You finish a session, scribble half a page of notes, and tell yourself you’ll review them before the next one. You don’t. Three sessions later you can’t remember whether the recruiter story was new or the fourth time it came up. The notes exist; the insight doesn’t.
This skill turns a single Custom GPT into a coaching session notes app that does two jobs at once: it cleans your raw notes into a tidy, neutral summary with clear action items, and it keeps a running map of the themes your client keeps circling back to. So the moment a fear or pattern shows up for the third time, you can see it and name it with them. By the end of this page you’ll have the skill built, a full worked example, and the reasoning that makes it reliable.
When to use this
- Right after a session, when your notes are messy and your memory is still fresh.
- When you carry 10+ clients and can’t hold each person’s arc in your head.
- When you sense a client keeps returning to the same theme but want proof before you reflect it back.
- When you’re preparing for the next session and want last time’s summary in 30 seconds.
- When you’re handing notes to a supervisor or doing your own CPD review and need them structured.
The skill
This is a Custom GPT instruction set, not a one-off prompt. Paste the whole block into the Instructions field of a ChatGPT Custom GPT (or a Claude Project / Gemini Gem):
ROLE
You are a meticulous coaching session note-taker and theme tracker. You work for a professional coach. Your job is to turn one set of raw, messy session notes into a clean, structured summary AND to maintain a running map of the recurring themes a client returns to across sessions. You are precise, neutral, and you never invent details the coach did not record.
INPUTS YOU NEED
The coach will paste:
- Client name or alias: {{CLIENT_NAME}}
- Session number: {{SESSION_NUMBER}}
- Session date: {{SESSION_DATE}}
- What this person is coached on: {{COACHING_FOCUS}}
- Raw notes from the session: {{RAW_NOTES}}
- Themes carried over from prior sessions: {{PRIOR_THEMES}}
Before you write anything, scan the raw notes. If something essential is missing or ambiguous (for example: no clear action items, an unclear pronoun, or a possible safety concern that needs a human decision), ask up to 3 clarifying questions first. If the notes are complete enough to work with, skip the questions and proceed.
PROCESS
1. Read the raw notes and separate what the CLIENT said or did from what the COACH observed or interpreted. Keep them distinct.
2. Identify 1-3 themes present in this session. Compare them to {{PRIOR_THEMES}}. Mark each theme as NEW, RECURRING, or RESOLVING.
3. Pull out concrete commitments the client made (action items) and any commitments the coach made.
4. Note one thing that seemed to shift or land well, and one open thread to revisit next session.
5. Update the running theme map: merge this session's themes with the prior ones, count how many sessions each theme has appeared in, and flag any theme appearing in 3+ sessions as a PATTERN worth naming with the client.
OUTPUT FORMAT
Return exactly these sections, in this order, using Markdown headings:
## Session Summary — {{CLIENT_NAME}}, {{SESSION_NUMBER}} ({{SESSION_DATE}})
A 3-5 sentence neutral recap. No advice, just what happened.
## What the Client Brought
- Bullet list of what the client said, felt, or did (their words/experience).
## Coach Observations
- Bullet list of what you (the coach) noticed or interpreted. Label these as observations, not facts.
## Themes This Session
- Theme name — [NEW | RECURRING | RESOLVING] — one line on how it showed up.
## Action Items
- Client: the commitments the client made, with any stated deadline.
- Coach: the commitments the coach made.
## Landed / Open Thread
- Landed: the one thing that seemed to shift.
- Revisit next time: the one open thread.
## Running Theme Map (paste this into next session's PRIOR_THEMES)
- Theme name — appeared in N session(s) — [PATTERN if 3+].
RULES
- Never invent quotes, results, diagnoses, or details the coach did not write down. If you infer something, label it clearly as an inference.
- Stay neutral. You summarize and organize; you do not coach the client or give the coach advice unless asked.
- Do not use clinical or diagnostic language (e.g. 'depression', 'trauma') unless the coach used it first.
- Keep client information self-contained in this conversation. Do not reference other clients.
- If the notes suggest risk of harm to the client or others, say so plainly in a short note at the top and recommend the coach follow their own safeguarding protocol. Do not attempt to assess or treat it.
- The Running Theme Map must be copy-paste ready so the coach can feed it into the next session.
How to set it up
The variables aren’t replaced in the instructions field — they’re the inputs you paste fresh after each session. Here’s each one and a one-time build:
| Input | What to paste | Example |
|---|---|---|
{{CLIENT_NAME}} |
A name or privacy-safe alias | Maya R. |
{{SESSION_NUMBER}} |
Which session this is | Session 6 |
{{SESSION_DATE}} |
The date | 2026-05-27 |
{{COACHING_FOCUS}} |
What you coach them on | career transition and confidence after a layoff |
{{RAW_NOTES}} |
Your messy notes, exactly as written | (your shorthand) |
{{PRIOR_THEMES}} |
The Running Theme Map from last session (blank for session 1) | (paste from last output) |
Build it once:
- In ChatGPT, go to My GPTs → Create a GPT (Claude: new Project; Gemini: new Gem).
- Name it Session Note-Taker & Theme Tracker and add a short description.
- In Configure → Instructions, paste the entire skill block above.
- Turn off web browsing and external actions you don’t need, to keep notes private to the chat.
- Save it Private and pin it to your sidebar.
- After each session, open it, paste your six inputs, and save the Running Theme Map for next time.
See it in action (full example)
Here is the complete input a career coach pastes after a session. Nothing is hidden:
Client name or alias: Maya R.
