The exact AI plays
for the job you actually have.

Here's a draft I wrote yesterday: [paste]. Read it once. Then tell me: (1) the weakest paragraph and why, (2) the strongest sentence and why, (3) one concrete rewrite of the weakest paragraph that keeps my voice. Do not be diplomatic.

::what to notice · AI defaults to flattery. Add 'do not be diplomatic' or 'tell me what would embarrass me at a journalism awards night.' Forces real critique.

::trap · Don't accept the rewrite verbatim. Use it as a target to write toward in your voice.

I'm writing about [TOPIC]. Find the three most important primary sources from the last 18 months — academic papers, court filings, regulatory documents, leaked memos — that a serious journalist would cite. For each, give me: source name, date, the one claim that matters, the page or section where it appears, and the URL.

::what to notice · Perplexity grounds against the live web. Claude is better at follow-up synthesis. Use Perplexity for first-pass discovery, then paste the sources into Claude for analysis.

Here's my draft: [paste]. Give me 12 headline options across these registers — 3 plain-spoken, 3 sharp/contrarian, 3 quietly intriguing, 3 SEO-prioritized. For each, also write the 25-word dek. Mark which one you'd ship and why.

::what to notice · Twelve options force range. Three registers force taste. The 'mark which you'd ship' line forces a defensible opinion you can react against.

Here are my last 4 published pieces in this column: [paste all 4]. Audit them for: characters/sources I mentioned that I haven't followed up on, claims I made that need updating, voice drift (am I sounding like myself?), and one unanswered question the reader is probably asking that I haven't addressed.

::what to notice · Long context is Claude's strength. ChatGPT will skim. Use Claude when you need true cross-document analysis.

I cover [BEAT]. Generate 15 story pitch one-liners for editors at [PUBLICATION 1, 2, 3]. For each pitch: the angle, why it matters this week specifically, my unique access or insight, the rough word count and form (feature / Q&A / first-person / investigation). Score each 1-10 on how strong an editor would find it.

::what to notice · The score forces tier-ranking. The 'this week specifically' grounds the pitch in news cycle.

Here are 8 paragraphs of my writing from over the years: [paste]. Reverse-engineer my voice fingerprint. Identify the sentence-length pattern, the rhetorical moves I rely on, the words I overuse, the words I never use, the cadence. Then tell me where in this paragraph from this week [paste] my voice slips and reads like AI.

::what to notice · Doing this once gives you a reusable rubric. Save the output to your skills/ folder.

Here's the file I need to modify: [paste]. Here's the task: [task]. Don't write code yet. Give me 3 implementation approaches. For each: (1) the one-line summary, (2) what it breaks or trades off, (3) lines-of-code estimate, (4) the test you'd write to catch regressions. Then pick one and tell me why.

::what to notice · The 'don't code yet' constraint forces real planning. Most failures come from the AI rushing to code.

Here's my diff: [paste git diff]. Review it like a hostile principal engineer. Look for: (1) edge cases I missed, (2) error paths I didn't handle, (3) data flow that could leak, (4) tests I should have written, (5) where I'm assuming behavior I haven't verified. Be specific. Cite line numbers.

::what to notice · Always run this before push. The 'hostile principal engineer' framing kills the AI's default flattery.

Here's the stack trace: [paste]. Here's the relevant file: [paste]. Here's what the user did to trigger it: [steps]. Don't fix it yet. Walk me through the exact sequence of calls that produced this trace. Identify the most likely root cause. Then list 3 candidate fixes and the trade-offs.

::what to notice · Stack-trace → root-cause → candidate fixes. Skipping straight to 'fix it' gives you the symptom-patch, not the cause.

Here's the function I just wrote: [paste]. Write me the test suite: happy path, the three most likely edge cases, the most likely security edge case, the most likely concurrency edge case. Use [test framework]. Include both the test cases and the assertions.

::what to notice · Always ask for the security edge and the concurrency edge by name. The AI will skip them otherwise.

Here's my codebase structure: [paste tree]. Here are 3 files I touched most this week: [paste files]. Identify 5 refactor opportunities worth doing this sprint — duplication, dead code, abstraction that paid off, abstraction that should be deleted. For each, the rough cost and the rough payoff.

::what to notice · Weekly cadence catches drift before it compounds.

Here's a bug we shipped: [description]. Here's the root cause: [from your diff or your review]. Walk me through: (1) the test we should have had that would have caught it, (2) the lint rule that would have prevented it, (3) the code review checklist item that would have flagged it. Give me a one-paragraph postmortem suitable for a team channel.

