Jeff Bullas

Good point: I like the reminder that short, focused practice beats long, aimless sessions — AI is a tireless, non-judgmental partner. Let me add a practical, ready-to-use routine you can start today.

Why this works

Small, repeatable cycles let your mouth learn new habits. Focus on one thing at a time (a sound, word ending, or rhythm) and use the AI to point out patterns — then repeat the exact corrective drill until it feels natural.

What you’ll need

• A smartphone or computer with a microphone (built-in is fine).
• A quiet corner and 10–15 minutes a day.
• An AI tool that accepts voice recordings or transcribes speech and can return feedback (speech-recognition or language app).
• A short list of 5–10 sentences or tricky words.

Quick do / do-not checklist

• Do keep sessions to 10 minutes and focus on one target.
• Do save your recordings so you can compare over time.
• Do use slow and natural-speed models to shadow.
• Do-not try to fix every sound at once.
• Do-not practice in a noisy room — the AI needs clean audio.

Step-by-step (10–15 minutes)

1. Choose one goal (example: final consonant /t/ clarity in English).
2. Record one sentence from your list that uses that goal twice.
3. Ask the AI to analyze and name the top 2 issues and give one short drill per issue.
4. Do the drills: repeat 5–10 times, first slow, then at normal speed shadowing the model.
5. Record the sentence again and compare. Note one measurable change (e.g., clearer /t/ sound at the end).
6. Use the sentence in a short imagined reply to practice real use.

Copy-paste AI prompt (use this exactly)

“Listen to my recording and identify the top two pronunciation issues. For each issue give one short drill I can repeat 5 times. Then provide a slow version and a natural-speed version of the sentence for me to shadow. Tell me one simple tip to check progress.”

Worked example

Sentence: “I packed the last boat at eight.”

• Expected AI feedback: missing /t/ at the end, reduced vowel in “last”.
• Drill for /t/: say /t/ alone 10 times, then say “last—t” (hold the t for 0.5s) 5 times, then say the sentence slowly 3 times.
• Drill for vowel: exaggerate the vowel in “last” (open jaw) 5 times, then normal pace 5 times.
• Record again and listen for a clearer /t/ release and a slightly fuller vowel in “last.”

Common mistakes & fixes

• Noisy recording — move to a quieter spot or use simple earbud mic.
• Trying to fix too many things — pick just one target per session.
• Not saving progress — keep a dated log (one clip per day).

7-day micro action plan

1. Days 1–2: focus on one consonant sound with the routine above.
2. Days 3–4: switch to a rhythm/intonation goal.
3. Days 5–6: role-play the sentence in short dialogues with the AI.
4. Day 7: listen to week’s earliest and latest recording and note one clear improvement.

Small, consistent steps win. Start with 10 minutes today — record, ask the AI, do the drills, and save that second clip. Celebrate the tiny improvements; they add up fast.

Quick win (try this in 5 minutes): Paste one interview chunk (200–400 words) into your AI and run the prompt below. You’ll get 3 pain points, 3 direct quotes, and a 2‑sentence TL;DR — enough to start a strong case‑study opening.

Nice point — structure is the lever. I like the KPI focus you added. Use AI to surface facts fast, then you (the human) verify and shape the story. Here’s a compact, practical add‑on to make that repeatable.

What you’ll need

• Raw artifacts: interviews, usability notes, screenshots, metrics (labeled).
• An AI editor (GPT-4 style) and any text editor.
• A 30–90 minute block per case study for 2–3 iterations.

Step-by-step (do-first mindset)

1. Chunk & label: Break notes into 200–500 word chunks. Give clear labels: Interview A, Usability B, Metrics 2024-06.
2. Run the quick prompt: Get pain points, quotes, TL;DR (use the prompt below).
3. Map to headings: Problem, Role, Research, Decisions, Outcome, Lessons.
4. Draft section-by-section: Prompt AI for one section at a time (60–150 words each). Keep outcome first in each section headline.
5. Visual brief: Ask AI for 3 visuals + captions + alt text (wireframe, flow, before/after screenshot callouts).
6. Verify & quantify: Check each quote and metric against originals. Replace uncertain items with [UNVERIFIED] or exact numbers.
7. Polish & export: Edit voice, shorten sentences, add TL;DR and a 1‑line impact highlight. Export PDF or portfolio page.

