What changed recently
as of June 2026Model
Claude Opus 4.8 is now in Microsoft 365 Copilot, including Copilot Cowork on Frontier. It is built for complex, multi step work, the exact kind these routines lean on. It is also the model PromptLeadz assistants are tuned for.
Now GA
Agent Mode in Word, Excel and PowerPoint. Copilot can plan and execute multi step edits inside the file, showing each change. The pattern: create from scratch in Copilot Chat, then switch to in app Agent Mode for precise edits once the file exists.
Useful
Researcher exports in one click. A Researcher report now converts to a PowerPoint, PDF, infographic or audio overview without redoing the work. Pair it with the Cowork deep research prompt below.
Heads up
Free in app Copilot ended in April 2026. Draft, summarise and rewrite inside Word, Excel and PowerPoint now need a paid Microsoft 365 Copilot licence. Copilot Chat still works without one. Step one below checks this.
Get Frontier and Cowork switched on
Verified against the June 2026 Frontier rollout. Frontier is Microsoft's early access track inside Microsoft 365. Opt in and you get experimental Copilot features before general release, including Copilot Cowork. Tick each step off as you go. Your progress saves on this device.
Cowork is not showing up for me
Three usual causes. First, the admin account is not enrolled in Frontier, which hides Cowork from Agent management even when users are enrolled. Second, your tenant or region is outside the current rollout, which started with the United States in English. Third, you have Copilot Chat but not a full Microsoft 365 Copilot licence. Check in that order.
Is it safe to let Cowork send things?
Cowork runs inside your Microsoft 365 security and permissions boundary, shows each step as it works, and asks for approval on actions. The practical rule: for the first few weeks, end every delegation with a line like "draft only, do not send anything without my approval". You can loosen that once you trust the output.
What is the difference between Copilot Chat and Cowork?
Chat generates content and answers inside one conversation. Cowork completes multi step work across your apps: it plans, acts, creates files, sends emails and posts in Teams. Use Chat when you want words, Cowork when you want a finished task.
Will these features change?
Yes. Frontier features are previews and Microsoft iterates on them, so names, menus and capabilities shift. The prompts on this page are written to survive that because they describe outcomes, not menu paths.
Part 2 · The new piece
Copilot Cowork: Copilot that does the work, not just the words
Copilot Chat answers you. Cowork acts for you. Describe an outcome and it plans the steps, then sends emails, schedules meetings, builds Word, Excel, PowerPoint and PDF files, posts in Teams and searches across your organisation. There are three levels of trust, and most people should climb them in order.
Level 1 · Ask
Single answers, single drafts
Classic Copilot in the apps. Summarise this thread, draft this section, explain this formula. Fast, low stakes, you stay in the driving seat. The Outlook, Word and Excel packs in part 3 live here.
Level 2 · Delegate
Multi step tasks with checkpoints
Where Cowork earns its name. Give it a finished outcome, let it plan, and review at the approval points. Weekly status packs, inbox triage with drafted replies, a document built from three sources.
Level 3 · Schedule
Repeating workflows
Cowork can run prompts on a schedule. A Monday morning briefing, a monthly budget review, recurring meeting prep. Set it once, refine the output over two or three cycles, then let it run.
Delegation builder
Pick your options, fill in the detail, and this writes the Cowork delegation for you. Built on the structure that gets the best results: outcome, sources, output, checkpoint.
Your delegation
Want this prebuilt, tested and ready for any AI, not just Copilot? Assistants from $9.99 at promptleadz.com
Level 1 · Prompt
One task, one use
You paste it, it runs once. The engineered prompts on this page. Free, and locked to the app you run them in.
Level 2 · Routine
Scheduled, it runs your week
Set once in Cowork and it fires at the moment it names: Monday reset, daily shutdown, Sunday triage. The Routines pack below. Also free, also Copilot only.
Level 3 · Skill
A capability your AI carries everywhere
Cowork supports skills and plugins that package whole workflows into a reusable capability. That is exactly what a PromptLeadz assistant is: one complete skill file, built to this page's standard, that works in Cowork, ChatGPT, Claude and Gemini. Tuned for Claude Opus 4.8, the model now inside Copilot. From $9.99, yours forever.
Get a skill at promptleadz.comOne habit that changes everything: give Cowork outcomes, not instructions. "Get me ready for Thursday's client review" beats a ten line list of steps, because Cowork plans the steps itself and you correct the plan, not the prose.
Your week, scheduled
Set these routines once in Cowork and your week runs itself. Tap any one to jump to it in the Routines pack.
Mon
Tue
Wed
Thu
Fri
Sat
Rest. The point of the other six.
Sun
And when it matters, not on a clock
The five words people mix up
Copilot Chatfree tier
The conversation. Ask, get an answer or a draft, in one window. Works without a paid licence, but no longer reaches inside your documents.
Agent Modepaid
Copilot working inside Word, Excel or PowerPoint, now able to plan and execute multi step edits in the file itself. Needs a Microsoft 365 Copilot licence since April 2026.
CoworkFrontier
The agent that acts across your apps: plans, sends, schedules, builds files, posts in Teams. Where the routines and delegations on this page run.
Frontieropt in
Microsoft's early access track. Opt in at tenant and user level to get preview features like Cowork before general release.
Skillportable
A complete capability in one file that any AI carries, in Copilot and beyond. The level above a routine, and what PromptLeadz sells.
Part 3 · Start here
Your first 7 days with Copilot
Seventy prompts is a lot to land on cold. Do one thing a day for a week and Copilot stops being a novelty and becomes a habit. Progress saves on this device.
Part 3 · Prompt packs
70 engineered prompts and routines, built to a paid standard
These are engineered, not collected. Every prompt carries a method, an exact output contract, a verification step and a rule for missing information, so the output never bluffs you. Tap any highlighted field to add your detail, star the ones you use, and open "Why this works" to learn the engineering behind each one. Prefer it offline? Download all 70 as a text file.
New here? Start with your role
Tap who you are and we will surface a starter set built for your week. Tap again to see all 70.