::what to notice · Three layers of prevention (test / lint / review) means the bug class doesn't ship twice.

I'm a solo founder. Yesterday I did: [paste yesterday's accomplishments]. My calendar today: [paste]. My inbox subject lines: [paste]. Tell me: (1) what I dropped that compounds, (2) the 60-minute highest-leverage action available right now, (3) what to say NO to today without guilt, (4) one thing to delete from my calendar this week.

::what to notice · The 'NO without guilt' is the killer prompt. Most founders die from too-much, not too-little.

I'm responding to a customer who said: [paste their message]. Context they don't know: [paste]. My voice is direct, warm, no-nonsense. Write the response in 3 versions: (1) the empathetic-first version, (2) the solution-first version, (3) the boundaries version. Mark which you'd send and why.

::what to notice · Three versions = options. One version = the AI's bias dominates. Always ask for three.

I'm sending a monthly update to my advisors. Here's what happened this month: [paste]. Here are my metrics: [paste]. Here's what I'm worried about: [paste]. Draft the update in my voice. Lead with the hardest truth. End with one specific ask. Keep it under 400 words.

::what to notice · 'Lead with the hardest truth' kills the usual update fluff and makes the email worth reading.

Here's my last 7 days of revenue / signups / cancels: [paste CSV]. What's the trend? What's noise? What's the one number I should be watching this week that I'm not? If I had to bet on one action this week to move the trend, what would it be?

::what to notice · Trend / noise / blind spot / bet. Four-part frame for any metric review.

[Perplexity] Find the top 3 product updates / pricing changes / strategic moves from my 4 main competitors [list] in the last 7 days. [Then Claude] Here are the moves: [paste Perplexity output]. Tell me: (1) which moves matter to me, (2) which are theater, (3) what one move I should consider this month to stay ahead.

::what to notice · Perplexity for facts, Claude for synthesis. Two-model workflow.

Here's my current roadmap: [paste]. Here's what I've learned from customers this week: [paste]. Tell me: which items on the roadmap are now wrong? Which order should change? Which item, if I shipped it next, would teach me the most? Defend each call.

::what to notice · Roadmaps die from inertia. Weekly revisit forces honesty.

I teach [subject] to [grade]. Tomorrow's standard: [standard]. Today's class energy was [low / chaotic / focused / curious]. Generate a 45-minute lesson plan: 5-min warm-up, 15-min direct instruction, 20-min guided practice, 5-min exit ticket. Give me 3 different angles for the same standard so I can pick.

::what to notice · The energy variable is the secret ingredient. The same standard needs different framing for a tired Friday vs. a sharp Tuesday.

Here's the concept: [topic]. My class has 3 levels — strong, on-grade, behind. Generate a single worksheet with 12 problems, but mark each problem 1 / 2 / 3 by level. Same skill, different scaffold. Include an answer key with one-sentence explanations for the trickiest problems.

::what to notice · One worksheet, three levels, marked. That's the whole inclusion strategy in one prompt.

I need to email a parent about [situation — DO NOT include student name or identifying info, use 'the student']. Tone: direct but warm. End with a clear next step. Keep it under 200 words. Generate 2 versions: (1) the kid is having a hard time, (2) the parent is going to push back.

::what to notice · Always redact identifying details. Cloud AI is not FERPA-compliant. Use Ollama (local) if the message must include specifics.

I taught [concept] today. Generate 5 exit-ticket questions — 1 easy (rephrasing), 2 on-grade (application), 1 stretch (transfer), 1 misconception probe (a wrong answer that sounds right). Include answer key + what a wrong answer to each question would TELL me about the student's misunderstanding.

::what to notice · The 'misconception probe' is the diagnostic gold. Most exit tickets miss this.

I just finished [unit]. The class data: [average score, range, common mistakes]. The unit standards: [list]. Tell me: (1) which standards landed and which didn't, (2) what I'd change next year, (3) the one re-teach lesson I should slot before moving on, (4) the formative check question that would have caught the misconception earlier.

::what to notice · Retro every unit. The pattern across retros is what becomes your teaching wisdom.

Generate a 60-word weekly newsletter blurb for families. Topic: what we covered this week + one specific question they can ask at the dinner table that surfaces what their kid actually learned. Tone: warm, specific, no jargon.

::what to notice · 'A question to ask at the dinner table' is 10× more useful to families than 'what we covered.' Reframe accordingly.

I'm sending an email about [topic] to [audience description]. The body is: [paste]. Generate 12 subject-line variants across 4 registers: 3 plain-spoken, 3 sharp/contrarian, 3 curiosity-gap, 3 specific-promise. For each, predict the open-rate winner type for this audience and explain your reasoning. Tell me your pick of 3 to ship.

::what to notice · Always ask for predictions + reasoning. You'll start spotting which registers your audience actually rewards.

[Claude] Here's the offer: [paste]. Here's the audience: [paste persona]. Generate 8 ad headline + body-copy pairs across 4 hooks (problem-aware, solution-aware, brand-aware, social-proof). Mark which 3 to test first. [Then Imagen] Generate 4 thumbnail concepts that fit the top headline.

::what to notice · Two-model workflow: language model for copy, image model for thumbnails. Claude doesn't generate images natively; don't waste tokens trying.

Here's last week's funnel: visits → signups → trial → paid. [paste numbers]. Where did we leak? Was the leak above or below the previous baseline? What's the one experiment that would move the worst stage by 10%? What's the read of the leak data IF we're in a noisy week with a holiday or news cycle?

::what to notice · Always include the 'noise check.' A bad funnel week with a holiday in it is signal-poor.

Here are 5 of our most successful posts: [paste]. Here's a draft I'm about to ship: [paste]. Does the draft sound like the same brand? Where does it drift? Give me a tighter rewrite in the brand voice.

::what to notice · Run this weekly across all channels (email, LinkedIn, X). Voice drift is the silent killer of brand equity.

What are the top 5 trending topics in [my industry / my target persona's world] this week? For each, give me: the volume signal (search trend, social volume, press mentions), the angle that would let our brand voice show up, the risk of jumping on it, the timing window.

::what to notice · Risk + timing window are the parts most marketers skip. They're where trend-jacking goes wrong.

[Perplexity] Pull current pricing for [competitors]. [Claude] Compare our pricing to theirs in a 3-column table. Flag: where we're more expensive (and is the value gap real?), where we're cheaper (and is that signaling cheap-not-better?), where we have unique pricing structure that's confusing buyers.

::what to notice · Update the table monthly. Pricing pages move quietly and the buyer always notices before the marketer does.

I'm drafting [document type] for a matter involving [redacted facts]. Don't draft yet. List the 5-7 issues a senior partner would expect this document to address. For each: (1) the issue in one sentence, (2) the most common drafting mistake at this issue, (3) the clause language pattern that protects against the mistake.

::what to notice · Never let the AI draft first. Issue-spot first, then draft. Otherwise you're correcting AI's instinct for the average matter.

Here's the agreement: [paste]. Read it once. Then identify: (1) the 5 most one-sided terms in favor of [counterparty], (2) the 3 ambiguities a litigator could exploit, (3) the standard clauses I'd expect to see that are missing, (4) any clause that conflicts with [jurisdiction] statutory limits.

::what to notice · Long context is Claude's killer feature for legal work. Don't paste 200-page agreements into ChatGPT.

[Claude] I'm researching [issue] in [jurisdiction]. What are the 5 most-likely-cited cases a practitioner would surface? For each, give me: (1) the case name and citation, (2) the holding in one sentence, (3) why it matters here. — [Then verify EVERY citation in Westlaw/Lexis before relying on it.]

::what to notice · Claude will hallucinate citations confidently. Always verify. Treat the AI output as a research-direction memo, not a research product.

I need to explain [legal concept] to a client who isn't legally trained. Their context: [redacted summary]. Generate the email in 3 versions: (1) the plain-language version, (2) the analogous-to-everyday-life version, (3) the strict-formal version. Mark which I should send to a sophisticated business client vs. a first-time founder.

::what to notice · The 'analogous' version is what wins long-term client trust. Use it freely.

Here are my matters this week: [redacted list with time]. Tell me: (1) which matter ate more hours than the typical, (2) where I should have delegated and didn't, (3) the matter type that's becoming my unprofitable cohort, (4) the matter type I should turn down going forward to protect my margin.

::what to notice · Most solos die from underpriced matter cohorts. Weekly retro catches the pattern before the quarter ends.

What statutes, regulations, or significant case decisions in [jurisdiction] affecting [practice area] have been published or decided in the last 14 days? For each, the 2-sentence summary and the link to the official source.

::what to notice · Stay current. Pair with a manual scan of bar-association alerts; AI signal is the second source, not the first.

Here are the last 6 months of comps in [neighborhood]: [paste from MLS]. Subject property: [details]. Tell me: (1) the right list price range and why, (2) the most comparable 3 sold comps and what makes them comparable, (3) the days-on-market reality check, (4) the one comp that would be misleading to cite and why.

::what to notice · The 'misleading comp' line forces honesty. A bad comp baked into a listing presentation will hurt you all year.

Here's the property: [bullet-point facts, room sizes, key features]. Generate 3 listing description versions: (1) the warm-buyer-imagine-themselves-here version, (2) the architecture-detail-rich version, (3) the lifestyle-positioning version. Each under 250 words. Mark which one fits this property best.

::what to notice · Three versions = options. The one that 'feels right' usually is. Trust your gut over the AI's pick.

Buyer: [persona]. Last interaction: [paste notes]. Generate a 4-touch follow-up sequence: (1) immediate same-day note, (2) 3-day check-in, (3) 7-day market update, (4) 14-day re-engagement. Each in my voice (warm, direct, no spam-feel). Each under 60 words.

::what to notice · Sequence > one-shot. Use this template and you'll never forget a buyer.

Here's the situation: subject property listed at [X], offer at [Y], buyer wants [terms], seller wants [terms]. List the BATNA for each side, the 3 most likely counter scenarios, the move that would unstick the deal, and the one mistake an inexperienced agent would make here.

::what to notice · BATNA framing is the difference between negotiating and reacting.

[Perplexity] What are the top 3 macro signals affecting [my market] this week — interest rates, jobs data, local employer news, school news, infrastructure? [Then Claude] Translate that into a 200-word client email that lands as informed without being preachy.

::what to notice · Clients don't want a market report. They want one operator's read on what to do next.

Here's my pipeline: [paste — stages, dollar values, last touch]. Tell me: (1) the deal most likely to close this week, (2) the deal most likely to stall if I don't touch it, (3) the deal I should drop because it's eating my time, (4) the next 3 touches I should make today.

::what to notice · Most agents die from too-many-stagnant-deals. Weekly drop-list is the discipline.

I'm about to read [chapter / paper / lecture notes — paste the table of contents or the abstract]. Generate 5 specific questions a strong reader would have answers to after finishing. Include 1 question that should change my mind if the chapter is doing its job.

::what to notice · The 'change my mind' question is the test of whether you actually engaged or just skimmed.

I'm studying [concept]. Explain it: (1) like I'm 12 — analogy + everyday example, (2) like I'm a grad student — formal definition + edge case, (3) like a textbook problem — solve a worked example. Then quiz me with 3 questions of escalating difficulty.

::what to notice · Three levels of explanation surface where YOUR understanding actually breaks down. The level you can't follow is your study target.

I'm writing an essay on [thesis]. Generate an outline: introduction with the strongest version of my argument, 4 body paragraphs each with a topic sentence + the strongest evidence + the most-likely counter, conclusion. For each paragraph, suggest 2 sources I should verify exist in [your library / Google Scholar / JSTOR].

::what to notice · Outline only. Don't have the AI write the body. The thinking happens in the writing.

I have an exam on [topics]. Generate 10 practice questions across difficulty levels: 3 recall, 4 application, 2 synthesis, 1 stretch. Give me the answer key but withhold it until I commit my answers. I'll paste my answers back and you'll grade.

::what to notice · Withholding the answer key is the discipline. Otherwise the AI just shows you the answers and you 'learn' nothing.

Here are the notes from this week's classes: [paste]. What concept showed up across multiple classes? Where did my notes get thin? What's the one concept I should re-study before next week's lectures build on it?

::what to notice · Compound concepts are where grades quietly slip. Weekly retro catches them.

I learned these skills this week: [paste]. Generate one cross-cutting practice problem that uses all of them at once. Mark which step uses which skill so I can self-check.

::what to notice · Integration > isolation. One mixed problem teaches more than five single-skill problems.

My [age]-year-old asked: '[paste their question verbatim].' Explain it: (1) at their age level using everyday language, (2) one analogy that helps it stick, (3) one related question they might ask next so I can stay ahead of the curiosity wave.

::what to notice · The 'related question' is the magic. Kids' curiosity is a chain. AI tells you what link comes next.

I have these ingredients: [paste]. My kid will eat [list]. My kid won't eat [list]. I have [time] to cook. Give me 3 dinner ideas tonight + tomorrow's school lunch. Each under 30 minutes. Include any pantry swaps that don't break the dish.

::what to notice · The 'won't eat' list saves more time than the 'will eat' list. Lead with constraints.

My [age]-year-old is having trouble [getting to sleep / processing a hard day / settling after a tantrum]. Write me a 3-minute script — warm, present, no screens, no bribes — that I can read or improvise from. Generate 3 options across tones: cozy / playful / mindful.

::what to notice · Three tones lets you match the kid AND your own energy that night. Pick whichever you have the bandwidth for.

My kid is going to ask me about [topic — death, divorce, body, identity, race, money, scary news]. I want to be ready. Give me: (1) 3 age-appropriate framings I can use, (2) the 2 mistakes parents commonly make on this topic, (3) the question they might ask as a follow-up so I'm not surprised.

::what to notice · Practice the conversation before you have it. AI lets you rehearse without burning the kid's real moment.

We have a weekly family meeting. This week's context: [what happened, what's coming, any conflict]. Generate a 20-minute agenda: warm-up (one thing each loved this week), main topic, kid agenda items, one decision to make as a family, closing.

::what to notice · The 'kid agenda items' slot trains kids that they have legitimate voice in family decisions. It's a long game.

My [age]-year-old is into [current obsession]. Generate 7 days of 30-min activities that ride the obsession into related skills (motor / literacy / numeracy / curiosity). Each day: the activity, the skill, the prep (5 min or less), the question to ask after.

::what to notice · Ride the obsession. Trying to redirect curiosity kills it. AI is great at mapping their interest into adjacent skills.

Here's the brief: [paste]. Don't generate visuals yet. Give me 3 design strategies: (1) the safe-credible direction, (2) the bold-distinctive direction, (3) the contrarian-defensible direction. For each: the one-sentence summary, what it solves, what it does not solve, the risk.

::what to notice · Three strategies = three real options, not three flavors of the same thing. The 'what it does not solve' line forces honest tradeoff.

Here's the surface: [screenshot or description]. The action: [what we want the user to do]. Generate 5 microcopy variants for the CTA + 5 for the supporting line. Across registers: minimal, helpful, confident, playful, urgent. Tell me which pair you'd ship and why.

::what to notice · Always ask for register variants, not just synonyms. Register is what most microcopy drafts miss.

[detailed prompt for the visual exploration tool] · style: [reference], composition: [details], lighting: [details], aspect ratio: [ratio]. Generate 8 variations. Mark which 3 you'd refine.

::what to notice · Image AI is variant-fast, taste-blind. Use it to explore, not to commit. The composition you'd never have drawn yourself is sometimes the answer; sometimes it's the trap.

Here's the design: [paste screenshot]. Critique it against Nielsen's 10 heuristics + the 3 most-likely accessibility issues. Be specific. Cite the element you're talking about. Skip the diplomatic version.

::what to notice · Vision-capable Claude is the right tool. Don't try this in a text-only chat.

I shipped these projects this week: [paste]. Write the portfolio case-study draft for the strongest one: the problem, the constraint, my approach, the design decision that matters most, the outcome. Voice: confident, specific, no jargon. Under 400 words.

::what to notice · Weekly case-study updates compound. By year-end you have a portfolio that walks-the-talk.

Here are 5 screens we shipped this week: [paste]. Identify: (1) where we drifted from the design system, (2) where the system is being violated because it's wrong (and should be updated), (3) where we made one-off choices that should become a system token.

::what to notice · Drift is normal. Catch it weekly and the system stays alive instead of becoming a museum.

Here are 5 paper abstracts: [paste]. My research question: [your question]. Rank them: which to read first, which to skim, which to skip and why. For each 'read first,' the one question I should be ready to answer after reading.

::what to notice · Daily triage saves a week of reading the wrong papers.

Here's the methods section: [paste]. Critique it: (1) the 3 biggest threats to internal validity, (2) the assumption that should be defended but isn't, (3) the missing control I'd flag as a reviewer, (4) one alternative analysis that would strengthen or weaken the claim.

::what to notice · The 'alternative analysis' line catches the 'this could go either way' papers. Most methods sections only defend one analysis.

Find the most-cited papers on [topic] from the last 36 months. For each: the citation, the central claim, the methodological approach, the most-cited critique. Identify the 3 most likely open questions a new PhD project could legitimately attack.

::what to notice · Pair with manual database scan; AI signal is suggestive, not definitive.

Here's a finding from my work: [technical statement]. Write 3 versions: (1) for an undergraduate intro course, (2) for a journalist with a science beat, (3) for the public Twitter audience. Each accurate, each calibrated to the audience.

::what to notice · Three audiences = three drafts you can use across the year (lectures, press, social).

Here are my notes + relevant papers: [paste]. My thesis: [statement]. Generate a section outline: claim · evidence · counter-evidence · disposition. For each paragraph, the 2 most-likely citations I should verify.

::what to notice · Outline only. Never let AI write the body. The thinking happens in your writing.

Here's the call for proposals: [paste]. Here's my research direction: [statement]. Draft the 1-page summary in 3 versions: (1) leading with novelty, (2) leading with feasibility, (3) leading with impact. Tell me which fits this funder.