Copy-paste prompt (use as your quick test)

“You are an expert UX case study writer. I will paste an interview or research chunk. Extract: 1) the top 3 user pain points, 2) three direct user quotes (verbatim), and 3) a 2-sentence TL;DR that highlights the main outcome. If any quote or metric is unclear, mark it with [UNVERIFIED]. Keep tone professional and concise.”

Example (before → after)

Before: messy notes saying “onboarding is confusing, app slow, users drop off.” After: TL;DR, 2 quotes, 3 pain points, one design change (progressive onboarding), and an outcome line: “Activation ↑ 38% in 8 weeks (A/B test).”

Common mistakes & fixes

• Using paraphrased quotes: Fix — keep verbatim quotes and tag edits as [EDITED].
• Too much process: Fix — lead with impact; put methods in a short appendix.
• Not verifying metrics: Fix — always cross-check numbers and dates before publishing.

3-step action plan (today)

1. Gather and label one interview chunk (10 min).
2. Run the quick prompt above and capture TL;DR + quotes (5–10 min).
3. Draft Problem + Outcome sections in the AI, verify facts, and save a one-page case study draft (30–60 min).

A small, verified iteration beats one perfect draft. Start quick, verify often, then polish for impact.

Nice quick win — mirroring the client’s first line is the fastest trust-builder. Your three-signal rule (understand problem, show a result, ask for a small next step) is the practical filter every freelancer needs.

Here’s a compact, do-first method to turn that quick win into consistent wins on Upwork or Fiverr.

What you’ll need:

• One live job post open
• One concrete past result (metric or short case line)
• AI tool (ChatGPT or similar) and 60–90 seconds for human tweak
• Simple tracking sheet (date, job, template, reply/interview/hire)

Step-by-step (do this now):

1. Read the posting and write the client’s exact phrase for the main pain (2–6 words).
2. Write a single-sentence hook using their words + main benefit (10–15 seconds).
3. Run the AI prompt below to generate a 150–200 word draft.
4. Edit for 60–90 seconds: insert your hook, add one metric from your past work, tighten the CTA to a 10–15 minute call.
5. Send and log results. Repeat — after 10 sends, compare templates and opening lines.

Copy-paste AI prompt (use as-is):

“Write a concise freelance proposal (150–200 words) for this job: [paste job description]. Start with a one-line hook that uses the client’s exact language and states the main benefit. Include: 1) a three-step plan (three short bullets or sentences), 2) one measurable outcome based on similar projects (framed as an aim, not a promise), 3) one line of social proof (past result), and 4) a single call to action asking for a 10–15 minute call. Keep tone professional, confident, friendly, and clear.”

Example hook + short proposal (copyable):

Hook: “You need a landing page that converts visitors into leads — I’ll help increase lead rate by ~30% in 60 days based on similar builds.”

Short proposal (150 words): I read your brief and can deliver a high-converting landing page focused on the offer you described. Plan: 1) Audit your current page and analytics, 2) Build a focused layout + persuasive copy, 3) Run A/B test on headline and CTA for two weeks. Typical outcome: similar clients saw ~30% lift in leads within 60 days. Social proof: helped an e-commerce client increase signups from 2% to 5% in 8 weeks. Next step: can we do a 10–15 minute call tomorrow to align on goals and timeline?

Common mistakes & fixes:

• Generic hook: Mirror the client’s language and state the benefit.
• No metric: Add a conservative, real metric from your work.
• Too long: Trim to 150–200 words — people skim.

7-day action plan:

1. Day 1: Gather 10 job posts + your top 3 case metrics.
2. Day 2: Create 3 template prompts and test the AI output.
3. Days 3–6: Send 2 proposals/day; log outcomes.
4. Day 7: Review replies — keep the best hook and metric; iterate.

Takeaway: Use the client’s words, show one real metric, ask for a tiny next step. AI speeds drafting; your quick human tweak wins trust. Start with one post now and send your first improved proposal in under 5 minutes.

