Jeff Bullas

Spot on: your triad — outcome, metric/time, and the emotional trigger — is the cleanest foundation. Let’s layer an insider tactic on top so you get bigger, safer wins with the same reviews.

Insider move: Build a Proof Ladder and triangulate. Single quotes are good; grouped, qualified proof is better. You’ll turn scattered reviews into a clear promise backed by multiple voices without over-claiming.

• Do: combine 2–3 reviews that point to the same result and add a qualifier (“in our pilot”, “most”, “on average”).
• Do: keep one short verbatim phrase to preserve the human voice.
• Do: show the number near the promise (don’t bury the proof in paragraph four).
• Don’t: average tiny samples into big claims; label small-n results as “in this case” or “among these customers”.
• Don’t: mix personas in one proof block (freelancers vs. enterprise) — segment for relevance.

What you’ll need

• Your filtered, specific reviews (as you outlined).
• A short list of top objections or decision criteria (price, speed, reliability, support).
• Simple tags for persona/use case (role, industry, plan).
• One human checker to validate numbers and consent.

Step-by-step: from raw quotes to a Proof Ladder

1. Cluster: Group reviews by outcome theme (e.g., “faster setup”, “lower cost”, “better support”). Aim for 3–10 reviews per cluster.
2. Score: For each review, rate 1–5 on Specificity, Outcome Strength, Relevance, and Differentiator. Prioritize 4–5 scores.
3. Triangulate: Pick 2–3 high scorers in one cluster that mention a similar number or timeframe.
4. Qualify: Decide the right qualifier for your sample size (“among 27 recent reviews”, “in Q3”, “in our beta”).
5. Create the Ladder:
   • Level 1: Verbatim quote — short emotional line.
   • Level 2: Quantified quote — number + timeframe from one review.
   • Level 3: Aggregated proof — a carefully worded summary across 2–3 reviews with a qualifier.
   • Level 4: Mini case snippet — 2 sentences: situation, change, result.
6. Place: Use Level 3 on hero/above the fold, Level 2 near CTAs, Level 1 as pull-quotes, Level 4 in email or below-the-fold sections.

Copy-paste AI prompt: Triangulated Proof Block

“You are a trustworthy copy editor. Using these customer reviews: [PASTE 2–3 REVIEWS], do the following: 1) Extract shared outcome; 2) List any consistent numbers/timeframes; 3) Identify one short verbatim phrase to keep. Then produce a Proof Ladder: A) Level 1: a 10–14 word verbatim quote; B) Level 2: a single quantified proof line tied to one review; C) Level 3: an aggregated proof line that combines the reviews with a clear qualifier (e.g., ‘among [count] recent reviews’ or ‘in our Q3 beta’); D) Level 4: a two-sentence mini case snippet (situation, change, result). Keep tone: clear, non-salesy, specific. Do not invent numbers. If numbers conflict, say so and default to a non-numeric qualifier.”

Worked example

• Review A: “Setup took under 30 minutes — we launched the same day and finally stopped firefighting.”
• Review B: “Was live in 25 minutes and saved us a week of back-and-forth.”
• Review C: “From signup to first result in half an hour. It just worked.”

• Level 1 (Verbatim): “Setup took under 30 minutes — we launched the same day.”
• Level 2 (Quantified): “Live in 25–30 minutes — customers report same-day launch without the back-and-forth.”
• Level 3 (Aggregated + qualifier): “Among three recent reviews, setup took ~30 minutes and enabled same-day launch.”
• Level 4 (Mini case): “Before: launches dragged for days. After: live in ~30 minutes with same-day results. Less firefighting, more doing.”

Turn it into assets

• Hero block: Headline promise + Level 3 line + one verbatim pull-quote.
• CTA section: Level 2 line directly under the button to reduce hesitation.
• Email subject: “Live in ~30 minutes (real customer proof inside)”
• Ad copy: 15 words using the verbatim phrase plus the timeframe.

Bonus prompt: Objection Crusher

“You are a helpful copywriter. Objection: [INSERT OBJECTION]. Reviews: [PASTE 3–5 RELEVANT REVIEWS]. Create: 1) a 10-word reassurance headline; 2) one proof line using a number/timeframe from a review; 3) a short qualifier to keep claims safe; 4) a 20-word CTA sentence. Keep one verbatim customer phrase. Tone: calm, credible.”

