The Deck That Built Itself

Welcome to Your AI Toolkit

You needed budget for something that mattered. You built the deck. Twenty-two slides, clean formatting, every number sourced. And somewhere around slide four you watched the person who decides stop looking at the screen. The information was all there. The argument never was. Everyone can make slides now. Making someone care is still the hard part. This playbook gives you one prompt that turns messy notes, a transcript, or a whiteboard photo into a presentation built to get a decision, plus the follow-up moves that keep it from falling apart under questioning.

• You don't need to use everything at once. Start with the next deck you actually have to build.
• This works across platforms: Copilot, ChatGPT, Gemini, Claude, or your company's internal AI tool.
• The prompt includes a main version (for licensed tools) and an alternate version (for free Copilot or tools with limited input).
• AI will get things wrong. Every prompt here is built to be checked, not trusted.

How to Use This Playbook

• Start with the Belief Gap prompt. Paste your raw material and let it build the argument before any slide exists.
• Answer the discovery questions yourself. This is the only moment where your expertise enters the work. If you let AI invent the answers, you get a confident deck built on facts nobody can source.
• Read the objection pass as a to-do list, not a critique. It tells you what to go build before the meeting.
• Close the weakest link. One real invoice beats a clean estimate every time.
• Then, and only then, build slides. Prompt, outline, deck. In that order.

The Belief Gap Framework

Every deck that fails, fails in the space between what your audience believes today and what they would have to believe to say yes. Slides do not close that gap. An argument does.

Key Insight: The deck was never the bottleneck. The argument was. Answer those four questions and the deck is basically written. Skip them and you get context, context, context, and a hopeful last slide. Headlines Are Claims, Not Labels

• A slide title that names a topic makes the reader wait for the point. "Q3 results" tells them nothing.
• A slide title that makes a claim delivers it. "Q3 margin held because we stopped expediting" is the whole slide in one line.
• Change nothing else about your deck and this single fix improves it. It also forces you to know what each slide is actually for.

The AI Efficiency Ladder

Know where you are. Know where you're going.

Your goal: Move from Level 1 to Level 2. The prompts in this playbook get you there.

Where Skills Fit (the Level 2 to Level 3 move) Level 3 has two different doors, and most people only know about one of them.

A Project knows who you are. A Skill knows how you work.

Why this prompt is a good Skill candidate It is not really a prompt. It is a five-step method with a pause in the middle, a fixed output order, an anti-fabrication rule, and a formatting standard for headlines. Every time you paste it, you are re-teaching the model something it could just know. As a Skill, it can also carry things a prompt cannot: the objection-closing playbook, the three-column comparison structure, the capacity comparison logic, and the tells list. So instead of "here are your three objections," you get "here are your three objections, and here is the comparison table you need to build to answer the first one." The honest bar: a Skill pays off when the work recurs. For one deck, a saved prompt is plenty. Persuasive decks recur.

The Bad Prompt (and Why It Fails)

This prompt circulates widely. It looks thorough. It produces a competent deck that loses the room.

ROLE: You are Dr. Alex Morgan, a renowned presentation strategist who has helped Fortune 500 CEOs create memorable, results-driven presentations.

CONTEXT: I need to create a [LENGTH] slide presentation about [TOPIC] for [AUDIENCE]. The goal is to [DESIRED OUTCOME], and the tone should be [TONE].

REQUEST: Create a complete slide deck including:
1. Engaging opening slide
2. Clear agenda/overview
3. Main content slides
4. Supporting data/examples
5. Call to action/next steps
6. Q&A preparation

QUESTION: Please ask any clarifying questions

The pattern to learn: this is a generation prompt wearing the costume of a strategy prompt. It produces slides about a topic. You need a deck that gets a decision. Every fix in the next section comes from closing one of these eight gaps.

Key Prompts from This Lab

One prompt, built to think before it builds. Everything else in this playbook supports it.

Prompt #1: The Belief Gap Deck Builder

What it's for: Turning messy notes, a transcript, or a whiteboard photo into a short presentation that gets a decision rather than covers a topic. When to use it: Budget requests, capital approvals, selling an idea internally. Anything where the why has to land before the details do.

ROLE
You are a senior communications strategist who builds decks for executive decision meetings in manufacturing and automotive. You have sat in the room when a budget request died on slide three. You care more about whether the argument holds than whether the deck looks good.