Session number: Session 6
Session date: 2026-05-27
What this person is coached on: career transition and confidence after a layoff
Raw notes from the session: talked about the recruiter call on tues - froze again, went quiet, said she 'sounded like an idiot'. mentioned dad's expectations AGAIN - he keeps asking when she'll get a 'real job'. did the box-breathing thing before a mock pitch with me, said it actually helped, voice steadier. wants to apply to 3 roles by friday but kept hedging - 'if I have time'. lit up when she talked about the portfolio project, totally different energy. didn't open the salary doc we agreed last time.
Themes carried over from prior sessions: fear of judgment from family (3 sessions); perfectionism stalling applications (4 sessions); voice/freezing in live conversations (2 sessions)
And here is the complete output the GPT returns:
Session Summary — Maya R., Session 6 (2026-05-27)
Maya described freezing during a recruiter call on Tuesday and judged her own performance harshly afterward. Her father’s expectations came up again as a source of pressure. A box-breathing exercise before a mock pitch noticeably steadied her, and she showed markedly higher energy when discussing her portfolio project. She set an intention to apply to three roles by Friday but qualified it, and she had not opened the salary document agreed in the prior session.
What the Client Brought
- Froze and went quiet on Tuesday’s recruiter call; said she “sounded like an idiot.”
- Raised her father’s repeated questions about getting a “real job.”
- Reported that box-breathing before the mock pitch helped; her voice felt steadier.
- Expressed visible energy and enthusiasm about her portfolio project.
- Said she wants to apply to 3 roles by Friday, but added “if I have time.”
- Did not open the salary document agreed in the last session.
Coach Observations
- Observation: the self-criticism after the recruiter call looks tied to the same judgment fear that surfaces around family, not just the call itself (inference).
- Observation: her energy shift on the portfolio topic suggests it may be an underused source of confidence to build on.
- Observation: hedging language (“if I have time”) tends to precede stalled applications for her (inference based on prior pattern).
Themes This Session
- Fear of judgment (family + self) — RECURRING — showed up in both the recruiter post-mortem and her father’s comments.
- Perfectionism stalling action — RECURRING — appeared as hedging on the 3-role goal and the unopened salary doc.
- Freezing in live conversations — RECURRING, RESOLVING — box-breathing produced a steadier voice, the first reported improvement.
- Portfolio as an energy source — NEW — distinct lift in energy and confidence.
Action Items
- Client: apply to 3 roles by Friday 2026-05-30; open the salary document before next session; use box-breathing before any live call.
- Coach: send the box-breathing one-pager; bring a short portfolio-to-pitch exercise to next session.
Landed / Open Thread
- Landed: box-breathing visibly steadied her voice before the mock pitch — first concrete win on the freezing theme.
- Revisit next time: the gap between stated goals and hedging language, and why the salary doc stays unopened.
Running Theme Map (paste this into next session’s PRIOR_THEMES)
- Fear of judgment from family — appeared in 4 session(s) — PATTERN
- Self-judgment after live conversations — appeared in 2 session(s)
- Perfectionism stalling applications — appeared in 5 session(s) — PATTERN
- Voice/freezing in live conversations — appeared in 3 session(s) — PATTERN, RESOLVING
- Portfolio as an energy source — appeared in 1 session(s)
Notice what the coach now has: a clean record, separated facts from interpretation, deadlines pulled out, and three themes flagged as PATTERNs — including “perfectionism stalling applications” at 5 sessions, which is the conversation worth having next time.
Why this works
Four LLM principles are doing the work here. Understanding them makes you better at building any Custom GPT:
- Role priming sets the standard. “You are a meticulous coaching note-taker… precise, neutral, you never invent details” tells the model not just what to do but how to behave. “Summarize my notes” gives you a vague paragraph; a defined role with stated values gives you discipline. The model writes in the register of the role you assign.
- Specificity in, specificity out. The skill demands six structured inputs and a fixed output with named sections. Because the output format is pinned down, the result is consistent every session — which is the whole point of a tracker. Loose prompts give beautiful-but-different answers each time; a coaching notes app needs the same shape every time so you can compare sessions and paste themes forward.
- Constraints are quality control. The RULES section is the most important part. “Never invent quotes or results,” “separate client words from coach observations,” “don’t use clinical language unless the coach did,” and the safeguarding note each remove a specific, dangerous failure mode. With client notes, a confident hallucination isn’t a typo — it’s a false record. Telling the model what NOT to do is as load-bearing as telling it what to do.
- Clarifying questions beat guessing. The “ask up to 3 clarifying questions first” instruction lets the model surface a missing action item or an ambiguous pronoun instead of inventing one. That single line is the biggest difference between notes you trust and notes you have to re-check.
Do this now
- Create a new Custom GPT (or Claude Project / Gemini Gem) and paste the skill block into the Instructions field.
- Turn off web access, save it Private, and pin it.
- After your next session, paste your six inputs (leave PRIOR_THEMES blank if it’s session 1).
- Copy the Running Theme Map it returns into a note for that client. Next session, it becomes your PRIOR_THEMES — and the tracker starts compounding.
Pro tips
- Use aliases, not full names. “Maya R.” or a code works fine for the model and keeps your records tidy and discreet. Confirm your own data and consent obligations before pasting any client information into an AI tool.
- Keep the Running Theme Map in one place per client. A single running note (one per client) that you update each session is what makes the PATTERN flags meaningful. The map is only as good as your habit of pasting it back in.
- Trust the PATTERN flags as conversation starters, not conclusions. When something hits 3+ sessions, that’s your cue to reflect it back to the client — “I notice this has come up a few times” — not a diagnosis.
- Feed it your real shorthand. Don’t tidy your notes before pasting. The messier the raw input, the more the structure it adds is worth. Cleaning notes yourself defeats the purpose.
0 comments
No comments yet.