Spot on about displacement maps — that’s the realism switch. Let’s add one pro move: a reusable “two‑pass realism stack” you can drop onto paper, plastic, metal, or fabric and get consistent, photoreal results in minutes.

Why this helps: Instead of rethinking every mockup, you build one template that handles curvature, shine, texture, and label thickness. Update one Smart Object, and the whole scene updates. That’s speed and quality on repeat.

What you’ll need

• Base photo or AI-generated substrate (paper/plastic/metal/fabric) at 2–4K resolution.
• Image editor with layers, masks, and displacement (Photoshop, Photopea, or GIMP).
• Optional inpainting-capable AI for subtle wear/reflections.
• Your artwork as a transparent PNG or layered file.

The two‑pass realism stack (build once, reuse forever)

1. Base texture: Place your clean substrate image. Duplicate it and keep a hidden “clean” backup.
2. Displacement map: Duplicate the base → desaturate → increase contrast so folds/shine pop → blur 2–4 px → save as a grayscale file. This is the map you’ll use repeatedly.
3. Artwork Smart Object: Place your label/art as a Smart Object (or editable layer). Transform to match perspective.
4. Pass 1 — Form: Apply the displacement to the artwork (start 5–15 px on 2–4K images). This sells the geometry (curves, folds, dents).
5. Pass 2 — Surface: Duplicate the artwork layer. Set one to Multiply (ink/dye) and the other to Overlay or Soft Light (shine/interaction). Mask the highlight areas on the Multiply layer so real speculars stay bright.
6. Edge thickness: Add a 1–2 px inner shadow or soft inner glow to the artwork group. Very low opacity (10–20%). It simulates label thickness or ink edge on paper.
7. Shadow catcher: From the base texture, duplicate → high-pass or levels to isolate shadows → set to Multiply above the artwork. This preserves tiny original shadows crossing the label.
8. Micro-variation (material-specific):
   • Paper: Add a subtle warm tint layer (2–5% opacity) to avoid cold, sterile whites.
   • Plastic: Paint a soft white highlight streak on a separate layer, set to Screen 20–40% and mask where the label sits.
   • Metal: Create a brushed look on the artwork duplicate: add tiny noise + motion blur 5–15 px in metal direction, set to Overlay 10–20% (clip to artwork).
   • Fabric: Duplicate the grayscale displacement, boost contrast, then use it as a layer mask on the artwork so threads intermittently “break” the ink. Add a 0.5–1 px blur to an artwork copy on Multiply 20–30% to mimic dye bleed.
9. Final polish: Tiny dodge on existing highlights, unify contrast with a very low-opacity curves layer, and convert a copy to CMYK if printing.

Material-by-material quick settings

• Paper: Multiply 70–90%; Overlay 10–25%; displacement 5–10 px.
• Plastic: Multiply 40–60%; Soft Light 20–40%; add Screen highlight streak; displacement 8–15 px.
• Metal: Overlay 30–50%; keep a Screen specular layer 20–40%; displacement 5–10 px.
• Fabric: Multiply 60–80%; dye bleed layer 20–30%; stronger mask from weave; displacement 10–18 px.

Insider trick: AI-assisted label curvature

1. Roughly transform your label to the surface.
2. Mask the label area and inpaint the edges with: “match the object’s curvature, keep geometry, add subtle edge lift.”
3. Use the inpainted edge as a guide; fine-tune with Warp for a perfect wrap.

Copy-paste AI prompts

Base substrate (swap [material]): “Photorealistic 3/4 view of a clean [material] surface with soft side lighting and visible fine texture; neutral studio background; allow a blank area for a label; no logos; high-resolution; natural shadows and highlights preserved.”

Edge realism inpaint: “Inpaint only the label area edges to add microscopic edge lift, faint scuffs, and preserve original specular highlights; do not change the object’s shape or perspective.”

Mini worked example — fabric pouch label (15–25 min)

1. Generate a cotton pouch base with soft folds. Save at 3000 px wide.
2. Create the displacement map (desaturate, contrast, blur 3 px).
3. Place your label Smart Object; transform to pouch angle.
4. Displace 12 px. Set Multiply 75% (ink), duplicate to Soft Light 25% (fiber interaction).
5. Use the high-contrast weave image as a mask on the artwork to let threads cut through. Add a blurred artwork copy on Multiply 25% for dye bleed.
6. Selective dodge on existing fabric highlights so they read through the ink. Export at 300 DPI.