Quality guardrails

• Sample size labels: under 5 reviews = “in these reviews”; 5–20 = “among recent reviews”; 21+ = include the count.
• Conflict handling: if numbers differ, use a range (“25–30 minutes”) or drop the number and keep the timeframe.
• Persona fit: tag and deploy proof blocks only where that persona lands (avoid one-size-fits-all).
• Placement rule: place the strongest quantified line within the first screen on mobile.

Common mistakes & easy fixes

• Mistake: Aggregating apples and oranges. Fix: Cluster by use case before you triangulate.
• Mistake: Hiding qualifiers. Fix: Add a short qualifier right next to the number.
• Mistake: Letting AI invent averages. Fix: Add the instruction “Do not invent numbers” to every prompt.
• Mistake: Proof buried below the fold. Fix: Put Level 3 proof within the hero or above the primary CTA.

48-hour action plan

1. Export last 90 days of reviews; tag by outcome theme and persona.
2. Pick two themes with 3–10 specific reviews each (e.g., speed, savings).
3. Run the Triangulated Proof Block prompt for each theme; produce Levels 1–4.
4. Deploy Level 3 in your top landing hero; add Level 2 near your main CTA.
5. Test two variations per theme this week (headline + Level 3 line).
6. Track conversion lift and save winners to your CMS snippet library.

What to expect: clearer, faster decisions from visitors because your promise is backed by multiple aligned voices. Results vary, but this structure reliably improves clarity and trust — and that’s what moves the needle.

Keep building your library. One strong proof block per week turns into a persuasive, proof-driven site in a month.

Your two-title A/B idea is spot on — fast lift with almost no risk. Let’s level this up so your brief not only guides writing, but also targets the featured snippet and closes gaps on the top result.

High-value add: the Two-Pass Brief — Pass 1 shapes direction and angle; Pass 2 produces a ready-to-use writer brief and publish checklist. This avoids bloated outlines and locks in search intent.

What you’ll need:

• Single target keyword.
• One competitor URL (the page to beat).
• One-sentence audience summary (who, why now).
• Desired CTA (lead, sign-up, download, purchase).
• Page type (blog, product, service, comparison).
• Optional: location/country if relevant, and 2–3 existing internal pages you can link to.

Step-by-step (do this now):

1. Pass 1 — Direction + Angle: Run the first prompt to identify intent, SERP features, and 3 content angles. Pick one angle and a word-count budget per section before moving on.
2. Pass 2 — Final Brief: Run the second prompt to generate the one-page writer brief and the publish checklist. Ask for two title variants (benefit + question) for your CTR test.
3. Optional — Gap Overlay: Add the competitor URL to get 3 concrete gaps and exact H2s to close them.
4. Review: Trim any redundant H2s, set reading level (Grade 7–8), and confirm CTA placement (top + mid or end).
5. Ship: Hand the one-page brief to the writer, keep the checklist for publish day, and schedule your title A/B.

Copy-paste AI prompt — Pass 1 (Direction + Angle)

“You are an SEO editor. For the keyword: [KEYWORD], audience: [AUDIENCE], page type: [PAGE_TYPE], market: [COUNTRY/REGION]. Do the following concisely: 1) Classify primary search intent (informational, commercial, transactional) and state the user’s core problem in one sentence. 2) List likely SERP features (featured snippet, ‘People Also Ask’, reviews, videos) and the content format most likely to win. 3) Propose 3 distinct content angles (e.g., checklist, comparison, template, beginner guide) and recommend one based on intent. 4) Provide a section-level word budget that sums to a target total (e.g., 1,200–1,600 words). 5) List 5 must-answer user questions. Keep it under 200 words.”

