ChatGPT daily productivity workflows: 8 templates for notes / email / docs / research

ChatGPT is more than a chat tool - 8 production workflows wire it into daily work: meeting summary, email drafts, long doc summary, research synthesis, customer comms, PPT outline, personal study, brainstorming. Custom Instructions + Projects + Custom GPTs make these workflows reusable one-click.

0. Base config (do once)

Custom Instructions

Settings → Custom Instructions. Two paragraphs:

What would you like ChatGPT to know about you?
- I'm XXX company XXX position
- My main work: product review / customer comms / document writing
- Tech I use: XXX
- Domains I care about: XXX

How would you like ChatGPT to respond?
- Default business but friendly tone
- Outline first before answering
- Tables / lists over paragraphs
- Avoid long pleasantries
- Don't invent facts, when uncertain say 'I'm not sure, suggest verifying'
- When mixing Chinese / English prefer Chinese

Applies to all conversations automatically - no re-prompting needed.

Projects (by scenario)

Build independent Projects for high-frequency scenarios:

Each Project gets independent Instructions (role + task), uploads relevant materials, multi-turn keeps context.

Custom GPTs (solidify workflows)

Solidify high-frequency workflows into Custom GPTs:

• meeting-notes: recording transcript -> 5-section minutes
• email-drafter: bullets -> business email
• pdf-summarizer: long PDF -> structured summary
• research-synthesizer: multiple papers -> comparison table

Each GPT with Actions (if calling API) or Code Interpreter (if data processing).

1. Meeting notes workflow

Scenario: 30-min product review meeting, need minutes right after.

Flow:

1. Use Whisper / Tongyi to transcribe meeting recording (with speaker labels)
2. Paste transcript (5k-10k chars) into ChatGPT
3. Use this prompt:

PROMPT:
Turn the following product review transcript into 5-section minutes:

1. [Decisions] Clear decisions made, each labeled with owner + deadline
2. [Open questions] Raised but unresolved, where the disagreement is
3. [Risks] Risks, dependencies, external blockers discussed
4. [Next steps] Concrete actionable tasks, labeled with owner + due date
5. [Next meeting agenda] Leftover / follow-up

Requirements:
- Strictly based on transcript, no additions not discussed
- Detail over 200 chars use bracket timestamp reference (e.g. 14:23)
- Output in English

Transcript: [paste]

Tips:

• 30-min meeting -> 1 page minutes (~800 words)
• Multi-speaker scenario must have speaker labels (otherwise AI can't tell who said what)
• Key decisions 'we decided to...' AI usually catches; implicit consensus 'we probably should...' needs prompt emphasis

2. Email draft workflow

Scenario: Reply to customer business email about order 12345 delay + solution.

Flow:

1. Paste bullets into ChatGPT:

Bullets:
- Customer: XX company procurement manager Zhang San
- Order: 12345, originally promised 8/15 delivery, now expected 8/22
- Delay reason: supplier material material problem
- Our compensation: 5% discount on original order + priority shipping
- Tone: sincere, not servile
- Length: under 200 words

2. ChatGPT outputs draft
3. Multi-turn refinement:
   - "Be more direct, remove pleasantries"
   - "Change 'we deeply apologize' to 'we own responsibility for the delay'"
   - "Add a line confirming timeline"
4. Final human copy

Tips:

• Bullets must include: audience + subject + key facts + desired tone + length
• Custom Instructions writes 'business email' style preference, global
• Multi-turn refinement is normal - first draft often needs 2-3 rounds

3. Long document summary workflow

Scenario: 100-page PDF product whitepaper, need 5-paragraph summary for team.

Flow:

1. Upload PDF to ChatGPT (not paste text - preserves structure)
2. Prompt:

PROMPT:
Generate 5-paragraph summary (English) from the uploaded whitepaper, each 100 words:

1. [Product positioning] What is it, what problem, target customer
2. [Core capabilities] 3-5 core functions + specific pain solved
3. [Differentiation] Unique advantages over competitors (per document, no additions)
4. [Business model] Pricing / deployment / after-sales / ecosystem

Requirements:
- Strictly based on uploaded doc, no external knowledge
- Specific numbers (performance / customer count) must cite page number
- Doc not mentioned content say 'doc not mentioned'

Tips:

• Upload PDF better than paste text (PDF preserves structure, AI understands better)
• ChatGPT handles 100-page PDF generally OK, over 200 pages split
• Grounding key: 'strictly based on uploaded doc' + 'doc not mentioned content say so'

4. Research synthesis workflow

Scenario: 5 papers on LLM inference optimization, need comparison table.

Flow:

1. Upload 5 PDFs to Projects (Projects can upload multiple at once)
2. prompt:

PROMPT:
Generate comparison table (English) from these 5 papers:

Comparison dimensions:
- Core method (one sentence)
- Performance gain (vs baseline)
- Memory overhead (vs baseline)
- Applicable model size
- Main limitations

Output markdown table + 50-word summary (which is best for our scenario)

Papers: [upload]

Tips:

• Projects good for multi-doc batch processing (preserves context to multi-turn)
• Comparison dimensions must be explicit (AI won't guess which you need)
• Limit to doc's conclusions, don't 'AI synthesizes speculation'

5. Customer communication workflow

Scenario: Customer in group chat asks 'how stable is your API?', need quick reply.

Flow:

1. In Projects ("work: customer comms") paste customer message + history
2. prompt:

PROMPT:
Customer asked: "How stable is your API?"