Common snags and fast fixes

• Looks “pasted on”: Increase displacement slightly and mask highlights on the Multiply pass so real shine comes through.
• Harsh ripple artifacts: Blur the displacement 2–4 px and reduce strength 10–20%.
• Edges too crisp on fabric: Add 0.5–1 px blur to an artwork copy; mask it into shadow areas only.
• Metal feels dull: Add a soft Screen highlight in the reflection direction; keep it under 40%.
• Color shock in print: Convert a copy to CMYK, reduce saturation 10–20%, and test a small swatch.

10-minute template setup (do this once)

1. Create a master PSD with groups: 01 Base Texture, 02 Displacement (grayscale), 03 Artwork Smart Object, 04 Form (displaced), 05 Surface (overlay/soft light), 06 Edge & Shadows, 07 Finishing.
2. Wire the displacement to your Form layer and save. Next time, just swap the base texture and update the Smart Object.

What to expect

• First time: 30–60 minutes. With the template: 10–20 minutes per mockup.
• Two iterations usually push realism over the line; the edge-thickness + shadow catcher combo is the noticeable leap.

Next best step (today)

1. Pick one substrate you use most. Generate or shoot a clean base at 2–4K.
2. Build the two-pass stack once (Form + Surface + Edge + Shadow).
3. Drop in two different labels, export, and get a quick thumbs-up from a colleague on realism (1–5). Iterate once.

Keep it simple, keep it repeatable. Your template becomes a production engine: swap the texture, update the Smart Object, export. Realistic mockups on paper, plastic, metal, and fabric — consistently and fast.

Quick win: Export a small sample of 200 contacts from your CRM and open it in Excel or Google Sheets. Use the UNIQUE function or conditional formatting to spot exact duplicate emails in under 5 minutes.

Cleaning, deduping and enriching a CRM doesn’t have to be technical or risky for privacy. Focus on small, repeatable steps: export, back up, clean locally, dedupe with clear rules, and only enrich with public or consent-backed data.

What you’ll need

• A CSV export from your CRM (always keep a backup copy).
• Excel or Google Sheets (for quick fixes) or OpenRefine (free desktop tool for stronger matching).
• A privacy rule: no uploading full contact lists to free cloud tools without consent.
• Optional: a privacy-focused enrichment service or manual lookup for high-value records only.

Step-by-step (practical)

1. Backup: export the full list and store a dated copy offline.
2. Sample: work on a 200–500 row sample for rules and testing.
3. Normalize fields: split full names, standardize phone formats, lowercase emails, remove spaces.
4. Exact dedupe: remove exact email duplicates first (email is usually the best key).
5. Fuzzy dedupe: use OpenRefine’s clustering or Excel’s Fuzzy Lookup to catch typos in names and companies.
6. Merge rules: create a simple policy—prefer non-empty email, latest update timestamp, and keep custom fields from the most recent record.
7. Enrich selectively: add verified company domain or industry from public sources; only enrich top 5–10% of contacts to limit privacy exposure and cost.
8. Re-import: test with a small batch back into your CRM, confirm results, then roll out.

Example

Use OpenRefine to cluster and merge “Jon Smith”, “Jonathan Smith”, and “J. Smith” as one contact. Keep the most recent email and copy non-empty custom fields into the merged record.

Common mistakes & fixes

• Rushing merges — always test on a sample first.
• Uploading raw PII to public tools — fix: anonymize or run locally.
• Over-enriching everyone — fix: enrich only high-value segments.
• No rollback plan — fix: keep backups and export a change log before import.

Copy-paste AI prompt (use locally or with a privacy-respecting provider):

“Clean this CSV data. Split Full Name into First Name and Last Name, lowercase and trim Email, standardize Phone to +CountryCode format if possible, normalize Company names (remove LLC/Inc), and flag possible duplicates. Output a cleaned CSV with columns: First Name, Last Name, Email, Phone, Company, Duplicate Flag, Notes. Do not share any data externally.”

