The most expensive part of a deep work session is usually the first fifteen minutes. Not because the work is hardest then, but because you spend that time reconstructing where you left off — scanning notes, rereading the last paragraph, trying to remember what the actual next step was. Research on attention residue by Sophie Leroy at the University of Washington suggests that incompletely processed prior tasks continue competing for cognitive resources even after you have switched to new work. Starting a session without clearing that residue means beginning at partial cognitive capacity.
The prompts below treat AI not as a thinking partner for the session itself — during focused work, the AI should be closed — but as a preparation and recovery tool for the moments immediately before and immediately after. The five prompts map to five specific friction points: context reconstruction, open-loop clearance, exit-point definition, mid-session recovery, and end-of-session handoff. Together they form a runway that gets you into focused work faster and exit it cleaner, so the next session starts with less friction.
Use them in sequence for the first few sessions until the pattern is familiar. Most experienced practitioners end up using prompts 1, 3, and 5 as the core set; prompts 2 and 4 are situational.
Prompt 1: Context Loading (Gate 1)
Use when: Starting any session where more than 24 hours have passed since your last work on the project, or at the beginning of any session where you cannot immediately recall the next specific action.
Use this at the start of every session to load your working memory before you touch the document.
I'm starting a [X]-minute deep work session on [project/task].
Here is my current context:
[paste: last draft, notes, task description, or relevant emails]
Give me:
1. Two sentences on where I am in this work right now.
2. The single most important thing to do in this session.
3. Three things I need to hold in mind to do it well.
4. One risk or complication to watch for.
Keep the output under 150 words.Why it works: Complex tasks require loading a mental model before you can make progress on them. This prompt produces a condensed brief that collapses 15–20 minutes of manual re-reading into 2 minutes of reading.
What good output looks like: Four distinct items in under 150 words — the “under 150 words” constraint in the prompt is doing real work here. A long, thorough response means the AI has not forced itself to prioritize. What you want is a brief that takes ninety seconds to read and leaves you feeling oriented. If the output is longer than two paragraphs, the prompt failed; tell the AI “trim to under 150 words, keep only what matters for the next action.”
Prompt 2: Interruption Triage (Gate 2)
Use when: You have two or more open items that you are actively monitoring and that might surface as intrusive thoughts during the session. If you have nothing pending, skip this prompt.
Use this when you have several pending items that might surface as distractions during the session.
I'm going offline for [X] minutes.
These open loops might pull my attention during the session:
[list them—be specific]
For each one, tell me:
- A 30-second action I can take right now to create closure, OR
- A short message I can send to hand it off, OR
- Whether it can genuinely wait until after.Why it works: Attention residue is generated by unfinished tasks as well as completed ones that linger. Triage creates cognitive closure on items you are actively monitoring, which frees attention for the session.
What good output looks like: A specific action for each item — a 30-second calendar block, a two-sentence Slack message, or a clear “this can wait until 4pm.” If the AI tells you an item “can wait” without explaining why, push back: “why can it wait, and what is the latest I need to address it?” A prompt that produces only “can wait” answers is being too permissive about what constitutes real closure.
Prompt 3: Exit-Point Definition (Gate 3)
Use when: Before every session, without exception. This is the most consistently high-value prompt in the set.
Use this before every session to define what success looks like.
My session goal is: [describe the task or project area]
I have [X] minutes.
Define a specific, observable exit point for this session.
Requirements:
- Ambitious but achievable in the time I have
- Observable: I can say "done" or "not done" definitively
- Narrower than my full project goal
Not "make progress"—name the actual deliverable.Why it works: Clear goals are a prerequisite for flow states (Csikszentmihalyi). A defined exit point also prevents sessions from drifting or extending indefinitely, which protects cognitive resources for subsequent days.
What good output looks like: One named deliverable — a document, a decision, a completed section, a working prototype — that you can point to at the end of the session. “Make significant progress on the analysis” is not a good output; “complete the first three rows of the competitive matrix with sources” is. If the AI returns something that includes the word “progress” or “continue,” ask it to try again with a concrete noun.
Prompt 4: Mid-Session Recovery
Use when: An interruption has actually occurred during a session — a colleague appears at your desk, an urgent call comes in, an alarm fires. Do not use proactively.
Use this only if you are interrupted during a session and need to re-orient quickly.
I was interrupted [X] minutes into a session on [task].
Before the interruption, I was working on [specific thing].
I have [remaining time] left.