CONTEXT
I am going to paste raw material. It may be messy: meeting notes, a transcript, bullet fragments, or text pulled from a whiteboard photo. It will be incomplete and out of order. I need to turn it into a short presentation that gets a decision, not a presentation that covers a topic. The audience is a specific person or small group with the authority to say yes or no. If a number or claim goes in front of this audience and it is wrong, I lose credibility I do not get back.

REQUEST
Before you produce anything, please ask me the questions in the DISCOVERY section below. Ask them all at once, in a single numbered list. Wait for my response. Once I answer, accept whatever I write, even if it's a paragraph or covers multiple questions at once. Do not re-ask the full list. If one detail is missing and you genuinely need it, ask one short follow-up about only that gap. Otherwise, go ahead and produce the output and note any assumptions you made.

Only after I have answered the DISCOVERY questions, produce:

1. THE BELIEF GAP
One short paragraph. What this audience believes today, what they need to believe to approve, and the single claim that has to carry the weight in between.

2. THE SPINE
The argument in six sentences or fewer, before any slides exist. If the argument does not hold in six sentences, say so and tell me what is missing.

3. THE SLIDE PLAN
One line per slide. Each line has the slide's headline written as a claim, not a label, and one sentence naming what evidence sits on that slide. Recommend the slide count based on what the argument needs. Do not pad to a target. Do not include an agenda slide unless I ask for one.

4. THE OBJECTION PASS
The three strongest objections this audience will raise, ranked by how likely each is to kill the request. For each one, name the slide that answers it, or tell me the deck currently has no answer.

5. THE WEAKEST LINK
The one claim in my material that is least supported. Say plainly what would make it hold.

Use only what I give you. Do not invent numbers, dollar amounts, dates, customer names, part numbers, or results. If something is unclear or missing, flag it inline as [CLARIFICATION NEEDED] instead of guessing. Keep dates, owners, numbers, and terminology consistent across all five outputs. Write slide headlines the way a person would say them out loud. No title case, no colons splitting a label from a subtitle, no "Key Takeaways."

DISCOVERY
1. Who is in the room, and which one person actually decides?
2. What exactly am I asking them to approve or do, and by when?
3. What do they believe about this today, including anything they have already pushed back on?
4. What is the cost of doing nothing, stated in their terms?
5. How much time do I have, and is this a live meeting or a send-ahead?
6. What is the one number or piece of evidence I am most confident in?

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

STEP 0 (do this before anything else)
Read the bracketed details below. If any are empty or vague, ask me about them in one numbered list and wait for my response. Don't search for context, don't draft anything, don't pull from prior conversations. Just ask the questions.

STEP 1 (only after I answer Step 0)
Produce the deliverable below.

---

You are a communications strategist who builds decks for executive decision meetings in manufacturing.

MY RAW MATERIAL: [paste messy notes, transcript, or whiteboard text]
WHO DECIDES: [name and role]
WHAT I'M ASKING FOR: [the decision, and by when]
WHAT THEY BELIEVE TODAY: [current position, past pushback]
COST OF DOING NOTHING: [in their terms]
MY STRONGEST EVIDENCE: [the one number I trust]

Produce, in this order:
1. The belief gap in one paragraph.
2. The argument in six sentences or fewer.
3. A slide plan, one line each, headline written as a claim not a label, plus what evidence sits on it. Recommend the count. No agenda slide.
4. The three objections most likely to kill this, ranked, each mapped to the slide that answers it or flagged as unanswered.
5. My weakest claim and what would make it hold.

Use only what I gave you. Do not invent numbers, dates, names, or results. Flag gaps inline as [CLARIFICATION NEEDED].

Here is a photo of the whiteboard from our Tuesday shift meeting. Three people wrote on it. One thing is circled in red and one line is crossed out.

What's on it: CMM #1 down 3x since March, four days each waiting on Zeiss service. Outside lab (Precision Metrology) at $1,850 per PPAP submission, eleven last quarter. Inspection queue backed up six days at its worst point in May. Two lots shipped on deviation because we could not measure in time. New Zeiss quote $312K installed, 14 week lead. Refurb around $180K with no service contract. Already declined once in January, told to run the existing one harder. Cap committee meets Sept 12. New crossover bracket program launches Q1 and adds about 40 PPAP submissions a year.

