A single prompt is a tool. A stack of prompts is a system. The difference between an amateur AI user and a pro is mostly that the pro has built systems for the work that happens every week.
The pro does not write a new prompt every morning to triage the inbox. They run an inbox triage stack. The pro does not improvise a brief before every meeting. They run a meeting prep stack. The pro does not stare at a blank document hoping the model gives them something usable on the first try. They run a writing production stack that drafts, critiques, and revises in three steps.
This post gives you the seven stacks that cover roughly 80 percent of what a knowledge worker does in a week. Each stack is three to five chained prompts, ready to copy and use. Together they reclaim about 10 hours every week. The math is simple. Each stack saves between 30 minutes and 2 hours per use. Most people run at least one of them every day.
What Is a Prompt Stack?
A prompt stack is a sequence of prompts where the output of each prompt feeds the next. The first prompt extracts. The second structures. The third drafts. The fourth critiques. The fifth revises. The output of the last prompt is the deliverable. Each prompt in the stack is small and focused on one job. The sequence is what produces the value.
A single prompt asks the model to do everything at once and hope it works. A stack asks the model to do one thing at a time, well. The output quality is materially higher because each step is constrained and each step gets the full attention of the model.
Stacks also compound across uses. Once you have a stack that produces good output, you can run it on similar inputs forever. The first time you build a stack it takes 20 minutes. Every time after, it runs in 2 minutes. The ratio is excellent and gets better with reuse.
Stack 1: The Inbox Triage Stack
Problem it solves. Two hours of email turns into 30 minutes of decisions.
Prompts in sequence:
Bulk classifier. "Act as my chief of staff. I am pasting subject lines and first sentences of unread emails. Sort them into four buckets: respond today (needs my voice), respond this week, delegate (to whom), and archive. Emails: [paste]."
One sentence drafts. "For every email in the respond today bucket, draft a one sentence response I can edit. Keep responses under 25 words each. Pair each draft with the original subject line. Output as a table."
Decline drafts for the deferrable. "For every email in the respond this week bucket that is a request I want to decline, draft a polite decline under 60 words. Acknowledge the request specifically. Offer one alternative if appropriate."
Time saved per use. 60 to 90 minutes for a typical executive inbox.
Pro tip. Run the stack twice a day, not after every new email. The classifier works better on batches than on single messages.
Stack 2: The Meeting Prep Stack
Problem it solves. Walking into meetings without a brief, then losing the meeting to people who did the prep.
Prompts in sequence:
Pre meeting brief. "Build a one page brief for the following meeting. Cover: stated purpose, my goal, what each attendee wants, three questions worth asking, two decisions that might come up, and the artifact I want to leave with. Meeting context: [paste calendar invite, agenda, prior notes]."
Counterargument check. "For each of my three goals in this meeting, identify the two strongest counterarguments the other side might raise. For each counterargument, give me one sentence I can use to respond without being defensive."
Talking points. "Convert my goals and the counterargument responses into a single talking points sheet I can glance at during the meeting. Use one line per point. Order by likely sequence in the conversation."
Time saved per use. 30 to 45 minutes per meeting that previously got no prep.
Pro tip. Run the stack 24 hours before the meeting, not 5 minutes before. The talking points need time to settle in your head, not just on the page.
Stack 3: The Document Analysis Stack
Problem it solves. Long documents that everyone says they read but actually skimmed.
Prompts in sequence:
Structured extraction. "Read the following document carefully. Output: one paragraph summary, the five key claims with the page or section each appears in, the three assumptions the document depends on, and the two questions a careful reader would ask. Document: [paste]."
Stress test. "For each of the three assumptions above, give me: the evidence in favor, the evidence against, and what would have to be true for the assumption to be wrong. Be specific."
Decision implications. "Based on the analysis above, what are the three decisions this document is implicitly asking me to make? For each, what is the recommended action and what are the trade offs?"
Time saved per use. 45 to 75 minutes per dense document.
Pro tip. Run this stack before reading the document yourself, not after. The structured output frames your own reading and makes the document much easier to engage with critically.
Stack 4: The Writing Production Stack
Problem it solves. The blank page, plus the second draft that should be better than the first but somehow is not.
Prompts in sequence:
Brief. "I need to write [type of document] for [audience] with the goal of [outcome]. The key points I want to make are [list]. The voice should be [voice profile prompt or features]. Draft the first version. Use the role anchoring, format specification, and constraint stacking from my standard pattern."
Self critique. "Critique the draft above as a professional editor. For each section, list one specific weakness with evidence. Identify any claims that are vague or unsupported. Note any sentences that sound like AI defaults."
Revise. "Produce a second draft that addresses each weakness identified above. Keep the structure. Tighten the language. Remove anything flagged as an AI default. Keep the same voice profile constraints."
Time saved per use. 45 minutes to 2 hours per piece, depending on length.
Pro tip. Stop after the second draft, not the fourth. Models hit diminishing returns after two passes. Past that you should be editing as a human, not asking the model for more revisions.
Stack 5: The Research Stack
Problem it solves. Three hours of tabs that produces nothing usable.
Prompts in sequence:
Research plan. "I need to understand [topic] to make the following decision: [decision]. Output a 5 step research plan: what specific questions I should answer, what sources are most likely to have each answer, and what 'good enough' looks like for each question. Do not produce the research itself yet, just the plan."
Source extraction. Repeat for each source (web articles, documents, reports): "Based on the research plan, extract from the following source: the relevant facts, the relevant claims, the date or recency signal, the credibility signal, and any conflicting view with another source I have already added. Source: [paste]."