History:
- Customer is XX company, Pro subscription
- Last month had one 5xx incident (resolved)
- Customer now considering upgrading to Business

Generate reply (English, business but direct):
- Give real stability metrics (99.9% SLA, monthly incidents < 1)
- Proactively acknowledge last incident
- Recommend Business's higher SLA + dedicated support
- Within 150 words

Tips:

• Projects prepares 'company intro + FAQ answers + case library', AI references don't rely on memory
• Custom Instructions writes 'no exaggeration, no over-promising'
• Specific numbers / policy / pricing must be human-verified (AI occasionally invents)

6. PPT outline workflow

Scenario: 30-min presentation on '2026 AI coding trends'.

Flow:

1. prompt:

PROMPT:
I need to give 30-min presentation on '2026 AI coding trends'. Generate 12-slide outline (English):

Each slide:
- Title (5-10 words)
- 3-5 bullet points (each 1 sentence)
- Visual suggestion (chart / screenshot / diagram)

Style: technical depth + business perspective, target audience is engineering directors

Outline:
1. Intro (30 sec)
2. Current state
3. Trend 1: ...
4. Trend 2: ...
... (12 slides)

Requirements:
- Trends must have specific company / product cases
- Data points must be verifiable (don't invent)

Tips:

• Outline phase use ChatGPT, PPT generation use Gamma / Slidev / KeyNote
• 30-min talk ~ 12 slides, 2-3 min per slide
• AI 'visual suggestions' saves 50% visual design time

7. Personal learning workflow

Scenario: Studying CS224N deep learning course, need Q&A.

Flow:

1. Upload textbook PDF to Projects ("learning: CS224N")
2. Any chapter has questions, ask directly:

Me: "Why does Transformer's self-attention use dot-product instead of cosine?"
ChatGPT: ...

Tips:

• Projects lets AI remember the whole book, no need to paste text each time
• More specific prompt, more accurate AI: 'explain in a way an 8-year-old understands' vs 'why scaled dot-product attention'
• AI gives 'real-world analogy' easier to grasp than direct equations

8. Brainstorming workflow

Scenario: Product direction stuck, need 50 ideas.

Flow:

prompt:

PROMPT:
I'm building 'AI coding assistant'. Give me 50 innovative feature ideas, grouped by category:

- Performance optimization (5)
- User experience (5)
- Business model (5)
- Developer ecosystem (5)
- AI model layer (5)
- Integration / compatibility (5)
- Security / compliance (5)
- Multi-modal (5)
- Data / evaluation (5)
- Long-term vision (5)

Each idea 1 sentence, include 'why might work'.

Tips:

• Don't limit 'feasibility' - brainstorm phase is wild growth
• Ask AI for 'anti ideas': which ideas are the opposite
• Put 50 ideas back into Projects, AI continues deepening 3-5 in subsequent conversations

Practical tips

Custom Instructions vs Projects

Custom Instructions is 'global preference' - 'use business tone' / 'list outline first' / 'avoid verbosity'. Applies to all conversations.

Projects is 'scenario workspace' - in 'product review' project upload docs + write system prompt + multi-turn.

Best practice: Custom Instructions for cross-scenario preferences, Projects for scenario-specific prompt + materials. Two layers stacked.

Workflow efficiency key isn't how strong ChatGPT is, but how clear your prompt + how complete your context

Change 'summarize this recording' to 'turn this 30-minute product review recording into 5 sections: decisions / open questions / risks / next steps / next agenda' - output quality doubles immediately.

Iterate prompts

After each workflow use, save useful micro-adjustments ('be more direct' / 'remove pleasantries' / 'outline first then detail') into Custom Instructions or Projects. Default next round.

FAQ

1. Custom Instructions vs Projects - which?

Custom Instructions is 'global preference' - 'use business tone' / 'list outline first' / 'avoid verbosity'. All conversations. Projects is 'scenario workspace' - in 'product review' upload docs + system prompt + multi-turn. Best practice: stack both layers.

2. Meeting notes quality?

Three tools: (1) upload recording transcript (Whisper); (2) 5-section prompt - participants / decisions / open questions / risks / next steps; (3) Custom Instruction - record key discussions in time order, highlight decisions, label owners. 30-min meeting -> 1-page key notes.

3. Email tone tuning?

Custom Instructions specify: (1) role; (2) audience; (3) tone; (4) length. Then paste bullets and let ChatGPT draft. Multi-turn: 'be more direct / remove pleasantries / change to active voice'.

4. Long doc hallucination avoidance?

Key tricks: (1) upload PDF (not paste text); (2) Projects combines multiple related docs + specify type and time; (3) prompt requires 'answer based on uploaded documents, no external knowledge'. If doc exceeds context (rare for PDFs under 100 pages), split.

5. Custom GPTs build?

Three steps in GPT Builder: (1) system prompt (role + output format + boundaries); (2) knowledge files (FAQ / templates / case library); (3) Actions (external API) or Code Interpreter (data processing). Each GPT after setup searchable in GPT Store or shared privately in team workspace.

Next steps

• New to ChatGPT? Read ChatGPT Complete Guide (2026): From Beginner to Expert.
• Want subscription comparison? Read ChatGPT Subscription Plans Compared (2026): Free vs Plus vs Pro vs Business vs Enterprise.
• Want Projects detail? Read ChatGPT Projects Complete Guide: File RAG, Custom Instructions, and Team Collaboration.