Answers to your six questions:

1. WHO IS IN THE ROOM: COO, CFO, Ops Manager, and the Inspection Supervisor. The CFO holds the cash veto. The Ops Manager sponsors it. The Inspection Supervisor owns the data.

2. WHAT I AM ASKING FOR: Approve capital for a new CMM by end of quarter, so we can hit the 14 week lead time ahead of the crossover bracket launch.

3. WHAT THEY BELIEVE TODAY: That we can keep paying overtime and rehab the machine that is down. This was already declined once in January. The read in the room is that measurement is not the constraint.

4. COST OF DOING NOTHING: A quality event we cannot contain, and a launch we cannot support. Deviation shipments continuing, and outside lab spend that does not stop.

5. TIME AND FORMAT: Send-ahead first, then a live meeting. It has to survive being read alone before anyone is there to defend it.

6. STRONGEST EVIDENCE: The two deviation shipments, and the six day inspection backlog at its worst point in May. Both are documented.

1. WHO IS IN THE ROOM: COO, CFO, Ops Manager, and the Inspection Supervisor. The CFO holds the cash veto. The Ops Manager sponsors it. The Inspection Supervisor owns the data.

2. WHAT I AM ASKING FOR: Approve capital for a new CMM by end of quarter, so we can hit the 14 week lead time ahead of the crossover bracket launch.

3. WHAT THEY BELIEVE TODAY: That we can keep paying overtime and rehab the machine that is down. This was already declined once in January. The read in the room is that measurement is not the constraint.

4. COST OF DOING NOTHING: A quality event we cannot contain, and a launch we cannot support. Deviation shipments continuing, and outside lab spend that does not stop.

5. TIME AND FORMAT: Send-ahead first, then a live meeting. It has to survive being read alone before anyone is there to defend it.

6. STRONGEST EVIDENCE: The two deviation shipments, and the six day inspection backlog at its worst point in May. Both are documented.

Where This Goes Next: One Prompt, Five Tools The same prompt runs in Claude, ChatGPT, Copilot, and Gemini. What changes is what happens after the thinking. In Copilot, open your branded template first and run the prompt in the side pane. The builder uses the layouts already in the open file, so your branding survives. In Claude or ChatGPT, save it as a Skill or a Project so it is one click away. In Gamma, paste the slide plan and get a formatted deck in seconds. Same prompt structure. Different ending. The people who get good at architecting prompts now are the ones ready when the tools around them get better.

Answering the Objection Pass

The prompt hands you three objections and a weakest link. That output is not a criticism of your deck. It is a to-do list with a deadline.

"You haven't shown this is financially better than the cheaper option."

This is almost always the objection that kills a capital request, and it is almost always answered with a number you do not have yet. What you’d build before the meeting: a three column comparison over the same time window.

You do not need precision. You need every column built the same way so the comparison is honest. A defensible estimate beats a missing number. The line that lands: the cheaper option is only cheaper until the year it isn't. Show them the year.

"Why can't we just refurbish for less?"

Refurbishment is rarely defeated on price. It is defeated on capacity and on risk. Three things to establish:

• Capacity. A refurbished machine is still one machine. If your constraint is throughput, refurbishing does not add a lane.
• Warranty and uptime. No service contract means the same waits that caused the problem in the first place. Name that explicitly.
• Useful life. How many years does the refurbish buy, and what happens at the end of them? The trap to avoid: do not argue that refurbishment is a bad machine. Argue that it solves a different problem than the one you have. That distinction is what keeps you from being walked into a cheaper yes that leaves you exposed six months later.

"The launch isn't actually at risk."

This one gets answered with a capacity comparison, not a story. Three numbers, one slide:

1. Inspection hours required during launch
2. Dependable available hours today, meaning hours you can count on including outages, not theoretical capacity
3. Available hours after the new equipment If line 1 exceeds line 2, you have proven the constraint. If it does not, you do not have this argument and you need a different one. Better to know that now than in the room. The supporting move: connect the deviation shipments directly to launch consequence. Two deviations during normal volume is a problem. Two deviations during a launch is a containment event with a customer watching.

The Weakest Link

The prompt names the one claim least supported by your material. Treat closing it as the highest value hour of work you can do before the meeting.