::what to notice · Funder fit > intrinsic quality at the proposal stage. Different funders reward different leads.

Here's my last month of notes on [report] (redacted): [paste]. Their recent work: [summary]. Tell me: (1) the pattern I haven't called out yet, (2) the conversation I've been avoiding and why I should have it, (3) one specific thing to bring up today, (4) one thing to ask them about their day-to-day that I haven't asked.

::what to notice · The 'avoiding' question is where good management happens. AI is excellent at surfacing the obvious thing you keep ducking.

I need to make a decision on [topic]. Here are the options: [paste]. Generate a decision memo: the call, the reasoning, the dissenting view, the metric we'd measure to know we made the right call, the rollback condition. Under 300 words.

::what to notice · The 'rollback condition' line is what turns a decision into a learning loop.

Here's what my team did this week: [paste]. The metric movement: [paste]. The blockers: [paste]. Draft the 4-bullet update to [exec]: progress / metrics / risk / ask. Tone: confident, honest about what isn't working, specific about what would unstick it.

::what to notice · Status updates that lead with the ask get attention. Status updates that bury it die in the doc.

Here's the sprint data: [tickets shipped, slipped, added mid-sprint]. The team's notes: [paste]. Generate the retro discussion prompts: (1) what worked that we should keep, (2) what didn't that we should stop, (3) the one process change to try next sprint, (4) the underlying pattern across sprints that we haven't named yet.

::what to notice · The 'underlying pattern across sprints' is where retros become useful. Otherwise they're sprint-local.

Here's my team's signals this week: 1:1 notes (redacted), velocity, on-call load, PTO requests, the chat tone. Tell me: what's the health pattern, where am I asking too much, who needs a check-in I haven't scheduled, what's the team-level conversation I should have at next standup.

::what to notice · Weekly pulse beats annual engagement survey. Patterns surface in the daily signal, not the once-a-year question.

Here's the team structure: [paste]. The current set of in-flight projects: [paste]. The 6-month strategy: [paste]. Where does the structure fit the strategy? Where is it misaligned? What's one team-design move I should consider this quarter?

::what to notice · Org design drifts behind strategy. Quarterly check keeps the structure honest.

[Perplexity] Pull the last 6 months of news, funding, leadership moves, product launches, and competitor pressure for [company]. [Claude] Synthesize: their likely top 3 priorities this quarter, the 2 most-likely buying triggers in our space, the 1 question that would land in a first call.

::what to notice · Two-model workflow. Perplexity for facts, Claude for synthesis.

I'm reaching out to [name, title, company]. Their recent signal: [paste — post, news, hiring page]. Our offer: [paste in 2 sentences]. Draft 3 outreach versions: (1) the warm relevance-first, (2) the hard pain-first, (3) the contrarian provocation. Under 80 words each. Mark which fits.

::what to notice · Three registers = three real options. The buyer's signal determines which one.

Here are my call notes (redacted client names): [paste]. Generate: (1) the recap email to the prospect, leading with what they care about, (2) the internal-CRM note on stage, blockers, next step, (3) the proposed next-step time-window with the specific value they'd get from accepting.

::what to notice · Always send the recap within 60 min. The 'value of accepting' line is the close before the close.

I'm about to propose [solution] to [persona]. Generate the 5 most-likely objections, ranked by frequency. For each: the surface objection, the underlying concern, the question that surfaces what's really blocking, the 60-word response.

::what to notice · Underlying concern > surface objection. Always. AI is great at the translation.

Here's my pipeline: [paste — stages, ARR, last touch, decision criteria]. Tell me: (1) the deal most likely to close this week, (2) the deal that needs an executive escalation, (3) the deal stuck because of [specific friction], (4) the deal I should drop because it's eating my time.

::what to notice · Weekly drop-list is the discipline. Most reps die from too-many-stagnant deals.

Here's a deal we won this month: [details]. Here's one we lost: [details]. Identify: the pattern in why we won, the pattern in why we lost, the question we asked in the won deal that we didn't in the lost, the disco step we'd add to the playbook.

::what to notice · Win/loss every month catches the pattern in real time. Quarterly is too late.

I'm publishing a video / post about [topic]. The 30-second hook: [paste script]. Generate 12 title variants across 4 hooks: curiosity-gap, specific-promise, contrarian, story-led. For each, predict the click-rate winner type for my audience and reasoning. Tell me your top 3 to test.

::what to notice · Title is 80% of the click. A/B/C/D before you publish, not after.

