Use Claude to turn 1:1 notes into talking points
Paste permitted notes, extract source-linked topics and questions, then refine a short agenda and checked follow-up.
Original editorial illustration.
An editorial Claude workflow using fictional notes. Talking points, agenda and follow-up are illustrative, not captured Claude responses. No live model test or measured saving is claimed.
The useful part
What you'll learn
A practical Claude workflow, from notes to a checked draft.
Extract talking points
Turn numbered notes into topics with a source for each one.
Refine the questions
Ask Claude to remove judgements the notes do not support.
Check the result
Verify a short agenda and a follow-up against what was agreed.
What you get: source-linked talking points, a 30-minute agenda and a follow-up draft you can check.
Why try this workflow?
A clearer draft. Fewer details to lose.
For youClaude may reduce the manual work of regrouping scattered notes. You still choose the input and check the draft.
For the teamSource IDs can make a changed deadline or an unagreed action easier to spot before the draft is shared.
For the businessA checked follow-up may make ownership and missing commitments clearer. It does not complete the work or guarantee a better meeting.
Use the same fictional notes to prepare one draft yourself and one with Claude. Record the time taken including corrections, unsupported claims found and whether each agreed action has the correct owner/date. Compare the checked results before deciding whether the workflow helps you. No saving has been measured here.
Claude can turn a few permitted work notes into talking points, draft questions and a short agenda. The useful part is making each suggestion traceable to something you supplied, so you can spot a question it has invented rather enthusiastically.
This walkthrough uses a normal text chat. You paste the notes yourself; Claude is not connected to your calendar, employee records or previous meetings. An optional Project lets you reuse the instructions next time.
What you'll make
source-linked talking points, a 30-minute agenda and a checked follow-up draft.
What you'll need
an approved Claude account and a small set of permitted notes. A new chat is enough; no paid upgrade or connector is needed for this fictional exercise. Normal account limits apply.
Start with fictional notes. For real work, use only information your organisation permits in its approved AI account. Removing a name does not necessarily anonymise a distinctive situation. Keep health, pay, complaints, disciplinary matters and other sensitive employee information out. Do not ask Claude to judge a person’s performance or make employment decisions.
1. Paste a small, numbered source
The transformation you are asking for
Open claude.ai, sign in to your approved account and start a new chat. Copy the prompt in step 2 with these fictional practice notes underneath it:
N1: Meeting length: 30 minutes.
N2: Colleague A is preparing a handover guide.
N3: I promised a clear priority order last Tuesday and have not sent it.
N4: The guide is waiting for a test account from another team.
N5: A asked to lead the next product demo.
N6: Leave at least 10 minutes for whatever A wants to discuss.The IDs are a simple checking aid. A topic based on N4 should concern the test account, rather than a guess about why somebody has not worked harder. Anthropic’s getting-started guide describes starting a conversation by typing a prompt.
2. Ask Claude to extract talking points
Send this prompt with the notes:
Use only the numbered notes below to prepare talking points for a 1:1.
Make a table with: topic, supporting note IDs, one neutral question
about the work or support needed, and any missing context.
Include commitments I owe the colleague. Keep facts separate from
questions. Do not infer personality, motivation, wellbeing or
performance. Do not recommend an employment decision. If information
is missing, mark it as missing rather than filling it in.
[Paste N1–N6.]A useful result would look like this illustrative extract, written here to show the structure, not captured from Claude:
| Topic | Source | Draft question | Missing context |
|---|---|---|---|
| Priority order I have not sent | N2, N3 | Which part of the guide needs the priority order first? | A date for me to provide it |
| Test account blocking the guide | N2, N4 | What help is needed to obtain the test account? | Who can supply it and when |
| Request to lead the demo | N5 | What would you need to prepare for the demo? | The demo date and preparation needed |
Where one talking point comes from
The supplied notes
N2 A is preparing a handover guide.
N4 The guide is waiting for a test account.
The illustrative draft
Topic: test account blocking the guide.
Question: what help is needed to obtain it?
Still missing: who can supply it, and when.
Check every source ID. N5 supports a request to lead a demo; it does not prove readiness, approval or a development need. If Claude adds one of those, remove it and ask for the supporting note.
3. Turn the checked topics into a short agenda
In the same chat, ask for a smaller output:
Using the checked talking points, draft a one-page 30-minute agenda.
Put my outstanding commitment first. Reserve at least 10 minutes
for the colleague's own topics. Keep the test-account and demo topics.
Show time ranges that total 30 minutes. Keep supporting note IDs.
Do not add new facts, promises or dates.An illustrative agenda could be:
| Time | Topic | Source |
|---|---|---|
| 0–5 minutes | Priority order I owe A; ask what is needed first | N2, N3 |
| 5–15 minutes | A’s own topics | N6 |
| 15–25 minutes | Test account and demo preparation questions | N4, N5 |
| 25–30 minutes | Note any actions actually agreed | N1; proposed agenda structure |
Check the 30-minute limit
5 + 10 + 10 + 5 = 30 minutes
The time allocation is a proposed structure, not a record of agreement. Check that it totals 30 minutes and protects the ten-minute space. If it does not, reply: “The agenda exceeds 30 minutes. Shorten it while preserving N3 and the ten minutes required by N6.” Check the revised arithmetic yourself.
4. Ask Claude to repair unsupported questions
If the draft contains “Why haven’t you taken ownership?”, it has added a verdict the notes do not contain. Give it a specific revision task:
Audit the questions against N1–N6. For each question, show its source
IDs and identify any unsupported judgement or assumption.
Rewrite affected questions as neutral questions about the work or
support needed. Remove questions with no source. Keep missing context
visible. Do not infer a person's motives or performance.For N4, “What help is needed to obtain the test account?” is grounded in the blocker. “Why are you avoiding the task?” is not. Check the revision against the notes, rather than accepting Claude’s own audit as proof that it is right.
5. Reuse the instructions in a Project, if useful
This step is optional. A saved prompt and a new chat work too. For a repeated task, open Projects → + New Project and name it “1:1 preparation”. On Team or Enterprise, choose private visibility when prompted. Leave it unshared and check who has access.
Choose Set project instructions, paste the brief below and save it:
Help me turn permitted, numbered work notes into 1:1 talking points.
Use only the notes supplied for this task. Cite their IDs for facts
and suggested questions. Separate facts, suggestions and missing context.
Do not infer personality, motivation, wellbeing or performance.
Do not make employment decisions or add commitments, owners or dates.
Keep the output short. Ask for missing information instead of guessing.Anthropic documents Projects and their instructions, including up to five projects on free accounts. Availability and workplace policy can vary. Instructions apply to project chats; they do not fetch your calendar or approve employee data.
Next time, start a chat inside the Project and explicitly paste the permitted current notes with the step-2 prompt. Do not rely on earlier chats or memory as the source. Keep reusable rules in the instructions, rather than building an employee-history file in project knowledge. If Projects are unavailable, paste the brief at the start of a normal new chat.
6. Extract a follow-up from actual agreed notes
After the meeting, manually supply only permitted action notes. These fictional follow-up notes are separate from the preparation notes; do not let Claude turn the earlier questions into commitments:
F1: I agreed to send the priority order by Thursday 8 October 2026.
F2: I agreed to ask the other team about the test account. No date agreed.
F3: A will draft a demo outline. No date agreed.In the chat, send:
Use only F1–F3 below to draft a short follow-up table: action,
explicitly agreed owner, agreed date and supporting note IDs.
Keep missing dates as "to agree". Do not import commitments from
the preparation agenda or add a deadline. Put unresolved details
in a separate section. Do not send the draft anywhere.
[Paste F1–F3.]A correct illustrative row is: “Ask the other team about the test account | Me | to agree | F2”. It must not acquire the 8 October deadline from F1. Check all three rows against the notes; correct any invented owner or date before copying the draft into your chosen document or message. Nothing is sent or added to a task system by this workflow.
Keep this: Give Claude a small source, ask for a specific transformation, then check each suggestion against that source.
Accounts and privacy
An approved account and suitable data selection both matter. Anthropic says commercial-product inputs and outputs are not used for training by default; feedback and opt-in choices can change handling. Consumer policies differ. Training defaults do not establish permission to share employee information or zero retention. Follow your organisation’s policy and check its actual settings.
Sources and status
Instructions and privacy claims checked on 4 October 2026 against Get started with Claude, create and manage Projects and Anthropic’s commercial model-training policy.
All people and notes are fictional. Tables are illustrative editorial examples. We have not run this workflow in Claude, compared models or measured time saved. Read our editorial approach for how we label evidence.