Watch what different tools do here. Run the same prompt in two places and you may get two behaviors. One will invent a plausible cost and flag it as the weakest link. Another will refuse to invent it and tell you to go get a real invoice. Neither tool is reliably the careful one. That is why the check is you, not the tool.

The Tells That Make a Deck Look AI-Built

These are the adjustments I found the hard way. Most take under a minute.

One caveat on the agenda slide. Some leadership teams expect one, and some people genuinely use it to orient themselves. If that is your culture, keep it. The point is not that agenda slides are forbidden. The point is that AI adds one by default, and a default is not a decision.

Two things to watch for in the output itself The tools quietly make decisions in the gaps you left. One swapped a template footer without being asked. One reordered slides and mentioned it in a sentence at the end. Read the summary paragraph after the build, not just the deck. And the prettier deck is not the more accurate deck. Check every number against your source. You cannot tell by looking.

The Eight-Deck Comparison

Same whiteboard. Same six answers. Same hard rule: invent nothing, flag every gap. Eight builds across every tool. Then we had Claude score them against weighted criteria and rank them.

The criteria, and what carried the most weight

The three results worth remembering

• Copilot was the most sendable. It invented nothing and flagged every missing number. Least flashy deck in the set, and the only one you could send tomorrow with confidence.
• Gamma was the most dangerous. Best looking deck by a wide margin. Invented numbers presented as fact, with the disclosure banners and notes stripped out. That is the version that loses the room.
• The bad prompt was the floor. Same tool, same whiteboard, same six answers. The number that makes the point

Same tool. Same source material. The entire difference was the prompt. One more finding worth sitting with The free version of Copilot Chat scored highest of all eight decks. It invented nothing and told us exactly where clarification was needed. Read that carefully, because it is easy to take the wrong lesson from it. It does not mean the free tier is where you should live. It means that on this one narrow task, with a well-built prompt and six good answers, the tier was not the variable. The prompt was. If your company has you on free Copilot Chat, you are not blocked from doing this work well. Start now. But the capabilities we cover in the rest of this series (memory, plugins and connections, skills, scheduled tasks, and grounding in your own files) live on the paid side. If you are serious about this, a licensed version is worth the money, whether your company provides it or you buy your own.

Failure is not invention. It is invention without disclosure. Every tool in this comparison had to fill some gaps. The ones that scored well said so, with banners, notes, and inline flags. The one that scored badly stripped the flags out and let invented numbers read as sourced fact. When you hand a deck to a decision maker, the risk is not that AI helped. The risk is that nobody can tell which parts it made up.

Six lessons from the run

1. The prompt matters more than the tool.
2. That is not a case for staying on the free tier. It is a case for building better prompts, then giving them a tool that can actually reach your files, your calendar, and your systems.
3. The best-looking argument in the set was built on invented numbers.
4. Failure is invention without disclosure, not invention itself.
5. Only the disciplined tools flagged gaps instead of guessing.
6. Most tools default to label headlines. Claim headlines have to be asked for.
7. Watch for the quiet errors, not just made-up dollars. Overclaims and misread numbers do the same damage.

Quality Check: Before This Deck Goes Anywhere

• Does every slide headline make a claim? If a title names a topic instead of stating a point, rewrite it.
• Can you say the argument in six sentences without the deck? If not, the deck is carrying weight the argument should carry.
• Is the ask specific? What exactly are you approving, who owns it, by when. If that is fuzzy, the answer will be "let's discuss next quarter."
• Did you close the weakest link, or at least name it? Assume they will find it. Better that they find it labeled.
• Does it survive being read alone? For a send-ahead, nobody is there to fill in the gaps. Read it cold.
• Did you verify every number against your source? Especially the ones you did not personally put on the whiteboard.

How to Build the Deck in Each Tool

The thinking is the same everywhere. Only the buttons change.

The pattern underneath all five. Do the thinking in a chat tool. Get the slide plan. Then hand the plan to whatever builds slides. The builder is a formatter, not a thinker, and it will format a bad argument just as happily as a good one.