The anatomy of an engineered prompt
Every card below is built from four layers. Tap a layer to see it inside a real one.
Read the entire thread before writing anything. Then give me three labelled sections. DECISIONS: what was agreed and by whom; if agreement is implied rather than stated, mark it ASSUMED. OPEN: unresolved questions and who owes the answer. ACTIONS: owner, deliverable, date; write NO DATE where none was given rather than inventing one. Close with one line: the single message in this thread I cannot afford to miss. If the thread contradicts itself, surface the contradiction instead of smoothing it over. 150 words maximum.
What is on fire right now?
Tap your situation and we will hand you the 3 cards that fix it. Tap again to see everything.
Outlook · 10 prompts
Use these in Copilot inside Outlook. They work on open emails, threads and your calendar.
What good output looks like · Thread triage
DECISIONS · Launch moves to 3 July, agreed by Priya and Marc · Vendor B selected, ASSUMED, Marc wrote "let's lean B" with no confirmation OPEN · Budget sign off, owed by finance, quiet for 6 days ACTIONS · Sofia: revised timeline, Friday 19 June · Marc: vendor contract, NO DATE Cannot afford to miss: Priya's note that legal review now blocks the 3 July date. Conflict found: message 4 says budget approved, message 11 says pending. Check with finance before relying on either.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Thread triage
Read the entire thread before writing anything. Then give me three labelled sections. DECISIONS: what was agreed and by whom; if agreement is implied rather than stated, mark it ASSUMED. OPEN: unresolved questions and who owes the answer. ACTIONS: owner, deliverable, date; write NO DATE where none was given rather than inventing one. Close with one line: the single message in this thread I cannot afford to miss. If the thread contradicts itself, surface the contradiction instead of smoothing it over. 150 words maximum.
The ASSUMED and NO DATE markers are the engineering. They stop Copilot doing what it loves: quietly filling gaps with plausible fiction. You get what the thread actually says, with the gaps visible.
Reply draft
Draft a reply to this email that does exactly three things: confirms [what I am agreeing to], moves the deadline to [the new date] with a reason that is honest but brief, and asks one precise question about [the unclear point]. Before you write, identify the sender's real ask, which may not be their stated one, and make sure the draft answers it. Match the formality of their email, one notch warmer. Under 120 words, no bullet points, and no openers like I hope this finds you well.
The real ask instruction is the difference maker. Half of all email friction comes from replying to the stated question while ignoring the actual one underneath it.
Saying no
Draft a decline to this request. Rules: the decision lands in the first sentence, one true reason, offer [an alternative] as a genuine option not a brush off, and close in a way that protects the relationship. One apology maximum. Before showing me the draft, reread it as the recipient and strip anything that invites them to negotiate the no. If declining would contradict a commitment visible earlier in this thread, warn me before drafting anything.The reread as recipient pass removes the soft phrases that turn a no into a maybe. The commitment check has saved more reputations than any wording ever will.
Inbox catch up
Go through everything I received since [yesterday]. Exclude newsletters, notifications and threads where I am cc only, unless I am named directly in the body. For each remaining email: sender, their real ask in one line, the deadline stated or implied, and a 2 sentence draft reply I can edit. Rank by the cost of ignoring each one, not by arrival time. If two emails conflict with each other, tell me that at the top before anything else.Ranking by cost of silence instead of recency is how senior people actually triage. The cc exclusion alone usually cuts the list by half.
Meeting prep
I am meeting [the attendees] about [the topic]. Build a one screen prep note from our email history: 1) current position in 3 lines, 2) what each attendee has asked for or pushed back on, with the email it came from, 3) any commitments I made that they may call in, 4) the 3 points I should land and the one concession I could trade. If the topic has gone quiet for over 2 weeks, say so, because the first question in the room will be why.
Point 3 is the one people skip and regret. Your own open commitments are the ambush risk in any meeting, and Copilot can find them faster than your memory can.
Escalation email
Draft an escalation to [the decision maker] about [the issue]. Structure: the situation in 2 sentences with zero backstory, the quantified impact if unresolved by [the date], the 2 things I already tried and why each failed, the specific decision I need framed so it can be answered yes or no, and the date I need it by. Strip every adjective that signals frustration. Then run this test: if the recipient forwards it to their boss unedited, do I still look measured? Fix anything that fails the test.
The forward test is the quality gate. Escalations travel, always upward, and the version that exists is the version that gets forwarded.
Long thread, late arrival
I was just added to this thread. Give me the 60 second version in four lines: 1) the decision on the table, 2) who wants what, with names, 3) where the real disagreement sits, which may not be the stated one, 4) what I am being added to do, stated or implied. If my role is not clear from the thread, write UNCLEAR rather than guessing, and point me to the one message worth rereading in full.
Line 3 is what you cannot get by skimming. Threads have a stated disagreement and a real one, and walking in knowing only the first is how you say the wrong thing confidently.
Follow up nudge
Draft a follow up to my email from [the date] about [the topic]. First check the thread and my recent mail: if they already replied somewhere I missed, tell me that instead of drafting. The nudge itself: 3 sentences, assumes good intent, restates the one thing I need in concrete terms, proposes [a new date], and gives them an easy out to delegate or decline. Banned phrases: just bumping this, sorry to chase, gentle reminder.
The reply check before drafting prevents the worst follow up failure: chasing something that arrived two days ago. The easy out raises response rates because a graceful exit beats silence.
Status update
Draft a status email to [the recipient] on [the project]. Line 1: status as one word, green, amber or red, with the reason in the same sentence. Then 3 progress bullets that name finished things, not activity. Then the single risk most likely to change the status, with my mitigation. Close with exactly what I need from them and by when. If I genuinely need nothing, end with no action needed from you. 150 words maximum, and never report activity as if it were progress.
Finished things, not activity is the line that changes the output. Worked on, aligned with and progressed are how slippage hides in status reports.
Tone rescue
Below is my draft. Rewrite it so it reads calm and professional while keeping every substantive point, including the difficult ones; soften the delivery, never the substance. Then list what you changed in 3 lines and name each pattern, for example sarcasm, hedging, blame. Finally, tell me the one sentence in my original that would have cost me the most, and exactly why.