Synthesis brief. "Synthesize all the extracted facts and claims into a one page brief that answers the original questions in the research plan. Note where sources agreed, where they disagreed, and where the evidence is thin."
Time saved per use. 60 to 120 minutes per research session.
Pro tip. Keep the extracted facts from step 2 in a separate document. The synthesis from step 3 changes each time you add a new source. The facts compound.
Stack 6: The Decision Stack
Problem it solves. Decisions that drift for a week because no one structured the thinking.
Prompts in sequence:
Frame the decision. "I am facing the following decision: [describe]. Help me frame it. Output: the actual decision in one sentence, the options on the table, the criteria that should drive the choice, the weight of each criterion, and the timeline within which the decision must be made."
Score the options. "Score each option against each criterion (1 to 5). For each score, give me one sentence of reasoning. Calculate a weighted total per option. Output as a table."
Counterfactual stress test. "For the leading option, answer: what would have to be true for this to be the wrong choice? What is the evidence for and against each of those conditions? What is the cheapest way to test the riskiest assumption before committing?"
Decision document. "Write a one paragraph decision memo. State the decision, the reasoning, the trade off being accepted, and the conditions under which the decision should be revisited. Tone: factual, no hedging."
Time saved per use. 60 to 90 minutes per non trivial decision. More when the decision was about to drift.
Pro tip. Save the decision memo. Three months later, re read it. Compare to what actually happened. The pattern that emerges over a dozen memos is more valuable than any single one of them.
Stack 7: The Weekly Review Stack
Problem it solves. Friday afternoons that should produce strategic clarity and instead produce more inbox.
Prompts in sequence:
Week in review. "Run my weekly review. Here is what I committed to last Monday: [paste]. Here is what actually got done: [paste]. Output: honest one paragraph assessment, three things that worked, two things that did not, one pattern to change, three commitments for next week. Plain language. No hedging."
Strategic question. "Based on the assessment above and the following context [paste broader context], generate the one strategic question I should hold in my head next week. The question should be hard to answer in one sentence, point at a real decision, and force me to choose between two genuinely different paths."
Calendar pre commitment. "Based on the three commitments from step 1 and the strategic question from step 2, recommend three specific calendar blocks I should add to next week. For each block: day, length, goal, and the artifact that should exist when the block ends."
Time saved per use. 30 to 45 minutes. The real value is not the time saved. It is the strategic week that would otherwise have not happened.
Pro tip. Do this on Friday before you stop, not Monday before you start. Friday self is more honest about what happened. Monday self is too eager to look forward.
How to Build Your Own Stacks
The seven stacks above cover most of the recurring work in a knowledge worker's week. There is no reason to stop there. Any recurring task in your week is a candidate for its own stack.
The pattern is consistent. Identify a task you do at least once a week. Break it into 3 to 5 small steps the model can handle in sequence. Write the prompt for each step. Test the chain on a real instance. Iterate the prompts until each step produces output the next step can use. Save the final stack.
The first stack you build takes 30 to 45 minutes. The fifth takes 10. By the time you have built a dozen, you are looking at every recurring task and asking why it does not have a stack yet. That is the pro mindset.
The compounding here is the point. Each new stack adds another hour or two of weekly leverage. Six months in, the stacks together replace the equivalent of a full day of work every week. That is the difference between the amateur and the pro.
Frequently Asked Questions
Do prompt stacks only work in ChatGPT?
No. The same stacks work in Claude and Gemini. The structure is model agnostic. Claude tends to follow long chained instructions more precisely, which makes it the strongest model for stacks that depend on tight handoffs between steps. ChatGPT and Gemini both work well with all seven stacks above.
Should I run the whole stack manually each time?
You can. Better is to save each stack as a template with placeholders and paste the placeholders one at a time. Even better is to set up an automation that passes output between steps automatically. Most people start manual, then automate the stacks they run daily.
How long should each prompt in a stack be?
Most working prompts are 50 to 200 words. The whole stack might be 500 to 1500 words total. Longer than that and the prompts probably duplicate constraints across steps. Shorter than that and the steps probably skip the role anchor, format specification, or constraints that make output reliable.
Do I need different stacks for different roles?
Yes, and that is a feature. A founder's inbox triage stack looks different from a head of operations' stack. Both follow the same structural pattern. The specifics are calibrated to the role. The seven stacks in this post are written generically. You will tune them to your context over the first two weeks.
What if a stack produces bad output?
Almost always the issue is the first prompt in the chain. Garbage out of step one becomes worse garbage by step three. Fix the first prompt. The downstream prompts usually start working without other changes.
How is this different from automation tools like Zapier?
Automation tools route data between services. Prompt stacks route output between prompts. They complement each other. A mature setup uses automation to feed inputs into the first prompt of a stack and to send the final output to wherever it needs to go.
Can I share stacks with my team?
Yes, and that is one of the highest leverage moves in a team setting. A shared stack library means the team's collective prompt knowledge compounds rather than each person rebuilding from scratch. Most teams that build shared stacks see productivity gains across the whole team within a month.
Get the Pro Stack Library
The PromptLeadz Pro Stack Library expands the seven stacks above into 30 plus production ready stacks across operations, sales, marketing, finance, and executive workflows. Every stack is formatted three ways (XML for Claude, Markdown for ChatGPT, PTCF for Gemini) and includes the calibrated prompts, the integration notes, and the time savings expected.
Browse the Pro Stack Library and the rest of the PromptLeadz catalog in the shop. Free starter stacks in the Freebie Vault.
コメントを残す: