one for post-conference notes.

What to Do Next

• This week: Build the GPT with your framework document and test it with 3 observations from this week's batch.
• This month: Add your teacher context document and test the teacher-specific calibration.
• Advanced: At the end of each marking period, update your teacher-context.txt based on what you observed. The GPT becomes a running record of where each teacher is in their development, and the write-ups get sharper as it accumulates context.

Advanced guide for charter school administrators. All observation documentation should comply with your school's evaluation policy and collective bargaining agreements where applicable. AI-generated write-ups require review before sharing with teachers.