[paste your draft here]
Soften the delivery, never the substance is the guard rail. Without it, tone rewrites quietly delete your hardest and most necessary point. The pattern naming trains you out of your own tells.
PromptLeadz
Ten prompts sharpen your inbox. One assistant runs it.
Every PromptLeadz assistant is a complete skill file built to this same standard: method, output contract, anti bluff rules. Drop it into ChatGPT, Claude, Gemini or Copilot and it works first time. $9.99 each, yours forever.
Browse the assistants at promptleadz.comWord · 10 prompts
Use these in Copilot inside Word, on a blank page or with a document open.
Agent Mode tip: use these in Copilot Chat with the Word Agent to create from a blank page, then switch to in app Agent Mode for precise, iterative edits once the draft exists.
What good output looks like · Consistency sweep
1. Page 2 states 14 percent growth; the table on page 7 shows 12 percent. The table cites the source file, so 12 is more likely correct. 2. "Phase two" is defined as Q3 on page 1 but used to mean post launch on page 5. 3. The churn claim appears as "definitely" on page 3 and "likely" on page 9. Same claim, two strengths. No date conflicts found. Otherwise clean.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
First draft
Draft a [type of document] on [the topic] for [the audience]. Before writing, state in one line the single decision this document should drive; if you cannot tell from what I gave you, ask me instead of guessing. Then write: recommendation first, 3 arguments each anchored to evidence the reader can check, and risks with the strongest counterargument stated fairly, not a strawman. Sentences under 25 words, zero buzzwords. Mark any factual claim you are not certain of with [VERIFY]. About [600] words.
The decision line and the [VERIFY] markers are the premium parts. The first stops you getting a document about a topic instead of a document for a decision. The second makes hallucinations visible instead of fluent.
Executive summary
Write a 5 sentence executive summary of this document. Sentence 1: the conclusion, with the key number in it. Sentences 2 to 4: the reasons, in order of strength. Sentence 5: the ask or next step with a date. The test: a reader who sees nothing else can make the decision. If the document does not actually support one clear conclusion, tell me that directly instead of papering over it with confident language.
The escape clause at the end matters most. A summary tool that admits the document is inconclusive is worth ten that manufacture certainty on command.
Hard edit
Cut this document by 30 percent. Protected and untouchable: every number, every commitment, every caveat that limits our exposure. Fair game: throat clearing, repeated points, adjectives doing no work, and background this audience already knows. List what you removed in 3 bullets so I can veto. If you cannot reach 30 percent without touching protected content, stop where you can and tell me exactly what the last stretch would cost.
The protected list is what makes aggressive cutting safe. Caveats and commitments are precisely what generic shorten this prompts delete first, because they read as clutter to a model.
Audience shift
Rewrite this section for [the new audience]. Keep every fact and figure identical; change the framing, vocabulary and depth to match what they care about and already know. Before rewriting, state in one line what this audience is most likely to challenge, and make the rewrite answer it early. Afterwards, list anywhere the original emphasis did not survive the shift, so I can decide whether that loss matters.The challenge prediction turns a translation into a defence. The emphasis audit catches the subtle failure: same facts, different document, because what got stressed quietly changed.
Structure check
Act as the most sceptical reviewer this document will face. Find the 3 weakest points in the argument. For each: quote the exact sentence, state the objection a hostile reader raises, rate the damage as fatal, serious or cosmetic, and propose the minimal fix. Check specifically for claims without evidence, evidence attached to no claim, and conclusions that outrun the data. Do not rewrite anything yet, and do not pad the list with cosmetic issues to appear thorough.
The damage rating forces honesty about what matters. The no padding rule matters because models inflate finding counts to seem diligent, which buries the one fatal flaw under five trivial ones.
Outline first
Before writing anything, build the outline for a document about [the topic]: a working title that states the argument, the one sentence thesis, then section headings where each carries one line on what that section must prove and what evidence proves it. Flag any section where I have asserted something you cannot see evidence for in what I gave you. I approve the outline before you draft a single paragraph.What it must prove and what proves it is the discipline. Sections that cannot answer both questions are padding, and you find that out in a 2 minute outline review instead of a 2,000 word rewrite.
Meeting notes to record
Turn these raw notes into the official record. DECISIONS: exactly as worded in the notes, do not improve them. ACTIONS: owner and date, marked NO OWNER or NO DATE where the notes are silent, never assigned by you. OPEN: unresolved questions. NEEDS CONFIRMATION: every place my notes are ambiguous, listed rather than interpreted. Then a 3 line summary on top. Drop everything else.
Do not improve the decisions sounds odd until a cleaned up decision gets disputed and the official record no longer matches what anyone remembers agreeing to. Exact wording is legal hygiene.
Counter argument
Argue against this document as its toughest reviewer. List the 5 strongest objections in order of damage. For each: the exact claim it attacks, the evidence or logic the objection rests on, and how the document should answer it, or the honest admission that it cannot. At least one objection must target the framing itself, not the content. No softball objections to pad the count.
Or the honest admission that it cannot is the clause that earns its keep. Knowing which objection you cannot answer before the meeting is worth more than five rebuttals you will never need.
Consistency sweep
Audit this document for internal conflicts: numbers that disagree with each other or with their own totals, terms defined one way and used another, dates that contradict, the same claim made twice at different strengths, and figures in the text that match no table. Report each as: location, both conflicting items quoted, and which is more likely correct based on context. If the document is clean, say clean. Do not invent findings to justify the audit.
The same claim at different strengths is the conflict humans never catch: definitely on page 2 becomes likely on page 9, and a careful client will find it even if you do not.
FAQ builder
From this document, write the 8 questions readers will actually ask, with answers of 2 sentences each grounded only in what the document says. At least 3 must be hostile questions, the kind asked in the corridor rather than the meeting. Where the honest answer is we do not know yet, write exactly that plus what would resolve it, rather than bluffing. Order by how likely each question is to come first.
Grounded only in what the document says is the leash. Without it, FAQ answers drift into claims the document never made, and you end up defending positions you never took.
PromptLeadz
This is one prompt. Our writing assistants carry the whole document.
Ready made skill files that draft, edit and pressure test like a senior reviewer, in any AI chat you already use. No setup, no prompt engineering. $9.99 each, yours forever.
Browse the assistants at promptleadz.comExcel · 10 prompts
Use these in Copilot inside Excel with your data in a table. Clear column headers double the quality of everything below.
Agent Mode tip: Excel now has Edit with Copilot and a Plan mode that maps steps before changing anything. Run the orientation and clean up prompts first, approve the plan, then let it edit.
What good output looks like · Variance story
Total change: +184K, +6.2 percent. Up: Germany +91K (+14%), France +63K (+9%), Spain +22K (+5%) Down: Italy -19K (-4%), Nordics -8K (-2%) Top 3 movers explain 78 percent of the total change: concentrated, not broad. Check: movers sum to +176K against a total of +184K. Gap of +8K shown as Other. Flag: Portugal shows +212 percent, but the base is 4K. Dramatic on paper only. One line: growth is real, but it is a two market story.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Data orientation
Before any analysis, profile this dataset: each column with its type, range and blank count; the date span and any gaps inside it; rows that look duplicated; and values that look like entry errors rather than real outliers, with your reasoning for each call. Then give the 3 most decision relevant patterns, each one sentence with its supporting numbers. No charts, no recommendations yet, and flag anything that makes this data unsafe to conclude from.
Errors versus real outliers, with reasoning, is the judgement call that separates profiling from describing. The unsafe to conclude flag has stopped more bad decks than any review meeting.
Clean up worklist
Build a cleanup worklist for this table. For each issue: column, problem type, rows affected, 2 example rows, and your proposed fix. Split fixes into SAFE, mechanical with no judgement required, and JUDGEMENT, where a human must choose, for example whether an outlier is an error or a real event. Change nothing yet. If any fix would alter downstream totals, state by how much before I approve anything.
The SAFE versus JUDGEMENT split is the control. Auto cleaning judgement calls is how a real but unusual sale becomes a deleted typo, and nobody notices until the quarter does not reconcile.
Formula explainer
Explain the formula in [the cell reference] step by step in plain words, including what happens at the edges: blanks, errors, and text where numbers are expected. Then tell me 1) one simpler way to write it, 2) one way it could silently return a wrong answer with this specific data, and 3) whether it survives new rows being added below the range. If it references other sheets, trace that chain one level deep.The silent wrong answer question is the premium part. Formulas that error are annoying; formulas that confidently return the wrong number are dangerous, and every inherited workbook has at least one.
Formula builder
Write a formula that [the calculation, described in plain words]. Requirements: it returns something sensible rather than an error when a lookup fails or a cell is blank; it survives new rows; and it hardcodes no value that belongs in a cell. Give me the formula, a one line explanation I can paste alongside it as documentation, and the one scenario where it will mislead me, so I know its limits before I trust it.The one scenario where it misleads is the line nobody else asks for. Every formula has a failure condition, and knowing it up front beats discovering it in a board pack.
Variance story
Compare [the first period or column] against [the second]. Output: total change in absolute and percentage terms, the 5 biggest movers each way with both measures, and the share of total change the top 3 movers explain. Then one plain language line on whether the shift is broad or concentrated. Verify before answering: do the movers sum back to the total? If not, show the gap as Other. Flag any mover whose percentage looks dramatic only because its base is tiny.
The reconciliation check and the tiny base flag are the credibility savers. A 400 percent increase on a base of 12 has derailed more variance meetings than any genuine number ever has.
Pivot with a purpose
Propose 3 pivot views of this data, each stated as the business question it answers, not the fields it uses. For each: the question, the fields, and what result would be surprising versus expected. Then build the one with the highest decision value and read its answer back to me in 2 sentences. Skip any pivot the data cannot actually support, and say why it cannot.
Surprising versus expected is the analytical move. If you cannot say in advance what would surprise you, the pivot is decoration, and this prompt refuses to build decoration.
Chart with a verdict
Create one chart that proves or disproves this claim: [the claim or comparison]. Choose the type that makes the verdict visible in 3 seconds, title it with the finding stated as a sentence rather than the topic, and strip every element carrying no information. First check whether the data actually supports a verdict; if it is genuinely ambiguous, build the chart that shows the ambiguity honestly and title it that way.Prove or disprove this claim beats show me a chart of X every time. And the ambiguity clause means the chart never pretends to a conclusion the data does not hold.
Top and bottom
Show the top 10 and bottom 10 rows by [the metric], with [2 or 3 context columns]. Then answer in 3 sentences: what measurably separates the groups, quoting the numbers; whether anything in the bottom group looks like a data problem rather than genuine performance; and which single bottom row would improve most with the least intervention, with your reasoning shown.
The data problem check protects people. Bottom ten lists travel fast inside organisations, and at least one entry is usually a coding error wearing a performance problem's clothes.
Scenario sketch
Estimate the effect on [the output metric] if [the input] changes by [the amount]. Show: every assumption as a numbered list, each tagged with its source in the data or its status as a guess; the calculation chain; the headline as a sensible range rather than false precision; and the one assumption the result is most sensitive to. If the data cannot support the estimate, state what is missing instead of forcing a number.
Tagging assumptions as sourced or guessed is the integrity layer. A range built on labelled guesses is honest; a precise figure built on hidden ones is a guess wearing a suit.
Handover sheet
Write the handover guide for this workbook as if the reader inherits it cold next Monday. Cover: each sheet's purpose in one line; inputs versus calculated cells, identified precisely so nobody types over a formula; the update routine, what changes each [week or month] and in what order; the 3 most fragile parts and how each one breaks; and the single check that confirms the workbook is right after an update. If you find something fragile while writing this, flag it. Do not describe it as fine.How each fragile part breaks is what a handover usually omits and a successor always needs. The closing rule stops the guide quietly papering over problems it discovered.
PromptLeadz
Like the engineering in these? It is our whole catalogue.
PromptLeadz assistants apply the same verification gates and failure handling to full workflows, not single tasks. They run wherever you chat with AI. $9.99 each, yours forever.
Browse the assistants at promptleadz.comTeams · 8 prompts
Use these in Copilot inside Teams: meetings, channels and chats. The live tracker runs during a meeting; the rest work on recaps and threads.
What good output looks like · Missed meeting recap
DECISIONS · "We ship the beta on the 20th, no further scope." (Lena, verbatim) ASSIGNED TO YOU · Pricing one pager by Thursday. Context: "Commercial framing sits with your team, the rest of us stay out of it." YOUR NAME CAME UP · Marc questioned whether the timeline survives your review. Worth hearing in full: 22:40 to 24:30.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Missed meeting recap
I missed [the meeting]. From the recap and transcript give me: decisions exactly as worded, actions with owners, anything assigned to me with the surrounding discussion quoted, and the 2 minutes I should hear in full, with timestamps. If my name came up, tell me how, and do not soften criticism if there was any.The surrounding discussion quoted is the protection. An action without its context is how you complete the wrong task confidently. The unsoftened criticism rule is there because you need the room's version, not the polite one.
Channel catch up
Summarise what happened in [the channel] over the last [7 days]: decisions made, open debates with both sides stated fairly, and anything awaiting my input. Skip social chatter unless it carries a decision inside it. If a debate got heated, characterise it neutrally. Do not pick a winner for me.
Do not pick a winner matters more than it looks. AI summaries quietly side with whoever wrote the longest messages, and you want the debate, not a verdict you did not ask for.
Live meeting tracker
During this meeting, track three things: every action the moment it is agreed, with its owner; every question parked for later; and any moment where two people are agreeing to different things using the same words. Surface that last one immediately rather than at the end. It is the expensive kind of agreement.
Same words, different things is the failure mode transcripts hide and meetings run on. Catching it live is worth more than the entire recap.
Post meeting channel post
From this meeting, draft a post for [the channel]: a 3 line summary, decisions, actions with owners and dates. Neutral register, no editorialising about how the meeting went. Mark anything left ambiguous as NEEDS CONFIRMATION rather than resolving it yourself. Draft for my review before posting.The no editorialising rule exists because a channel post is the permanent record. Great discussion today ages badly when the decisions it announces unravel by Thursday.
Pre standup brief
Before my [9:30am] standup, prepare my 3 lines: what I finished yesterday, from my sent mail and files; what I am doing today, from my calendar and tasks; and any blocker visible in threads where I am waiting on someone, with how long I have waited. Honest and specific. No activity dressed up as progress.The waiting duration on blockers is the standup upgrade. I am blocked is a complaint; I have been waiting on X for 4 days is a fact the room can act on.
Thread to decision
This thread has been circling. Lay out: the actual question being decided, which may not be the stated one; each position with its strongest argument, not its loudest; what evidence would settle it; and a proposed decision with the reasoning shown. Label the proposal as a starting point for the group, never a verdict.
Strongest argument, not loudest is the instruction that makes this fair. Threads are won by volume; decisions should be won by reasoning, and this forces the swap.
New joiner brief
Build an orientation brief for [the person] joining [the team or project]: what we are doing and why, in plain words; the key decisions already made, so they do not relitigate them; who owns what; and the 3 threads worth reading in full. Include nothing beyond their access level, and mark anything you inferred as INFERRED.
The decisions already made section is the onboarding accelerator nobody writes. New joiners reopen settled questions because nobody told them they were settled.
Tone check before posting
Before I post this to [the channel], review it: how does it land with [the audience], what is the least charitable reading someone could take, and what single change removes that reading without weakening the point. If the draft is fine as written, say fine and stop.
[paste your draft post]
The least charitable reading is the test that matters in a channel, where tone has no voice and screenshots are forever. The say fine and stop clause prevents invented problems.
PromptLeadz
Eight prompts tame Teams. A skill sits in every meeting.
PromptLeadz assistants carry this engineering into any AI you use. One complete file, zero setup, output shaped by your input every time. $9.99 each, yours forever.
Browse the assistants at promptleadz.comPowerPoint · 8 prompts
Use these in Copilot inside PowerPoint, or in Copilot chat with the deck open. Structure prompts before slides exist, audits once they do.
Agent Mode tip: build the deck from a document or notebook with the PowerPoint Agent in chat, then refine in app. Copilot in PowerPoint can also ground slides in a public webpage you point it at.
What good output looks like · Headline audit
Slide 3: "Market overview" names a topic. Rewrite: "The mid market is growing twice as fast as enterprise." Slide 8: "Our approach" names a topic. Rewrite: "We win by shipping in weeks, not quarters." Broken link: slide 6 ends on cost pressure, slide 7 opens on hiring. Missing between them: why cost pressure forces the hiring change. Titles 1, 2, 4 and 5 carry the story as written.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Deck from document
Build a deck from this document for [the audience]: one idea per slide, every headline a full sentence that states the point, maximum 25 words of body per slide, and a final slide with the ask. Everything that did not make a slide goes into the notes, never deleted. If the document only supports [6] strong slides, 6 is the correct number. Do not pad.
Headlines as full sentences is the discipline that survives every trend: read the titles alone and the argument should hold. The do not pad rule fights the urge to hit an expected slide count.
Headline audit
Read only the slide titles of this deck, in order. Do they tell the complete story on their own? Rewrite any title that names a topic instead of making a point, and flag any two adjacent slides whose titles do not connect logically, with what is missing between them. Touch nothing else in the deck.
This is the 5 minute review that finds structural problems a 50 minute review misses. If the titles do not carry the story, the slides never will.
Executive cut
Cut this deck to [5] slides for an executive audience. Keep: the recommendation, the evidence that carries it, the biggest risk, and the ask. Move everything else to an appendix rather than deleting it. Then tell me in 3 lines what the cut version loses, so I can cover it verbally if asked.Appendix, not deleted is the safety. Executives ask appendix questions precisely because the appendix exists, and what the cut loses is your verbal briefing list.
Speaker notes
Write speaker notes for every slide: 3 to 5 spoken sentences that add to the slide rather than reading it aloud, plus the one question this slide is most likely to provoke and a one line answer. Conversational register, written for the ear. Where a slide needs no notes, write none needed rather than padding.
Adds rather than reads is the line between notes that help and notes that make you sound like a screen reader. The provoked question per slide is rehearsal compressed.
Chart honesty pass
Review every chart in this deck: does the slide title claim match what the chart actually shows, is any axis, scale or baseline flattering the story, and is any comparison missing the context that would make it fair? Report slide by slide. Where a chart is honest, write honest and move on. Do not invent findings to justify the review.
The audience always contains one person who reads axes. This pass finds what they will find, while you can still fix it instead of defending it.
Story spine first
Before I build any slides, structure the argument for [the topic]: situation, complication, question, answer, then the 3 supports for the answer, one line each. If the complication is weak, say so directly, because a deck without a real complication has no reason to exist and no amount of design will fix that.Situation, complication, question, answer is the spine every persuasive deck hangs on. Testing it in 6 lines costs nothing; discovering its absence on slide 14 costs the meeting.
Rehearsal partner
Quiz me on this deck as [the toughest person in the room]. Give me the 6 hardest questions in the order they would actually come up in the meeting. Then, for each, the answer this deck supports, or NOT COVERED where it does not. The NOT COVERED list is what I prepare tonight, so do not shrink it to be kind.Order they would actually come up is the realism layer; tough questions arrive at specific slides, not at the end. The do not shrink it instruction keeps the gap list honest.
Consistency sweep
Sweep this deck for mechanical problems: fonts and sizes that vary without reason, slides that break the template, numbers that disagree with other slides or with the source document, dates that conflict, and orphaned bullets. List by slide number with both conflicting values quoted. This is an audit, redesign nothing.
Cross slide number conflicts are the credibility killer: slide 4 says 12 percent, slide 11 says 14, and the client notices in silence. Machines catch this in seconds; humans almost never do.
PromptLeadz
Slides are where weak prompts get exposed.
Our assistants carry complete deck methods, spine, build, audit and rehearsal in one capability, and they run in Copilot, ChatGPT, Claude and Gemini. $9.99 each, yours forever.
Browse the assistants at promptleadz.comCopilot Cowork · 12 prompts
Written as outcomes, not steps, because Cowork plans the steps itself. Expect it to show its plan and ask approval before acting.
What good output looks like · Deep research
WHAT WE KNOW · Category adoption grew roughly 40 percent year on year (vendor report, high confidence) · Two of our three competitors launched equivalents this quarter (their announcements, high confidence) WHAT IS CONTESTED · Payback period: one analyst house says 6 months, another says 18. Both cited. WHAT WE DO NOT KNOW · Usage beyond our pilot team. Resolved by: a 30 day telemetry pull. NEXT: three options follow, with trade offs. The evidence on option 2 is thin. That is the finding.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Morning briefing
Every weekday at [8am], prepare my briefing in priority order, never chronological: 1) anything that became urgent overnight, 2) today's meetings with one line each on what I must walk in knowing, 3) emails awaiting my reply, ranked by the cost of further delay, 4) deadlines within 3 days. One screen maximum. If nothing qualifies for a section, omit the section instead of padding it. Never inflate something to urgent to fill space.The omit empty sections rule is what keeps a scheduled briefing alive past week two. Padded briefings train you to skim, and skimmed briefings miss the one thing they existed to catch.
Minutes to actions
From the transcript and notes of [the meeting], produce two drafts for my approval: a Word record with decisions exactly as worded, actions with owner and date, and a NEEDS CONFIRMATION list for everything the discussion left ambiguous; and a follow up email to attendees containing each person's actions only, not the full record. Where the transcript is unclear about who owns an action, list it as unassigned rather than guessing. Show me both. Send nothing.Each person's actions only is deliberate. People act on a 4 line email naming their tasks and ignore a full record attachment. The unassigned rule stops Cowork volunteering people for work.
Status pack
Build my weekly status pack for [the project]: gather progress from emails, files and Teams across the last 7 days, then produce a one page Word summary, finished things first and activity second, plus a 5 line Teams post for [the channel]. Mark anything you inferred rather than found stated as INFERRED. If the week shows slippage, lead with it rather than burying it in bullet four. Draft both for my approval and send nothing.
The INFERRED marker is the trust mechanism. Cowork assembling a week from fragments will fill gaps; this makes the filling visible so you check the inferences, not the whole pack.
Inbox triage
Triage my unread email from the last 2 days into: NEEDS MY REPLY, ranked by the cost of silence; WAITING ON OTHERS, with who and since when; FYI; ARCHIVE. Draft replies for the first bucket in my voice, matched to how I actually write in these threads. Hard rules: send nothing, never archive anything containing a question addressed to me, and if a thread looks commercially or legally sensitive, surface it separately and do not draft a reply at all.
The sensitive thread carve out is the rule you set before you need it. Some replies should never start from an AI draft, and the time to decide which is now, not after one goes out.
Scheduling
Schedule 45 minutes with [the first person] and [the second person] within the next week. Constraints: avoid my focus blocks, avoid creating back to back meetings where possible, prefer their working hours where visible. Propose 2 options to me first. Once I choose, send the invite with this agenda: [the 3 agenda points], attaching any relevant pre read from our recent files on the topic. If no slot satisfies every constraint, show me the least bad 2 options and what each violates.
The least bad clause is the difference between an agent and a script. Real calendars rarely satisfy every constraint, and you want trade offs surfaced, not a silent failure or a silent violation.
Document from sources
Create a Word document on [the topic] from [the first file], [the second file] and my email thread with [the person]. Structure: summary, current position, options with a fair case for each, recommendation. Rules: every factual claim traces to one of the three sources; where sources disagree, show the disagreement in a clearly marked box rather than averaging it away; where all three are silent on something the document needs, write GAP plus what would fill it. Draft for my review.
Show the disagreement, never average it is the rule that matters. Multi source synthesis loves to blend conflicting figures into one smooth wrong number, and a marked box makes the conflict a feature.