7-day action plan

• Day 1: Export + backup + pick sample.
• Day 2: Normalize fields in sample.
• Day 3: Run exact and fuzzy dedupe rules.
• Day 4: Create merge policy and test merges.
• Day 5: Enrich high-value records only.
• Day 6: Test import small batch and review results.
• Day 7: Full import and set regular cadence (monthly or quarterly).

Small, consistent steps win. Start with a safe sample, create clear rules, protect privacy, and automate what you trust. Try the 5-minute duplicate check now — you’ll see instant value.

All the best,

Jeff

Quick win: Finish this in 20 minutes and you’ll likely avoid at least one unwanted renewal in the next 30 days.

Subscriptions hide in small charges and clutter. The answer is simple: one spreadsheet as your source of truth, two calendar nudges per item, and a 5-minute monthly habit. Here’s a tight, practical workflow you can do now.

What you’ll need

• Last 2 months of bank/credit card statements (PDF or CSV)
• Your email receipts or a forwarded receipts inbox
• A spreadsheet (Google Sheets or Excel) and your phone calendar
• 20 minutes now; 5–10 minutes/month ongoing

20-minute step-by-step (do it now)

1. Scan (5 min) — Open your latest statement and search email for keywords: “receipt”, “subscription”, “renewal”, “invoice”. Note repeating vendor names.
2. Create the sheet (3 min) — Columns: Service | Typical Amount | Frequency | Last Charge | Next Billing | Auto-renew (Y/N) | Cancel-by | Reminder Date | Notes.
3. Add top items (7–8 min) — Enter 8–12 obvious subscriptions: streaming, cloud storage, apps, memberships. Use last charge to estimate next billing.
4. Set calendar nudges (3 min) — For each item create an event titled: “Action: [Service] renewal — keep/cancel”. Primary alert 7 days before for annuals (3–5 for monthlies) + follow-up 1 day before. Put cancel-by date in the note.
5. Quick verify (2 min) — For the top 3 expensive items, log into the vendor’s billing page to confirm cancel rules and adjust Cancel-by in the sheet.

Worked example (copy into your sheet)

• Service: StreamPlus — Typical Amount: $12 — Frequency: Monthly — Last Charge: 2025-11-15 — Next Billing: 2025-12-15 — Auto-renew: Y — Cancel-by: 2025-12-08 — Reminder Date: 2025-12-08

Common mistakes & fixes

• Missing tiny charges — fix: filter statements for vendors under $10 and search email for short vendor names.
• Wrong dates — fix: use last charge date to estimate next billing and confirm on vendor account pages.
• Giving out passwords — fix: never share full credentials; use exports or a dedicated forwarding address.

7-day action plan (fast)

1. Day 1: Download statements and search inbox for receipts.
2. Day 2: Create spreadsheet and add top 8–12 subscriptions.
3. Day 3: Add billing dates, auto-renew flags, and set calendar nudges.
4. Day 4: Use the AI prompt below to cross-check your list (verify results).
5. Day 5: Reconcile missed items from statements and update sheet.
6. Day 6: Cancel 1–2 low-value subscriptions you don’t use.
7. Day 7: Schedule your 5-minute monthly review on the calendar.

Copy-paste AI prompt — extract recurring subscriptions

“I will paste lines from my bank statement or email receipts. Identify recurring subscriptions only. For each item output a CSV row with headers: Service, Typical Amount, Billing Frequency (monthly/annual/unknown), Last Charge Date (if shown), Estimated Next Billing Date, Likely Auto-renew (Y/N), Suggested Cancel-by Date (YYYY-MM-DD). Do not include one-off purchases. If data is missing, mark as ‘unknown’.”

Prompt variant — create calendar events from CSV

“Given this CSV of subscriptions (Service, Next Billing Date, Cancel-by), return a CSV with: Service, Event Title, Primary Reminder Date, Follow-up Reminder Date, Event Note (one sentence with cancel-by info). Use YYYY-MM-DD dates for reminders.”

Small reminder: start simple, protect your privacy, and keep the spreadsheet as your single source of truth. Do it now — 20 minutes and you’ve bought yourself several worry-free billing cycles.

Great point — keeping it simple and prioritizing follow-ups is the single most practical way to make a personal CRM stick. Small, regular actions beat big, rare efforts every time.

Here’s a clear, do-first plan you can set up this weekend. It’s low-tech, low-cost, and uses AI where it helps most: summarizing notes and drafting outreach.

What you’ll need

• A place to store contacts: a spreadsheet (Google/Excel), Airtable, or Notion — whichever you already use.
• Your calendar (Google/Outlook/Apple) for reminders.
• An AI assistant (chat tool) for summaries and message drafts. Automations (Zapier/IFTTT) are optional.

Step-by-step setup

1. Create one master table with these columns: Name, Relationship (dropdown), Last Contact Date, Next Action (short), Follow-up Date, Tags, Short Notes, Source.
2. Pick 5–8 practical tags (e.g., client, prospect, mentor, follow-up, referral). Too many tags = decision fatigue.
3. Decide simple rules for follow-ups (examples): new lead = 3 days, warm = 2 weeks, client check-in = monthly. Add these as a default note or formula.
4. Connect Follow-up Date to your calendar. If you can’t automate, block a weekly 20–30 minute review to set dates and send messages.
5. Create 3 short templates: check-in, value-share, next-step. Use AI to personalize each before sending.

Example (one contact row)

• Name: Sarah Lee
• Relationship: Prospect
• Last Contact: 2025-11-20
• Next Action: Send pricing overview
• Follow-up Date: 2025-11-25
• Tags: prospect, lead-source-email

Common mistakes & fixes

• Mistake: Over-tagging and over-detailing. Fix: Limit tags to 8 and keep notes to 1–3 sentences.
• Mistake: Letting automation run unchecked. Fix: Review automated items weekly so nothing looks robotic.
• Mistake: Waiting to add contacts. Fix: Add the contact within 24 hours with one-line notes.

Practical AI prompt (copy-paste)

“Summarize the following meeting note into three bullet points, suggest one clear next action with a deadline, and draft a two-sentence friendly follow-up email tailored to a professional contact. Meeting note: [paste meeting notes here].”

Simple 5-step action plan (this weekend)

1. Pick your tool and create the master table (30–45 min).
2. Add 10 recent contacts and a one-line note for each (20–30 min).
3. Set follow-up rules and a calendar sync or weekly review block (15 min).
4. Create 3 templates and stash your AI prompt for quick personalization (15–20 min).
5. Run your first weekly review: update dates and send 3 follow-ups (30 min).

Keep it tiny and consistent: 30–60 minutes upfront, then 10–30 minutes weekly. That rhythm creates momentum — and fewer missed opportunities.

Love the micro-sprint — clear, fast, and practical. Let’s add an insider upgrade: make every guide audit-proof and rollout-ready in a single pass. Two outputs, one prompt — a simple guide for users and an audit appendix for SMEs and auditors. That’s how you speed adoption and keep risk tight.

Why this helps

• Users get a 60-second checklist they can do immediately.
• SMEs get traceability to the exact policy clauses, exceptions, and evidence.
• You get painless updates when the policy changes (diff the appendix, not the whole guide).

What you’ll need

• One policy section (500–1,000 words) with clause or paragraph numbers.
• Target role(s) and one typical task they perform.
• Your existing micro-sprint template plus a second “Audit Appendix” template (obligations, exceptions, source map, evidence).
• Access to an LLM. One SME. One pilot user.

Run this enhanced workflow

1. Label the source: Add simple labels to the policy section (e.g., P1, P2…). This becomes your “traceability ribbon.”
2. Dual-output draft: Use the prompt below to produce both a Guide Card and an Audit Appendix in one go.
3. Scenario test: Ask the AI to generate two brief real-life scenarios per role and verify the checklist works step-by-step. Adjust wording to remove friction.
4. SME delta review: Send only the Audit Appendix and flagged lines to the SME for sign-off. Keep the review tight and fast.
5. Pilot in 10 minutes: Time one user completing a task with the Guide Card. Capture confusion and missing steps.
6. Publish: Post the Guide Card with version and review date. Store the Audit Appendix alongside it (internal-only).
7. Measure: Track time-to-task and one 3-question comprehension check. Iterate monthly or after any incident.

Copy-paste AI prompt (dual output: user guide + audit appendix)

“You are a senior compliance UX writer and audit specialist. Convert the policy text below into two outputs for [ROLE]. Keep language at Grade 7–8, start each checklist step with a verb, limit steps to 7–12 words, and avoid jargon.

Output A — Guide Card (for users): 1) Purpose (≤25 words); 2) Who/When (scope and frequency); 3) Checklist (3–6 steps); 4) Two examples (correct vs incorrect); 5) FAQ (3 short Q&As); 6) If unsure (escalation path); 7) 60-second recap.

Output B — Audit Appendix (for SME/auditors): a) Obligations list; b) Exceptions and conditions; c) Deadlines and frequency; d) Evidence required (screenshots/logs/forms); e) Systems/tools involved; f) Risk rating per step (High/Med/Low); g) Owner role and handoffs; h) Review triggers (what changes require update); i) Source map linking each checklist step to the exact policy clause IDs with short quotes; j) Flag lines containing must/shall/required/unless or any ambiguous phrases.

Return JSON with keys: guide {purpose, scope, checklist[], examples[{correct, incorrect}], faq[], escalation, recap}, audit {obligations[], exceptions[], deadlines[], evidence[], systems[], risk[], owners[], review_triggers[], source_map[{step, clause_id, quote}], flags[]}. Maintain clause IDs from the source text. Here is the policy: [PASTE TEXT].”

Quick variants

• Poster card: “Summarize the Guide Card into a 40-word poster with a 3-step checklist for [ROLE].”
• Role matrix: “From this policy, list obligations by role (Role → Action → Frequency → Evidence).”
• Delta update: “Compare OLD vs NEW policy. Output changed obligations, impacted roles, and the exact checklist steps to update. Include a one-line reason per change.”
• Scenario test: “Given the Guide Card, simulate two common edge cases and show where users might fail. Suggest fixes in plain English.”

Example (mini)

• Policy line: “Employees must report suspected phishing emails within 15 minutes.”
• Guide snippet: Purpose — Stop breaches fast. Checklist — 1) Click Report Phish in Outlook; 2) Do not forward or reply; 3) If tool fails, email security@ with subject: PHISH-URGENT.
• Audit appendix snippet: Obligation — Report phishing. Deadline — 15 minutes. Evidence — Tool log entry or email header. Source map — Step 1 → P3 “must report”. Risk — High.

Quality guardrails (use these every time)

• Verb-first, max 12 words: Turns “awareness” into action. If a step is long, split it.
• One-sitting test: A new hire should complete the task in one attempt without help.
• Traceability ribbon: Every step links to a clause ID and quote. This saves hours during audits.
• Evidence-first design: Ask, “What will prove this happened?” Add that to the checklist and appendix.

Common mistakes and fast fixes

• Problem: Guides drift from policy language. Fix: Keep clause IDs in the source map; update via the Delta prompt.
• Problem: Legal nuance gets flattened. Fix: SME reviews only the flags and exceptions; keep a 24-hour SLA.
• Problem: Steps are tool-agnostic and vague. Fix: Add system names and where to click. Include an offline fallback.
• Problem: Users still ask the same question. Fix: Promote that question to the FAQ and tighten the step text.

1-week action plan

1. Day 1: Choose one policy section. Label clauses P1–Pn. Identify one role.
2. Day 2: Run the dual-output prompt. Produce Guide Card + Audit Appendix.
3. Day 3: SME delta review of flags and exceptions only. Apply edits.
4. Day 4: Pilot with one user. Time-to-task, 3-question quiz. Update wording.
5. Day 5: Publish with version and review date. File the Audit Appendix. Add to your dashboard (time-to-task, quiz score).
6. Day 6–7: Run the Delta prompt on a second policy. You’ll improve speed by 30–40% on the second pass.

Closing thought

Ship the first dual-output guide this week. When every step ties back to a clause and evidence, you get clarity for users and confidence for auditors — that’s how compliance actually sticks.