Here's my first 30 seconds: [paste script]. Critique like a feed scroller: where does this lose me, what would make me stop, what's the line that should be the FIRST line, what's a hook variation that would test against this.

::what to notice · Feed-scroller framing kills creator-friendly diplomacy. You get the brutal truth.

Here's the transcript: [paste]. Generate: (1) the 200-word description optimized for discovery, (2) the chapter timestamps, (3) the 5 most-quotable lines for social cuts, (4) the 3 follow-up questions for community engagement.

::what to notice · The 'quotable lines for social' is where one upload becomes a week of distribution.

Here's the transcript: [paste]. Generate 5 short-form cuts (60s each) optimized for [platform]. For each: the in-point, the out-point, the title for the short, the caption text. Lead with the hook.

::what to notice · One long-form upload = five short-forms = a week of distribution. AI does the cutting work in minutes.

Here are my last 5 uploads: [titles, views, watch-time, click-through rate]. Tell me: (1) the title pattern that worked, (2) the title pattern that didn't, (3) the topic the audience over-indexed on, (4) the experiment to run next.

::what to notice · Weekly retro on 5-upload windows catches signal before the trend dies.

[Perplexity] What's trending in my niche [topic] this week — search trends, news, community questions? [Claude] Generate 7 video / post ideas across the next 14 days, each tied to one trend signal. For each: the title, the hook, the format, the call-to-action.

::what to notice · Always pair AI-generated calendar ideas with at least one off-trend original idea. Pure trend-chasing dies fast.

I have a call with [client]. The agenda: [paste]. The last 3 emails: [paste]. My objective today: [paste]. Tell me: (1) the unspoken question they're trying to ask, (2) the move that would unstick the engagement, (3) the one question I should ask to surface the real blocker, (4) the talking-point I should NOT lead with.

::what to notice · The 'should NOT lead with' is the killer. AI is good at flagging the move that would derail the conversation.

Here are 5 interview transcripts: [paste]. The research question: [paste]. Synthesize: (1) the 5 themes that recur, (2) the 2 dissenting views worth weighting, (3) the surprise insight nobody articulated cleanly but is implied across 3+ interviews, (4) the open question that needs more discovery.

::what to notice · Long context is Claude's edge for qualitative work. Don't try this in ChatGPT.

Here's the recommendation: [statement]. Outline a 12-slide deck: (1) the problem we're solving, (2) the data, (3) the options considered, (4) the recommendation, (5) the implications, (6) the risks, (7) the timeline, (8) the resourcing, (9) the metric we'd measure, (10) the next-step decision needed, (11) the appendix preview, (12) the close. For each slide, the headline + the 1 chart or 1 bullet that matters.

::what to notice · Outline only. Build the slides. AI-built slides look like AI-built slides.

I need to tell [client] [hard truth]. Their context: [paste]. My read of their likely reaction: [your prediction]. Generate 3 versions of the conversation: (1) lead with care, (2) lead with data, (3) lead with the question. For each, the 30-second opener, the likely counter, my response.

::what to notice · Rehearsal is the deliverable. AI lets you practice the conversation before you have it.

Here's the engagement: [scope, timeline, current state, client signals, fees]. Tell me: (1) the client's confidence pattern this week, (2) where we're off-scope and need to renegotiate, (3) where the client is silently disagreeing and won't say so, (4) the move to keep this engagement healthy through the next milestone.

::what to notice · Silent disagreement is what kills consulting engagements. Weekly check catches it.

[Perplexity] What are the top 3 signals in [my industry] this week that would create new consulting demand? [Claude] Translate into 3 outreach openers I could use this week with target clients who'd be affected.

::what to notice · Practice growth comes from signal-jacking — being the consultant who already has the answer when the question lands in the executive's inbox.

Write a patient handout on [condition / medication / procedure] at a 6th-grade reading level. Include: (1) what it is in everyday language, (2) what to expect, (3) the 3 things they should call us about, (4) when to go to the ER. Under 250 words. No medical jargon.

::what to notice · Verify every claim against your facility's reference. AI is the draft; clinical sign-off is the deliverable.

Here's the patient summary (de-identified): age range, diagnoses, current concerns, vitals trends, interventions, response, what I'm worried about, recommended next steps. Tighten this into a 60-second SBAR. Make sure recommended next step is specific and actionable.

::what to notice · Always de-identify before pasting. Better: use Ollama for this entirely.

I have an interdisciplinary care meeting on [topic — de-identified]. The team members: [roles]. Generate: (1) my 90-second update, (2) the question I should bring to the team, (3) the concern I haven't been able to articulate cleanly, (4) the family perspective I should remember to surface.