Deep research
Research [the question] across our internal files and the web. Output a 2 page brief in four sections. WHAT WE KNOW: each point with its source and a confidence level. WHAT IS CONTESTED: where credible sources disagree, with both sides presented. WHAT WE DO NOT KNOW: each gap with what would resolve it. WHAT I SHOULD DO NEXT: 3 options with trade offs, not one foregone conclusion. Prefer primary sources over commentary. If the evidence is thin, that is the finding. Report it.The four section structure exists because research has four honest states of knowledge and AI output usually collapses them into one confident voice. Thin evidence reported as thin is a premium answer.
Monthly review
On the first Monday of each month, pull [the budget file or report] and compare actuals to plan. Build an Excel summary of every variance over [the threshold], each with its driver where the data shows one and DRIVER UNKNOWN where it does not. Then draft an email to [the recipient] with only the 3 variances most worth a conversation, chosen by size and controllability together, not size alone. Verify totals reconcile before drafting; if they do not, the reconciliation gap becomes item one.
Size and controllability together is how a finance head actually picks talking points. The biggest variance is often uncontrollable and therefore the least useful thing in the meeting.
Pre meeting dossier
Two hours before my meeting with [the person or company] on [the date], deliver a dossier: our email history in 5 lines; every open commitment on both sides with its original date, mine listed first because mine are the ambush risk; relevant files with one line on why each matters; anything recent and public about them worth knowing; and 3 questions I should ask, at least one of which I will be tempted to avoid. If a commitment of mine is overdue, lead the dossier with it.
Mine listed first inverts the natural instinct to audit the other side. The question you are tempted to avoid is usually the one the meeting actually turns on.
Handover pack
I am away from [the start date] to [the end date]. Build a handover for [the person covering]: my active projects reconstructed from recent emails and files, each with status, the next decision due and what lands during my absence; who to contact for what; and a section marked JUDGEMENT CALLS for anything where I should write the guidance personally because you cannot know my intent. Mark anything you could not verify as UNVERIFIED rather than presenting it as fact. I edit before anything is shared.
The JUDGEMENT CALLS section is the honest boundary. A handover that admits what the AI cannot know is safer than one that fills every gap, because your cover will trust the wrong parts of a complete looking pack.
Stakeholder pulse
Monthly, rebuild my stakeholder picture for [the project] from our actual correspondence: who has gone quiet that used to be active, with the date of their last substantive message; who has newly entered the threads; and anyone I have not exchanged a real message with in over [3 weeks] that I probably should. Evidence only, quoted where it matters. No personality theories.
Gone quiet is the earliest signal a project gives and the last one anyone checks. The no personality theories rule keeps this an evidence report, not amateur psychology.
Decision log keeper
Maintain a running decision log for [the project]: every decision found in meetings, mail and Teams, each with date, decider and a link to where it lives. Every [Friday], show me the new entries plus anything that contradicts an earlier logged decision. The contradictions are the entire point of this log. Never smooth them over, and never log an implied decision as a made one.
Projects do not fail from bad decisions as often as from contradictory ones nobody noticed. A log that hunts contradictions is governance that runs itself.
PromptLeadz
Cowork runs these inside Microsoft 365. Our assistants run everywhere.
Complete, ready made AI assistants shaped by your own input every time. ChatGPT, Claude, Gemini, Copilot, one file, zero setup. $9.99 each, yours forever.
Browse the assistants at promptleadz.comRoutines · 12 prompts
The week, scheduled. Set each one in Cowork once and it runs at the moment it names: Monday reset, daily shutdown, Friday wrap, Sunday triage, back from leave, the night before a heavy day.
What good output looks like · Sunday evening triage
Clear runway, except one thing: the owner of Monday's 9am moved it to 8:30 and added a pre read, about 12 minutes. Everything else waits.
A real shaped response: labelled sections, the anti bluff markers doing their job, and nothing invented to look complete.
Monday morning reset
Every Monday at [7:30am], before anything else: 1) rebuild my week from the calendar, flagging any day with more than [4] hours of meetings as a focus risk, 2) list what I promised people last week that is still open, from my sent mail, 3) surface the emails from the weekend that change my plans, if any, 4) name the one thing this week that does the most damage if it slips. One screen. If the weekend was genuinely quiet, say quiet. Do not manufacture urgency.
Point 2 is the engine. Your own open promises, mined from sent mail, are the Monday list nobody keeps and everybody needs. The do not manufacture urgency rule keeps the reset trustworthy.
Daily shutdown
Every weekday at [5:45pm], close my day: 1) anything I was asked today and have not answered, with a drafted holding reply wherever the full answer needs tomorrow, 2) tomorrow's meetings that genuinely need preparation tonight, and only those, 3) one line on what actually moved today versus what merely generated email. Queue the holding replies for my approval and send nothing. If nothing needs me, say so in one line and let me leave.The holding reply is the underrated move: it buys you the evening without leaving anyone hanging. And moved versus generated email is the most honest daily metric there is.
Friday wrap
Every Friday at [4pm], build my week wrap: 1) what shipped this week, finished things only, 2) what slipped, with the honest reason from the email trail rather than the polite one, 3) a drafted status note to [my manager or team] that I can edit, 4) the 3 things Monday me will want already lined up. Mark anything inferred rather than stated as INFERRED. Draft only, send nothing.
The honest reason versus the polite one is the line that makes this useful. You can always soften the truth in your edit; you cannot recover the truth from a softened summary.
Sunday evening triage
Every Sunday at [6pm], prepare Monday without ruining Sunday: scan what arrived since Friday [5pm] and sort it into needs a reply Monday morning, can wait, and ignore. Flag anything that changes Monday's first meeting. Maximum 10 lines, no drafting, no actions. This is reconnaissance, not work. If nothing requires Monday morning attention, the entire output is one line: clear runway.
The reconnaissance not work rule is the boundary that makes a Sunday check healthy instead of corrosive. Clear runway as a complete output is the best sentence Cowork can send you.
