Jeff Bullas

Good point: focusing on classroom-ready, practical prompts makes AI actually useful — not just interesting. Here’s a simple, actionable way to design SEL activities and reflection prompts you can use tomorrow.

Why this works: AI helps generate age-appropriate language, scaffolded questions, and quick rubrics. That saves prep time and gives you several options to test with students.

What you’ll need

• Clear SEL goal (e.g., empathy, self-regulation, teamwork).
• Grade level and time available (5–10, 20–30 minutes).
• A device and an AI chat tool (simple web chat will do).
• Student profile or examples of typical responses.

Step-by-step process

1. Choose a single learning objective. Keep it narrow (e.g., “recognize feelings in others”).
2. Ask AI to create 3 activity options: quick warm-up, paired share, written reflection. Specify grade and time.
3. Generate 5 reflection prompts at increasing depth (surface → deeper → application).
4. Ask AI for simple success criteria or a one-point rubric (what “good” looks like for that prompt).
5. Test with one small group, collect quick feedback, tweak language or time, then scale.

Example (middle school — empathy, 20 minutes)

1. Warm-up: 5-minute emotion charades (students act, others guess).
2. Paired share: Tell a time you felt left out. Partner reflects what they heard in one sentence.
3. Written reflection prompts (choose 1):
   • What happened? How did you feel?
   • What might the other person have been feeling?
   • Next time, what could you say or do differently?
4. Success criteria: Student names emotion, describes one perspective, and lists one supportive response.

Common mistakes & fixes

• Mistake: Prompts too vague. Fix: Add grade, time, example student responses.
• Mistake: Overloading with too many tasks. Fix: Pick one objective per session.
• Mistake: Ignoring privacy. Fix: Use hypothetical scenarios or anonymize responses.

Quick 3-step action plan (do-first mindset)

1. Today: Pick one SEL goal and run the AI prompt below to produce 3 activity options.
2. Tomorrow: Try one option with a small group and collect 2 quick student notes.
3. Next class: Adjust language/time and roll out to the whole class.

Copy-paste AI prompt (use as-is)

“Design three classroom-ready SEL activities for [grade X] focused on [SEL goal]. Each activity should include: time required, step-by-step student instructions, one reflection prompt, and a one-sentence success criterion. Keep language age-appropriate and provide a shorter option for 10 minutes.”

What to expect: Within seconds you’ll have multiple drafts. Pick one, test fast, iterate. Small adjustments—simpler words, shorter times—will make it classroom-ready.

Reminder: Use AI to prototype and speed up prep, not to replace your teacher judgment. Try one activity this week and refine from real student responses.

Hook: Great question — detecting spam leads and low-quality traffic is one of the fastest wins for small teams. You don’t need fancy tools: tidy data, a few rules, and an AI helper will do most of the heavy lifting.

Quick correction: Don’t paste full, sensitive lead data (emails, phones, full IPs) into a public chat. Mask or anonymize personal data before sending samples to any shared AI service.

What you’ll need

• Lead export (CSV) with: IP (or hashed), timestamp, referrer, user agent, email domain, phone (masked), form answers, UTM tags, session duration/pages if available.
• A spreadsheet (Google Sheets or Excel) and basic filters.
• An AI assistant (chat model you trust) or a low-code automation to call an API.

Step-by-step workflow

1. Export a 2–4 week sample (200–500 rows). Mask emails/phones (e.g., jan***@domain.com).
2. Add helper columns: email domain, time-to-submit (seconds), pages viewed, repeated-email-count, submissions-per-IP (windowed), user-agent-score (empty/robotic).
3. Apply quick deterministic rules to flag obvious spam: disposable domains, time-to-submit < 3–5s (tune this), same IP > 5 in 1 hour, empty/referrer mismatch, suspicious UAs.
4. Take the remaining sample (50–100 rows, anonymized) and ask the AI to cluster and label entries with a short reason and confidence score (0–100).
   1. Output format: label (clean/likely-spam/low-quality), reason (one line), score (0–100).
5. Manually review flagged rows (expect false positives). Update thresholds or whitelist domains and rerun weekly.
6. Automate: tag leads in your CRM using combined rule + AI score. Route mid-score leads for manual review.