Copilot Basics (If You Use Microsoft 365)

• Work IQ (the toggle at the top left): this is what used to be the Work and Web split. Switched on, Copilot can see your organization's files, email, Teams, and SharePoint. Switched off, it works from what you give it and the open web. Turn it on when the deck depends on your actual documents. Turn it off when you want a clean draft with no interference. Work IQ is powerful and occasionally overeager, which is why the alternate prompt opens by telling it to ignore prior context.
• The Chat and Cowork tabs: Chat is the conversation you already know. Cowork is the newer mode where Copilot takes on longer multi-step work using Skills. Start in Chat for this prompt.
• The left rail: New chat, Search, Library, and Notebooks sit at the top. Below that is your Agents list, which includes Microsoft's built-in ones (Researcher, Analyst, PowerPoint) plus any your company built, plus New agent if you want to make your own.
• The PowerPoint agent is in that Agents list. It produces better slides than plain chat but pushes its own themes and asks for a slide count up front, so you lose your brand template. Use it for speed, not for anything client-facing.
• The side pane inside each app: Word, Excel, PowerPoint, Outlook, and Teams each have their own Copilot pane that works on the file in front of it. That is the one that respects your branded template.
• Notebooks: a grounded workspace. Load your sources and it answers only from them.
• The path that works for decks: open your branded template first, then open the Copilot side pane inside that deck, then paste the prompt with your raw material attached. Answer the six questions. Review the outputs. Then tell it to build the slides. Doing it in that order is what keeps your branding, because the builder uses the layouts already in the open file.
• Getting your template to show up: save your branded deck as a PowerPoint Template (.potx) into Documents, then Custom Office Templates, on your own machine rather than SharePoint. Restart PowerPoint. It appears under File, New, Personal. If IT locks that folder, use Design, then Themes, then Browse for Themes and point at the .potx. That carries fonts and colors but not always custom layouts, so treat it as the fallback.
• You can now choose the model. There is a model picker next to Work IQ at the top. You can select which underlying model runs your request, including Claude Opus 5. Worth knowing when one model gives you a weak answer and you want to try another without leaving Copilot.
• Prompt Lab is where saved prompts live. Under the message box you will see a row of suggestion chips: File Insights, Inbox Triage, People Search, Meeting Prep, and then three dots. Click the three dots to open Prompt Lab. The left side has Your saved prompts at the top, then Prompt topics (Analyze, Ask, Assist, Catch up, Code, Collect, Create, Edit, Explore, and more), plus a keyword search at the bottom. Microsoft's own prompts are marked Suggested by Microsoft, and each card has a bookmark icon to save it.
• You can schedule saved prompts to run on their own. A weekly calendar review, a daily email summary, a Monday morning news scan. Set it once and the output shows up without you asking.
• Save this prompt there. Run it once, then bookmark it so it lands under Your saved prompts. A prompt you have to hunt for is a prompt you will not use. This one is long enough that retyping it is not an option.
• A note on Copilot Cowork. It has moved to usage-based credits. If you build something substantial in it, check your credit balance before you rely on it live. Running out mid-demo is a real possibility.
• Reminder: follow your company's AI policy. If you are unsure whether a document can go into a tool, ask before you upload it.

Gemini Notebook: The Other Half of the Work

Everything above is about output. This is about input. If you walk into a room with an AI-built deck, you had better know what it says. This is how you get up to speed on a topic fast enough to defend it. Gemini Notebook was previously called NotebookLM. Same tool, new name, and you will still see the old name in places.

What it can produce from your sources Audio overview (two hosts talking, roughly twenty minutes, with an interactive mode where you can ask questions mid-conversation). Mind map. Infographic overview. Blog post. Quiz. Flashcards. Slide deck. Report. Video. Three ways people are actually using it

• Learning a topic before you have to present on it. Pull in a training channel's videos and listen to the audio overview on the drive in.
• Learning a piece of software. Load the vendor's training videos, then ask questions against only those videos instead of hunting through documentation.
• Capturing someone's thinking. Pull a speaker or author's full channel into a notebook, then use it as the back end for a custom GPT or project built around their frameworks. What makes it different: it does not pull from the web. It answers only from what you loaded. If the answer is not in your sources, it tells you rather than inventing one. That makes it a research tool, not a drafting tool. What not to use it for: building the persuasion deck. It will summarize your whiteboard rather than argue from it. Learn the subject here, then make the case with the Belief Gap prompt. ⭐ YouTube to Notebook Extension: Install it once, and a YouTube to Notebook button appears on any YouTube channel. Click it and it pulls every video from that channel into a new notebook and transcribes them all.