In three sentences: re-orient me to where I was
and give me the single best thing to work on
for the remaining time.Why it works: After an interruption, re-orientation takes 5–15 minutes unaided. This prompt compresses it. The key is using it quickly rather than letting the interruption expand into a longer break.
What good output looks like: Three sentences that put you back at the specific point you left off — not a general summary of the project, but a precise “you were here, doing this, the next thing is this.” If the output starts with a project overview rather than your immediate prior position, it has prioritized context over re-entry. Tell the AI: “Focus on where I was, not what the project is.”
Prompt 5: Session Handoff (Post-Session)
Use when: At the end of every session, before you close your notes or switch to another task. This takes two minutes and pays for itself in the next session’s ramp time.
Use this at the end of every session to create your Gate 1 input for next time.
My session on [project] is ending.
I completed: [brief description of what you produced]
I stopped at: [where exactly you left off]
The next step is: [best guess at the most important next action]
Write a three-sentence handoff note I can paste
into my next context-loading prompt.
Write it as if briefing a version of me who has
completely forgotten this session.Why it works: Cold starts—sessions where you have no handoff from the previous session—cost the most ramp time. A handoff note makes Gate 1 faster and more accurate next time.
What good output looks like: A three-sentence note written in second person — “You were working on X, you had just finished Y, the next thing to do is Z” — that reads like a briefing from your past self to your future self. If the output is a bulleted project status summary, ask again: “Write it as a handoff for someone who just sat down, not as a project update.”
The Sequence in Use
Before a session: run Prompts 1, 2 (if needed), and 3. Close AI. Work.
If interrupted: run Prompt 4 once. Close AI. Return to work.
At session end: run Prompt 5. Save the output somewhere you can find it.
Total AI time per session: 5–8 minutes. Total uninterrupted session time: whatever you have blocked.
Common Pitfalls
Pitfall: Keeping the AI open during the session. These prompts are runway tools, not co-working tools. The moment you open a chat window to ask a question in the middle of a session, you have introduced a context switch and an attention residue source. Fix: close the AI tool before you begin work, and do not open it again until the session ends.
Pitfall: Skipping the exit-point prompt because you already know the task. Knowing the task and naming a specific, observable deliverable are different things. The exit-point prompt forces precision that most people think they have but do not. Fix: even if you are confident you know what you are doing, run Prompt 3. The output will be more specific than your mental version — or it will reveal that you do not have a clear deliverable yet.
Pitfall: Using Prompt 1 with vague context. Pasting in a task name without the underlying material — the last draft, the relevant notes, the conversation thread — produces an orientation brief based on nothing. The AI will generate plausible-sounding output, but it will be reconstructed from generalities rather than your actual work. Fix: paste the real artifact, even if it is rough or partial. Bad context produces confident-sounding but useless output.
Pitfall: Running Prompt 4 as an extended break. Mid-session recovery should take three minutes, not twenty. If you find yourself having a long conversation with the AI after an interruption, you have turned the recovery prompt into avoidance. Fix: use Prompt 4, read the three-sentence re-orientation, close the window, and return to work within five minutes.
Start Right Now
Copy Prompt 3 and run it before your next session. Define what done looks like before you start. That single prompt will show you the value of the others.
The full deep work methodology behind this prompt sequence — including how to schedule sessions, how to protect them from calendar encroachment, and what Newport’s four philosophies imply for your specific role — is in the Complete Guide to Deep Work with AI Assistance.
Frequently Asked Questions
-
When should I use these prompts?
Prompts 1–3 are used immediately before a session (the runway). Prompt 4 is for mid-session recovery if you get interrupted. Prompt 5 is used at the end of every session to prepare the next Gate 1 input. Together they cover the full deep work session lifecycle.
-
Do I need to use all five prompts every session?
Not necessarily. Prompts 1, 3, and 5 are the core set—context loading, exit point, and handoff. Prompt 2 (interruption triage) is most useful when you have several open loops. Prompt 4 (recovery) is for use only when an interruption actually occurs.
Frequently Asked Questions
-
When should I use these prompts?
Prompts 1–3 are used immediately before a session (the runway). Prompt 4 is for mid-session recovery if you get interrupted. Prompt 5 is used at the end of every session to prepare the next Gate 1 input. Together they cover the full deep work session lifecycle.
-
Do I need to use all five prompts every session?
Not necessarily. Prompts 1, 3, and 5 are the core set—context loading, exit point, and handoff. Prompt 2 (interruption triage) is most useful when you have several open loops. Prompt 4 (recovery) is for use only when an interruption actually occurs.