Example (what AI might return)

• Label: likely-spam — Reason: disposable email + same IP as 12 others within 30 min — Score: 92
• Label: low-quality — Reason: session duration 8s, one page, UTM missing — Score: 42

Common mistakes & fixes

• Too aggressive time threshold — fix by sampling real users and setting a 5–10% false-positive target.
• Pasting raw PII into public AI — always mask first.
• Relying only on rules — combine rules plus AI scores and human review for edge cases.

Copy-paste AI prompt (anonymize real values first)

I have a CSV with these columns: timestamp, email_domain, masked_email, masked_phone, ip_hash, referrer, user_agent, time_to_submit_sec, pages_viewed, utm_source. Please review this 75-row anonymized sample and return a CSV-style list with: label (clean/likely-spam/low-quality), reason (one short sentence explaining the trigger), and score (0-100). Highlight common patterns and suggest 3 simple rule thresholds I can implement in a spreadsheet to reduce false positives.

Immediate 3-step action plan

1. Export 2 weeks of leads and mask PII now.
2. Run the quick rules above and sample 50–100 anonymized rows for AI review.
3. Tag and automate the obvious ones; queue mid-scores for manual review for two weeks.

Keep it iterative: weekly tweaks and a small review pool will turn noisy leads into a reliable pipeline quickly.

Nice and practical — that five-minute tactic is a fantastic quick win. It forces discipline: short problem, short ideas, quick human triage. Here’s a compact, do-first playbook to make that habit safe, repeatable and owned by you.

What you’ll need

• A chat AI you trust for brainstorming.
• A one-sentence problem statement.
• Your 5-item ethics checklist (privacy, fairness, legal, brand fit, harm potential).
• A provenance log (date, prompt, model/version, outcome tags).

Step-by-step: from live problem to owned idea

1. Prepare (5–10 min): Write the one-sentence problem and open your log. Set the ethics checklist where you can see it.
2. Run ideation (5 min): Paste the prompt below, ask for 10 one-line ideas with one-line ethical flags. Stop once you have output.
3. Scan & triage (5–15 min): Mark three to keep. Use tags: Keep / Revise / Reject. Note why (feasibility, ethics, client fit).
4. Provenance (2 min): Log prompt text, date, tool name, and which ideas you kept and why.
5. Micro-test (1–7 days): Turn one idea into a tiny experiment (one survey, one social post, one A/B email).

Copy-paste prompt (use as-is)

“You are an idea-generation assistant. For this problem: [one-sentence problem]. Produce 10 concise, one-line ideas. For each idea include: one-line description, one ethical concern, and one quick validation test. Do not write final copy or use proprietary or personal data.”

Worked example (real but simple)

Problem: Increase newsletter sign-ups from small business owners aged 40+.

• Idea 1 — Free 3-tip checklist tailored to legacy businesses. Ethical concern: may inflate expectations. Test: Run a small FB ad to a landing page and measure sign-up rate in 72 hours.
• Idea 2 — Short case study video featuring a peer. Ethical concern: consent & privacy for testimonial. Test: Ask one client for permission and track engagement.
• Idea 3 — Offer a live 20-min clinic Q&A. Ethical concern: giving advice without formal consulting disclaimers. Test: Promote to 50 prospects and count attendees.

Do / Do not checklist

• Do: Log prompt + model + date every session.
• Do: Require one ethical note per idea.
• Do not: Use AI outputs as final deliverables without human rewrite and sign-off.
• Do not: Feed proprietary or personal client data into the prompt.

Common mistakes & fixes

• Mistake: Treating AI ideas as finished work. Fix: Always add a human refinement step before sharing.
• Mistake: No provenance. Fix: Keep a one-line log entry for every session.
• Mistake: Skipping ethics checks. Fix: Reject or revise any idea flagged on your 5-item checklist.

1-week action plan (do-first)

1. Day 1: Create boundary list + 5-item ethics checklist (30–60m).
2. Day 2: Run two 10-minute ideation sprints with the prompt (20–30m total).
3. Day 3: Triage outputs, pick 3 ideas and log provenance (30–60m).
4. Days 4–7: Run one micro-test, gather results, update the checklist and repeat.

Keep it simple, repeat the habit, and treat AI as your rapid idea engine — you stay the owner of the decisions.

Spot on about speed and trust. Your confidence gate, timeboxed reviews, and clear KPIs are the backbone. I’ll add a few low-cost tricks that make the workflow sturdier without extra tools: an “abstain” rule, a tiny calibration pack, and a shadow QA sample. These three raise quality fast for small teams.

Big idea: Make the AI prove its choice. Require a short quote from the text as evidence, allow it to say “uncertain/abstain,” and keep a small set of hard examples for calibration. That combination cuts false positives and boosts stakeholder trust.

What you’ll need

• Transcripts in one folder (CSV or text, segments per row if possible).
• One-page codebook (4–8 codes) with: definition, 1 positive example, 1 “do not include” example.
• AI model that returns a confidence score (cloud or local) and a simple spreadsheet tracker.
• Calibration pack: 12 segments (1–2 per code + 2 ambiguous) pre-labeled by a human.
• Review cadence: 20–30 minute blocks; weekly 20–30 minute reconciliation.

How to run it (step-by-step)

1. Segment smart (optional if not already segmented)
   • Split interviews into 1–3 sentence segments. Avoid slicing mid-thought.
   • Keep the speaker label and transcript_id with each segment.
2. Upgrade the codebook
   • Add a one-line primary vs. secondary rule per code. Example: “If both Pricing and Ease-of-Use appear, set primary to Pricing.”
   • Add an explicit Uncertain/Abstain pseudo-code with when-to-use guidance.
3. Calibrate first
   • Run the calibration pack through the AI. Compare to human labels.
   • For any mismatch, add a single-line rule or example to the codebook. This tightens the model before you touch real volume.
4. Pilot 5–10%
   • Run AI, capture assigned_codes, confidence, and a one-sentence justification plus a direct quote.
   • Triage: confidence ≥0.70 = green; 0.55–0.69 = yellow; <0.55 = red.
   • Review all yellow/red. If a green looks odd, mark it for discussion.
5. Refine and lock
   • Document 3–5 recurring errors as rules: “If mentions ‘price comparison’ but no dissatisfaction, do NOT tag Pricing Pain.”
   • Freeze the codebook as Version 1.0 for the full run. Avoid silent drift.
6. Full pass + shadow QA
   • Run all data. Humans review yellow/red only.
   • Randomly spot-check 5% of green items (“shadow QA”). This catches quiet mistakes early.
7. Reconcile weekly
   • 20–30 minutes: confirm fixes for repeated patterns, update to Version 1.1, bulk reclassify matches, and re-run only affected rows.

High-value add: the “prove it” prompt pattern

• Requiring a direct quote forces the model to ground its label in the text.
• Allowing “uncertain/abstain” reduces bad auto-accepts that erode trust.
• Including a few negative examples (“don’t label when…”) tightens boundaries fast.

Copy-paste prompt (base classifier)

Role: You are a careful qualitative coding assistant. Use the codebook to label the segment. If the evidence is weak or ambiguous, select “uncertain” and explain why.

Codebook (name, definition, positive example, negative example): [paste your 4–8 codes with one positive and one negative example each]

Instruction: Return a single JSON object with fields: segment_id, primary_code, secondary_codes (array), confidence (0–1), justification (1 sentence), evidence_quote (exact short quote from the segment), abstain (true/false). If abstain=true, set primary_code=”uncertain” and confidence ≤0.60.

Segment (with speaker and context): [paste segment]

Optional prompt (segmenter)

Split the interview text into analytical segments of 1–3 sentences, keeping speaker labels. Avoid breaking mid-idea. Return JSON array with fields: transcript_id, segment_id, speaker, text.

Worked micro-example

• Codebook snippet: Pricing Pain = user expresses dissatisfaction with cost; Positive: “too expensive for what it offers”; Negative: “pricey but worth it” (do not tag as pain).
• Segment: “It’s a bit pricey, but it saves me hours each week.”
• Expected outcome: primary_code = Value Perception; evidence_quote = “saves me hours each week”; do not tag Pricing Pain.

Metrics that convince stakeholders

• % auto-accepted (greens) – target 70%+ after two iterations.
• Human review rate – aim to drop toward 10–15%.
• Agreement (human vs AI on shadow QA) – aim 80%+.
• Time saved per interview – baseline vs current.
• Simple ROI: (Manual hours − AI+review hours) × hourly rate − model cost.

Common mistakes and quick fixes

• Overlapping codes everywhere: Add primary/secondary rules and keep only primary for headline charts.
• No “uncertain” path: Forces bad greens. Add abstain and keep confidence ≤0.60 when used.
• Model can’t handle jargon: Add a short glossary to the codebook; include 1–2 jargon examples per code.
• Greens never reviewed: Shadow-check 5% of greens weekly.
• Segments too long: Cap at ~3 sentences; long blocks confuse the model.
• Moving targets: Version the codebook weekly. Re-run only impacted rows to save cost.

5-day, low-stress action plan

1. Day 1: Draft the one-page codebook with positive and negative examples. Build your 12-item calibration pack.
2. Day 2: Calibrate on the 12 items; write 3–5 one-line rules from mismatches. Set thresholds (≥0.70 green).
3. Day 3: Pilot 5–10% of interviews. Review all yellow/red; log errors.
4. Day 4: Freeze codebook v1.0; run full pass; shadow-check 5% of greens.
5. Day 5: Reconcile 20–30 minutes; bulk-fix repeating errors; calculate time saved and agreement; share a one-page results summary.

Closing thought: Small teams win with clear guardrails, not bigger tools. Make the AI show its work, allow abstention, and keep a tiny calibration set. You’ll move fast, protect trust, and get to consistent insights without blowing the budget.

Let’s turn one page into proof. In 60–90 minutes you can run a form-focused audit, ship 3 fixes, and see score and usability gains. Here’s the playbook, prompts, and example code you can copy today.

What you’ll need (10 minutes)

• One high-value page with a form (sign-up, checkout, contact).
• Automated scan export (axe/Lighthouse JSON) and a short HTML snippet of the form.
• 3 screenshots (top of page, form fields, submit area).
• Access to an AI assistant. Use sanitized HTML/screenshots only.
• QA: keyboard-only pass, plus a quick screen-reader check (NVDA or VoiceOver).

The Form Fix Pack micro-sprint (6 steps)

1. Run the scanner on your chosen page; export JSON and save screenshots.
2. Copy the form’s HTML (form tag to submit button). Remove user data and unique IDs.
3. Paste your JSON + HTML + screenshots into the Form Fix Pack prompt below.
4. Create tickets for the 3 quick wins the AI marks. Add acceptance tests.
5. Implement and re-scan. Run a keyboard-only pass and a short screen-reader pass.
6. If color fails appear, run the Color Tokenizer prompt to get a safe palette update.

Copy-paste AI prompt — Form Fix Pack (prioritized tasks + code)

You are an accessibility engineer. I will paste 1) an automated scan JSON, 2) 2–3 screenshots, and 3) the HTML of a form. Create a prioritized Form Fix Pack (max 12 items). For each item include: issue title, severity (High/Med/Low), exact location (CSS selector or HTML snippet), plain-English explanation, step-by-step fix, one copyable code example (HTML/CSS/JS), estimated dev time (Low <1h, Med 1–3h, High >3h), and a single acceptance test a QA person can run. Mark the top 3 quick wins. Also produce: a) one site-wide baseline suggestion if relevant, b) 3 ready-to-paste Jira ticket summaries with acceptance criteria. Prefer HTML truth when screenshots conflict. Sanitize any sensitive data.

Copy-paste AI prompt — Color Tokenizer (fast, safe contrast)

Act as a design systems accessibility specialist. Input: a list of brand colors and any contrast failures from a scan. Output: 1) an accessible color token set that meets WCAG 2.1 AA (4.5:1 for body text, 3:1 for UI components), 2) a CSS variables patch, 3) a migration map from old colors to new tokens, 4) 3 visual checks to confirm legibility in light and dark areas. Keep tokens under 12. Example output format: tokens, CSS variables, mapping, test steps.

Copy-paste AI prompt — Interactive Widgets (menus, modals, tabs)

You are an accessibility engineer. I will paste HTML/JS for one widget (menu, modal, tabs, tooltip). Return: correct roles/attributes, full keyboard map (Tab/Shift+Tab, Arrow keys, Home/End, Escape), minimal JS needed (focus management, aria-expanded toggling, trapping/returning focus), and one acceptance test per behavior. Include a copyable code block for the final widget markup and JS sketch.

Worked example — checkout form (3 high-impact fixes)

1. Missing labels and programmatic names (High)

Fix (HTML): <label for=”email”>Email</label> <input id=”email” name=”email” type=”email” autocomplete=”email” required aria-describedby=”email-help”> <small id=”email-help”>We’ll send the receipt here.</small>

Acceptance: Screen reader announces “Email, edit, required. We’ll send the receipt here.”

1. Error messaging not announced (High)

Fix (HTML + JS): Add a live region and link errors to fields. Example container: <div id=”form-status” aria-live=”polite” role=”status”></div> On validation fail: set aria-invalid=”true” on the input and point aria-describedby to the error text ID. Example error: <p id=”email-error” class=”error”>Enter a valid email.</p>

Acceptance: After submitting invalid input, the error is read out automatically and focus moves to the first invalid field.

1. Weak focus visibility and order (Med)

Fix (CSS baseline): :where(button,[role=”button”],a,input,select,textarea,[tabindex]):focus{outline:none} :where(button,[role=”button”],a,input,select,textarea,[tabindex]):focus-visible{outline:3px solid #1a73e8;outline-offset:2px;border-radius:4px} Ensure DOM order matches visual order; remove tabindex values >=1 unless required for components.

Acceptance: Tabbing moves logically through the form; each focusable shows a thick visible outline.

Insider trick — ship a site-wide “error pattern” once

• One status region: <div id=”form-status” role=”status” aria-live=”polite”></div>
• One error class: .error{color:#b00020;font-weight:600}
• Validation script sets: aria-invalid, adds/removes aria-describedby for the error ID, sends a brief message to #form-status.
• Result: every form on the site announces errors without per-page custom code.

Acceptance test template you can reuse

• Keyboard: Starting at the address bar, press Tab through the page. Focus is always visible and moves in a logical order. Shift+Tab works backwards.
• Forms: Submitting empty required fields announces errors via the screen reader; focus lands on the first invalid field.
• Color: Body text contrast ≥ 4.5:1; UI text and icons ≥ 3:1. Hover/focus states remain ≥ target contrast.
• Widgets: Escape closes modals/menus and returns focus to the trigger. Arrow keys navigate tablists or menus.

Common mistakes and quick fixes

• Using placeholder as label → Always use a <label for> and keep placeholder as a hint only.
• Forgetting aria-expanded on toggles → Sync the attribute on open/close and update focus.
• Color fixes per element → Standardize tokens once; map old colors to new variables.
• Over-ARIA → Prefer native HTML elements first (button, a, input) before adding roles.
• No sanitization → Strip emails, names, and IDs before pasting into AI. Consider a “Sanitize this HTML” AI step first.

What to expect

• A short, prioritized backlog with exact locations and code snippets.
• 3 shippable fixes in under 90 minutes on a single page.
• Measurable score lift after re-scan, and fewer user friction points in forms.

5-day plan (lightweight, repeatable)

1. Mon: Pick top 10 pages. Run scans. Save screenshots + key HTML.
2. Tue: Run the Form Fix Pack prompt on the top 3 pages. Create tickets.
3. Wed–Thu: Ship site-wide baselines (focus, skip link, landmarks) and form error pattern. Implement quick wins on your highest-revenue flow.
4. Fri: Keyboard + short screen-reader check. Re-scan. Capture metrics: issues by severity, score delta, hours vs estimate.
5. Next week: Run the Widget prompt on your menu/modal, and the Color Tokenizer if contrast is still failing.

Your next move: pick the form page (sign-up, contact, or checkout). Paste your scan JSON + 3 screenshots + form HTML into the Form Fix Pack prompt above. I’ll help you turn it into 3 quick wins you can ship this week.

You nailed two big habits: treat engagement lift as a hypothesis, and save your seed/settings with notes. That’s how you turn happy accidents into a repeatable style.

Quick win

Let’s level up your prompt with layout and print-defect language. These two additions make outputs look convincingly vintage without getting technical.

What you’ll need

• Any image generator you like.
• 10–20 minutes and 2–4 runs.
• One era picked and a 3-color palette in mind.

Era cheat sheet (pick one)

• 1950s travel poster: teal, coral, cream; flat shapes; clean sans-serif.
• 1970s psychedelic: burnt orange, mustard, brown; hand-drawn lettering; swirls.
• 1920s art-deco: gold, black, ivory; geometric symmetry; elegant serif.
• 1980s neon/vapor: magenta, cyan, deep purple; grids; chrome vibes.
• 1930s WPA: muted blue, brick red, tan; bold blocky shapes; slab serif.

Texture and print defects (use 1–3)

• Halftone dots, paper grain, subtle creases, edge wear.
• Screenprint look, risograph grain, ink bleed.
• Off-register ink (1–2 mm misalignment), overprint, sun-faded colors.
• VHS scan lines (80s), letterpress impression (20s–50s).

Layout cues (simple stage directions)

• Top banner for headline, large centered subject, footer strip for small text.
• Generous margins, big shapes, simple composition.
• If type gets messy: ask for “blank space for headline” and add text later in your design app.

Prompt template (copy–paste and fill brackets)

Prompt: A [ERA] [MEDIUM] of [SUBJECT], flat graphic style, limited 3-color palette ([COLOR 1], [COLOR 2], [COLOR 3]), [1–2 TEXTURES from list], subtle wear on edges, simple composition with [LAYOUT CUE: top banner, centered subject, footer strip], typography style: [retro sans / art-deco serif / hand-drawn], slightly faded colors, no photorealism, no modern logos, no glossy 3D effects

Ready-to-use prompts (robust, plain English)

• 1950s travel: A 1950s vintage travel poster of a coastal highway and seaside diner, flat graphic style, limited 3-color palette (teal, coral, cream), halftone dots and paper grain, off-register ink 1–2 mm, simple composition with top banner headline and centered car silhouette, retro sans-serif typography, slightly faded colors, no photorealism, no modern logos, no glossy effects
• 1970s concert: A 1970s psychedelic concert poster of a guitar under a sunset, warm palette (burnt orange, mustard, deep brown), hand-drawn lettering style, risograph grain and subtle ink bleed, swirling patterns, large central shape, footer strip for ticket info, muted and slightly desaturated, no photorealism, no modern logos
• 1920s art-deco ad: A 1920s art-deco advertisement for a luxury ocean liner, gold and black with ivory accents, geometric symmetry, metallic sheen simulated on paper, light letterpress impression, elegant art-deco serif typography, strong borders, simple layout with top headline panel and centered ship, no photorealism, no modern logos
• 1980s neon: An 1980s neon cityscape poster, vaporwave colors (magenta, cyan, deep purple), grid horizon, VHS scan lines, paper texture and slight edge wear, bold retro type area left blank for headline, high contrast but not glossy, no photorealism, no modern logos
• 1930s WPA: A 1930s WPA-style national park poster, muted palette (forest green, brick red, tan), screenprint look with visible halftone and overprint, large blocky shapes, centered mountain and winding trail, slab-serif headline panel at top, slightly sun-faded, no modern logos, no gradients