Load two or three sources, not one. The audio overview is noticeably better when it has more than one thing to synthesize. And the video needs captions, or it will not process.

Staying Current Without Drowning

This came up in the session. AI changes weekly, and nobody has time to keep up with all of it. You do not need all of it. You need the part that touches your job. Two sources worth your time

If TLDR feels too dense, covers similar ground with a more practical, how-to angle for general professionals. Better than reading everything: have AI filter it for you Save this as a scheduled task and it runs every week without you asking. In Copilot, save it in Prompt Lab and set a schedule. In ChatGPT or Claude, save it as a project or a scheduled task.

ROLE
You are a practical, human-centered AI adoption advisor. You write for people across all skill levels and life roles, and you care about what real people actually do with AI rather than what the technology can theoretically do.

CONTEXT
I want to stay meaningfully current with AI without drowning in noise. Here is my profile so you can filter everything through what actually matters to me:

MY ROLE OR PROFESSION: [job title or industry, e.g. quality manager, HR business partner, plant supervisor]
MY CURRENT AI EXPERIENCE LEVEL: [beginner / occasional user / regular user / power user]
THE AI TOOLS I CURRENTLY USE: [list tools, or none yet]
MY BIGGEST INTEREST IN AI: [e.g. saving time at work, writing better, automating reports]
PARTS OF MY PERSONAL LIFE AI COULD HELP WITH: [e.g. parenting, side business, health, finances, hobbies]
TIME I CAN REALISTICALLY SPEND ON AI EACH WEEK: [e.g. 15 minutes, 1 hour]

REQUEST
Generate my Weekly AI Update, a personalized briefing I can run every seven days to stay sharp without the overwhelm. Structure it exactly like this, with these five headings and no others:

THIS WEEK'S HEADLINES THAT MATTER TO ME
Filter the most recent AI news down to the two or three developments that actually affect someone in my role and life. Skip the hype. Tell me why each one matters to me specifically.

ONE THING TO TRY THIS WEEK
Recommend exactly one new tool, feature, or technique matched to my experience level. Give me a three-step try-it-right-now mini tutorial I can complete in under the time I said I have.

MY ROLE-SPECIFIC PROMPT OF THE WEEK
Write one ready-to-use prompt tailored to a real task someone in my profession or personal life actually faces. Make it copy-paste ready, with a brief explanation of why it works.

WHAT TO IGNORE THIS WEEK
Name one or two AI stories getting a lot of attention that I can safely skip given my level and goals, and tell me why they are not worth my time yet.

MY PROGRESS NUDGE
Based on my current experience level, give me one short reflection question or micro-challenge that helps me grow slightly beyond where I am today, without pressure.

Use plain headings. No emoji. Write the way a person would say it out loud.

BEFORE YOU START
Read my profile above. If any bracketed field is empty or vague, ask me about those specific fields in one numbered list and wait for my answer. If the profile is complete, go ahead and produce the briefing.

Why this beats subscribing to more newsletters. A newsletter tells you everything. This tells you the two things that touch your job, one thing to try, and what to ignore. The What to Ignore section is the one people skip and the one that saves the most time.

What to Try This Week

• Take the next deck you actually have to build. Run the Belief Gap prompt on it before you open PowerPoint.
• Answer the six questions in writing, even if you never run the prompt. That alone will change the deck.
• Rewrite every headline in one existing deck as a claim instead of a label. Notice how much shorter the deck wants to be afterward.
• Save the prompt as a Skill in Claude, a Project in ChatGPT, or a saved prompt in Copilot.
• Load one thing you have been meaning to learn into Gemini Notebook and listen to the audio overview on your commute.
• Set up the Weekly AI Update as a scheduled task so it runs without you remembering.
• Time the difference. How long does a persuasive deck take you now? Run it once this way and compare.

Resources & Next Steps

• Link to AI Learning Lab Library

Questions or feedback? cheryl@cherylthompson.net

The Mental Model: Argument Before Artifact

Today's session was built on one idea: making slides and making someone care are two different jobs. AI got very good at the first one. The second one is still yours.

Every tool in this lab can make slides. None of them know your plant, your customer, or the person who already said no in January.

The deck built itself. The argument didn't.

• Run the prompt on one real deck this week.
• Rewrite your headlines as claims.
• Your expertise is the judgment. Check its work.