Back from leave
I was away from [the start date] to [the end date]. Rebuild my world before I open the inbox: 1) decisions made in my absence that affect my projects, with who made them, 2) threads where I am currently blocking someone, ranked by how long they have waited, 3) what can be archived unread, as a count rather than a list, 4) the 3 conversations to have today. Mark anything you could not verify as UNVERIFIED. Do not reply to anything.
Threads where I am blocking someone is the post leave list that matters most and gets found last. Reading this before the inbox turns a brutal morning into a directed one.
Big day eve
Tomorrow is heavy: [describe the day]. Tonight, prepare: for each meeting, the one thing I must walk in knowing and the one thing I must walk out with; any open commitments of mine the attendees could raise, mine listed first; the files I should have open; and any gap where I am exposed with no time left to fix it, so I am at least not surprised. Deliver by [8pm] tonight. Honest beats reassuring.
Walk in knowing and walk out with turns a diary into a strategy. The exposed gaps section exists because the worst surprises are the ones you could have at least braced for.
Month end close
On the last working day of each month at [3pm], close my month: what I shipped, from sent mail and files; promises I made that land next month, with their dates; expenses or admin still open, judged from receipts and reminders in my mail; and one line per active project on trajectory, improving, flat or slipping, with the evidence for the call. Draft only, and mark inferred items INFERRED.The trajectory word per project, forced to come with evidence, is a monthly habit that compounds. Twelve of these and you have the honest year in twelve lines.
Quarter kickoff
On the first working day of each quarter, rebuild my commitments for the quarter from recent mail and documents: each with who expects it and when, flag any pair that conflict on timing or resources, and propose the [3] I should renegotiate now rather than miss later, with a draft message for each renegotiation. Honest beats comfortable. Send nothing.Renegotiate now rather than miss later is the senior move this routine automates. A deadline moved in week one is planning; the same deadline moved in week eleven is failure.
Recurring 1:1 prep
Every [Wednesday] at [8:30am], prep my 1:1 with [the person]: what they sent or flagged since we last met; open actions in both directions, mine listed first; anything I praised or promised and have not yet followed through on; and one suggested topic beyond the urgent. Five lines maximum. If there is genuinely nothing, say so. A short 1:1 is allowed.
The promised and not followed through line is the trust engine. People forgive a busy manager; they remember a forgetful one, and this routine makes forgetting impossible.
Travel day
The evening before I travel on [the date]: list what in my calendar clashes with transit, with a drafted reschedule note for anything that cannot survive; build an offline pack of the files I will plausibly need; and split the day's items into handleable by phone in transit versus waits until I land. Queue everything for my approval. Send nothing tonight.The phone versus waits split is the travel day saver: transit time becomes the call list, and nothing important gets attempted on airport wifi.
Post meeting debrief
Same day after [the big meeting], debrief me: what was actually decided versus what different people may believe was decided, and exactly where those versions differ; my commitments with dates; their commitments with dates; and a drafted follow up note that pins down the points most likely to drift, ready for my edit. The gap between decided and believed is the entire deliverable.Every important meeting produces two or three private versions of what happened. The follow up note that lands first usually becomes the official one, and this routine makes sure it is yours.
Friday admin sweep
Every Friday at [2pm], sweep my admin: expenses I have not filed, judging from receipts in my mail; approvals waiting on me, with how long each has waited; next week's recurring meetings that have no agenda, listed so I can cancel or fix them; and any renewal or subscription notice that arrived this week. Present it as a 15 minute checklist. Action nothing yourself.The agendaless recurring meetings line quietly returns hours to your calendar. Most standing meetings survive on momentum, and a weekly list is how momentum gets questioned.
PromptLeadz
Routines run your week inside Copilot. Skills run it anywhere.
A skill is the level above a routine: a complete assistant in one file that any AI can carry, built to the same standard as everything on this page. $9.99 each, yours forever.
Browse the assistants at promptleadz.com
Part 4 · Field notes
Seven ways people waste Copilot
From watching real teams adopt it. Each one has a fix on this page.
01
Topic prompts
"Summarise this" names a topic and specifies nothing, so you get a shape lottery. Every usable prompt carries an output contract: sections, caps, orderings.
02
Trusting the filled gaps
Copilot fills missing dates, owners and reasons with plausible fiction, fluently. The fix is forcing markers: ASSUMED, NO DATE, INFERRED. Visible gaps, not smooth lies.
03
One shot drafting
Asking for 2,000 words cold means restructuring 2,000 words later. Approve a 6 line outline first; fix structure when it costs 2 minutes.
04
Full trust on day one
Letting Cowork send on day one is how one bad draft ends the experiment. Start at draft only, climb the trust ladder deliberately.
05
Charts without a claim
"Make a chart of sales" produces decoration. "Prove or disprove this claim" produces a verdict. The title should state a finding, not a topic.
06
Measuring activity, not progress
Worked on, aligned with, progressed: that is how slippage hides. Force every status output to name finished things, and slippage has nowhere to live.
07
Rebuilding the same prompt every day
If you paste the same engineered prompt every morning, what you actually need is the level above: a skill your AI carries permanently, in Copilot and everywhere else. That is the one thing this page does not give away.
Fix it for $9.99 at promptleadz.com
Part 5 · The business case
What is this actually worth to you?
Set your real weekly volumes. We will estimate the time these prompts and routines hand back, and draft the message that justifies a licence to whoever signs off.
Emails you triage or draftper day
10
Meetings you prep or recapper week
5
Documents you write or editper week
3
Decks you buildper month
2
0
hours back per week
That is 0 hours a year, or
0 working days handed back.
0 working days handed back.
A deliberately conservative estimate, counting only minutes the engineered prompts realistically save per task. Treat it as the floor, not the ceiling. Even halve it and the case usually makes itself.
Want this working in every AI, not just Copilot? Skills from promptleadz.com, $9.99 each
Beyond Copilot
These prompts live inside Microsoft 365. Your assistant should not have to.
PromptLeadz builds ready made AI assistants: complete skill files you drop into any AI chat, including ChatGPT, Claude, Gemini and Copilot. No setup, and the output is shaped by your own input every time. Each one is $9.99, yours forever.