Step-by-step (10–20 minutes)

1. Pick your era and 3 colors from the cheat sheet.
2. Choose 1–2 textures/defects (halftone, paper grain, off-register ink).
3. Decide the layout cue (top banner, centered subject, footer strip).
4. Paste the template or one of the ready prompts. Generate once.
5. Save the best result. Note one tweak: palette, texture strength, or type style.
6. Regenerate with only that tweak. Do 2–3 passes max.
7. If printing: export large (at least 3000 px on the shortest side). Convert to print color later in your design tool if needed.

Insider trick: two-pass method for cleaner type

• Pass 1: Ask the AI for “blank space for headline and small footer text” to avoid messy letters.
• Pass 2: Add the headline yourself in a design app using a lookalike style (retro sans, art-deco serif, slab). This keeps the art authentic and the text readable.

What to expect

• 2–4 runs usually land a believable vintage look.
• AI may over-detail or modernize colors. Counter with “muted, slightly desaturated, worn paper, no glossy.”
• Type accuracy varies. Use the two-pass method if it doesn’t behave.

Mistakes and quick fixes

• Still looks modern: add “off-register ink, halftone, worn edges, no gradients.”
• Too busy: add “simple composition, large shapes, generous margins, minimal text.”
• Colors too loud: add “muted, sun-faded, 1970s print inks.”
• Flat/shiny: add “screenprint look, risograph grain, paper texture.”
• Type feels wrong: specify “retro sans” or “art-deco serif,” or leave space and add type later.

Fast A/B test plan (pragmatic)

1. Create two versions: modern vs. your retro design.
2. Post at similar times to the same audience within 48 hours.
3. Track a simple metric (likes or CTR). Keep whatever wins by a clear margin.

Closing thought

Retro magic is constraints plus texture. Name the era, limit the colors, add print flaws, and keep the layout simple. Small words, big difference—generate once, tweak one thing, and ship.

Nice point — nailed it: the focal point is the conversion engine of your landing page. Your steps are clear and practical. I’ll add a compact, ready-to-use playbook with exact prompts, mobile tips, common mistakes and a short action plan so you can do this today.

What you’ll need:

• A one-line value statement for your offer (what it does, who for, why it matters).
• One or two visual ideas (product close-up, person using product, product in context).
• An AI image tool that accepts text prompts (web UI is easiest).
• A simple editor for cropping and overlays (site builder or free web editor).

Step-by-step — do this now:

1. Set the goal (5 minutes): Write one sentence: who, what, and feeling. Example: “Show a relaxed 45–60-year-old woman enjoying a calming face cream, trust and relief.”
2. Pick the focal point (5 minutes): Decide what grabs attention first — the product, the face, or the headline area.
3. Generate 3 AI variations (10–20 minutes): Use the prompt below and ask for three variations (change angle, lighting, expression).
4. Crop for mobile and desktop (10 minutes): Create two crops: wide hero for desktop, tall/centered for mobile. Keep focal point high enough to leave room for headline and CTA.
5. Add text overlay treatments (5–10 minutes): Use a subtle dark gradient behind text or 40% translucent panel so text reads on small screens.
6. Launch two variants (A/B) and run 7–14 days: Measure click-throughs and signups; keep winners and iterate.