::what to notice · Family perspective is what most clinical conferences forget. AI is good at flagging the missing voice.

Write a plain-language explanation of [medication class] for a patient: what it does, the 3 most common side effects, the 2 dangerous interactions to watch for, when to call us. Under 200 words. 6th-grade reading level.

::what to notice · Always verify against Lexicomp or UpToDate. AI is the draft. The clinician is the source.

What new guidelines, FDA actions, or significant studies in [my specialty] have been published in the last 14 days? For each: the source, the 2-sentence summary, the practice implication, the official link to verify.

::what to notice · Always verify against official source. AI is the alert, not the truth.

Here are 5 of my recent notes (de-identified): [paste]. Audit: (1) where I'm being vague that an auditor would flag, (2) where I'm being too detailed and exposing the chart to risk, (3) the documentation patterns that don't reflect the actual care I delivered, (4) the one habit-change that would tighten everything.

::what to notice · Documentation drift is silent. Weekly scan catches it before an audit does.

Explain today's biggest news story in plain language — no jargon, no political spin, no acronyms unexplained. Three short paragraphs: what happened, why it matters, what to watch next.

::what to notice · Run this daily and you'll be better informed than half the people you went to high school with.

Explain '[word or phrase you heard]' in plain language. Use an example I can picture. Then tell me where it originally came from and how it's changed.

::what to notice · Bonus: you'll remember the word. AI gives you context, not just a definition.

I have these ingredients on hand: [paste]. I'd like to make something like [cuisine / dish]. Give me 3 recipe options with substitutions noted, prep time, and which one is the easiest. Imperial measurements.

::what to notice · AI is great at 'what can I make with what's in the fridge.' Use this daily and stop buying takeout.

I want to write to [grandchild / friend / sibling] about [topic]. They're [age / context]. Keep it warm, specific to them, under 200 words. End with a question they'd actually answer.

::what to notice · The 'end with a question' is the secret to keeping the correspondence alive.

I'm working on [woodworking project / quilt / garden plan / coin collection / family history / memoir]. Here's where I'm stuck: [paste]. Walk me through 3 ways to think about this, then suggest the one you'd try first.

::what to notice · AI is your hobby tutor. Free. Available at midnight. Doesn't sigh.

I have an appointment with my [doctor / specialist] next week. Here's what I want to ask about: [paste]. Generate the 5 questions I should bring — calibrated, specific, the questions a strong patient asks. Don't give me medical advice; give me the questions.

::what to notice · AI is the patient-advocate friend. NOT the doctor. Always bring the questions to the human professional.

A customer left this review: [paste]. Their rating: [stars]. Draft 3 responses: (1) the warm-thanks-and-acknowledge version, (2) the address-the-specific-complaint version, (3) the take-it-offline-with-care version. Each in my voice (warm, direct, local-business, no corporate-speak). Under 80 words.

::what to notice · Three versions = three options for tone. The one you ship depends on the review. AI is fast at all three.

I run a [business type] in [city]. Today's hook: [season / event / news / promotion]. Generate 3 social post variants for [platform]: (1) helpful-tip version, (2) behind-the-scenes version, (3) offer-driven version. Each under 150 words. Include a clear call-to-action.

::what to notice · Three a week is plenty for a local business. AI lets you batch a week of posts in 30 minutes.

I'm sending an email to my customer list this week. Topic: [paste]. My voice: [direct / warm / no-fluff]. Draft the email: subject line, one-paragraph hook, the offer or content, the call-to-action. Under 200 words.

::what to notice · Weekly cadence beats monthly. AI removes the 'I don't have time' excuse.

Here's next week's demand forecast: [paste]. Here are my employees with availability: [paste]. Build the schedule that minimizes overtime, hits coverage at peak hours, gives [person] their requested day off, and leaves me time for [tasks].

::what to notice · AI is fast at schedule puzzles. Faster than scratching it out at midnight.

Here's last week's numbers: revenue, costs, foot traffic, online orders, refunds, complaints, employee hours. Tell me: (1) the trend that's quietly compounding, (2) the cost line that's growing faster than revenue, (3) the one experiment that could move the needle this week, (4) the metric I should be tracking that I'm not.

::what to notice · Weekly review beats monthly. Compounding problems show up in weekly data first.

What's happening in [city / neighborhood] this week that affects a [business type]: events, weather, news, competitor moves, school calendar, local employer news? For each: the 1-sentence summary and the action I should take this week.

::what to notice · Local signal > national signal for a small business. Perplexity grounds against the live web.