Copy-paste AI prompt (use as-is, tweak product details):

“Create a clean, high-resolution landing page hero image: Close-up of a smiling 45–60-year-old woman holding a small jar of calming face cream, facing camera. Warm, soft lighting, shallow depth of field. Background blurred, muted pastel colors. Product label visible but not dominant. Leave negative space on the right for headline and CTA. Natural skin tones, optimistic mood, high contrast between subject and background. Provide three variations: 1) tighter crop, 2) wider with hands, 3) over-the-shoulder angle.”

Example (quick): If you sell a travel pillow: swap subject to “man on a train using a compact travel pillow, relaxed, window light, focal point on pillow, muted carriage interior.” Same prompt structure works.

Mistakes & fixes:

• Too-busy background: Fix with stronger blur or replace with flat color.
• Text unreadable: Add a subtle gradient or translucent panel under copy.
• Focal point off-screen on mobile: Re-crop to center the subject higher in the frame.
• Slow images: Export for web (JPEG/WEBP, 70–80% quality).

Action plan — next 48 hours:

1. Write your one-line goal and focal point (10 minutes).
2. Run the prompt, generate 3 variations (20–30 minutes).
3. Crop for mobile/desktop, add overlay, export optimized files (30 minutes).
4. Launch A/B test and check results in 7–14 days.

Quick reminder: Small visual changes win when they remove friction and guide the eye. Do one test this week — you’ll learn more from a live test than from perfecting a single image forever.

Nice point — wanting clearer user stories and acceptance criteria is the exact place to get big, fast wins.

Here’s a practical way to use AI to write crisp user stories that your team (and stakeholders) can act on today.

What you’ll need

• A short description of the feature or problem (1–3 sentences).
• Who the primary user is (role or persona).
• Any constraints or non-functional needs (performance, security, devices).
• An AI tool (chat or prompt-capable model) and a human to review outputs.

Step-by-step: how to do it

1. Write a single-sentence feature brief. Example: “Allow customers to save payment methods for future purchases.”
2. Feed that brief to the AI using the prompt below.
3. Ask the AI to return: a short user story, 4–6 acceptance criteria (Gherkin-like), edge cases, and test ideas.
4. Review with your team: pick missing conditions, simplify language, and assign priority.
5. Turn accepted criteria into tasks or test cases in your workflow tool.

Copy-paste AI prompt (ready to use)

Prompt: Act as a product coach. Given this feature brief, write one clear user story using the format “As a [role], I want [action], so that [benefit].” Then provide 5 acceptance criteria written as short, testable statements (use “Given/When/Then” where useful), list 3 edge cases, and suggest 3 manual test steps. Feature brief: “Allow customers to save payment methods for future purchases.” Constraints: PCI-compliant, must allow users to delete saved methods, mobile and desktop.

Variants

• Shorter: “Write a user story and 4 acceptance criteria for: [feature brief].”
• Role-focused: “Write stories for roles: customer, admin. Provide acceptance criteria and privacy considerations.”

Example output

User story: As a returning customer, I want to save a payment method so that I can checkout faster on future orders.

• AC1: Given I add a card, when I opt to save it, then the card is stored and shown in my payment methods.
• AC2: Given a saved card, when I select it at checkout, then it pre-fills the payment and completes the order.
• AC3: Given PCI constraints, when a card is saved, then only a token is stored (no card number visible).
• AC4: Given a saved card, when I delete it, then it is removed immediately from my account.
• AC5: Given mobile checkout, when I save a method, then it syncs across desktop and mobile.

Mistakes & fixes

• Vague benefits — Fix: insist on “so that” outcomes tied to user goals.
• Too many conditions in one AC — Fix: split into separate, testable ACs.
• Forgetting edge cases — Fix: always add at least 3 edge cases from AI output.
• Assuming implementation — Fix: keep acceptance criteria implementation-agnostic.

Action plan (do this in 30–60 minutes)

1. Write a one-line feature brief.
2. Run the copy-paste prompt above in your AI chat and generate output.
3. Review with one teammate, pick 3 ACs to start, and create tasks/tests.

Quick reminder: Use AI to draft, humans to validate. Start small, iterate, and you’ll get clearer stories that actually help delivery.