Copy-paste AI prompt — Pass 2 (Final Brief + Checklist)

“Create a concise SEO brief for [KEYWORD]. Audience: [AUDIENCE]. CTA: [CTA]. Angle chosen: [ANGLE]. Include: 1) Two page titles (<=60 chars: benefit + question) and one meta description (<=155 chars) optimized for CTR. 2) One 40–50 word featured-snippet paragraph that directly answers the primary query in plain English. 3) H1 and H2/H3 outline with a word-count budget per section that totals [TARGET WORD COUNT]. 4) 7 semantic keywords to use naturally. 5) 5 FAQs for on-page Q&A (aim to match PAA). 6) Internal link suggestions: 3 anchors + page types (category, product/service, related blog). 7) Schema recommendations (e.g., Article + FAQPage + ImageObject) and 3 image directives (filenames, alt text pattern, one diagram idea). 8) Evidence pack: 3 data points to cite (stat type + source type, no links) and one expert quote prompt for the author. 9) Readability guidance (Grade 7–8), tone notes, and exact CTA placements (top after intro + mid or end). Deliver two outputs: A) one-page writer brief; B) publish checklist (meta, schema, internal links, images, compression, accessibility). Keep it short and actionable.”

Variant — Add competitor gaps

“Add a comparison to this competitor: [COMPETITOR_URL]. List 3 content gaps vs that page and provide exact H2/H3s to close those gaps, plus any missing entities or questions. If their word count is X, recommend a range to win and where to invest that extra depth.”

Insider tricks that raise win rates:

• Snippet-first intro: Put the 40–50 word answer immediately after H1; it boosts featured snippet odds and hooks readers fast.
• Word-budget discipline: Cap each H2 with a budget. Overrun? Cut, don’t pad. Google rewards clarity.
• Evidence placeholders: Tell the writer exactly where stats or examples go — it stops generic fluff and strengthens E-E-A-T.

Example (what you’ll get):

• Title A: Solar Tax Credits 2025: Save More on Home Solar
• Title B: What Are the 2025 Solar Tax Credits? (Simple Guide)
• Meta: See current 2025 solar tax credits, who qualifies, and how to claim them step-by-step.
• Snippet paragraph: In 2025, most U.S. homeowners can claim a 30% federal solar tax credit on eligible costs. You qualify if the system is new, installed at your home, and you own it. File IRS Form 5695 and carry forward unused credit to future years.
• Outline: H2 Eligibility (350), H2 What’s covered (250), H2 How to claim (400), H2 State add-ons (300), H2 FAQs (400) — Total ≈1,700 words.

Common mistakes & fixes:

• Overstuffed outlines — Fix: Use the “rule of 4” main H2s. If you need more, merge or move to FAQs.
• Writing to keyword, not intent — Fix: Lock the primary intent in Pass 1 and echo it in the snippet paragraph.
• Thin E-E-A-T — Fix: Add one expert quote prompt and two data placeholders per article.
• Ignoring SERP features — Fix: Demand a snippet paragraph and FAQ schema in the brief.
• Weak internal links — Fix: Specify anchor text + page type; place one internal link in the first 30% of the article.

What to expect:

• A one-page writer brief plus a 10–12 point publish checklist.
• Cleaner drafts, fewer rewrites, and faster indexing with clear snippet targeting.
• Measurable lift in CTR from your title test within 7–14 days, and better time-on-page from the snippet-first intro.

1-week action plan:

1. Day 1: Run Pass 1, pick the angle and word budget; run Pass 2 to get the brief + checklist.
2. Days 2–3: Draft to budget; place the 40–50 word snippet under H1; add data placeholders.
3. Day 4: On-page check (titles, meta, schema, internal links, images, compression, accessibility).
4. Day 5: Publish, request indexing; queue outreach for one relevant backlink target type.
5. Days 6–7: Monitor CTR and rankings; if CTR lags, swap to the alternate title and tighten the meta.

Paste your keyword, audience, page type, CTA, and one competitor URL, and I’ll help you generate Pass 1 now.

Quick win: In the next 5 minutes export one week of your time-tracking as a CSV and paste 10 rows into an AI chat with this prompt (below). You’ll get immediate patterns and one change to try next week.

Thanks — that question about whether AI can analyze time-tracking data is exactly the right one. Yes, and it’s less mystical than you think. AI helps spot patterns and suggest practical tweaks, but you control the priorities and judgement.

What you’ll need

• A recent export of your time-tracking (CSV or Excel)
• Columns: date, project/client, task, duration (minutes or hours), billable (yes/no), notes
• Access to an AI chat tool (copy-paste works fine)

Step-by-step: use AI to find quick wins

1. Export one week of data. Keep it to 50 rows or less for the first run.
2. Open an AI chat and paste a 10-row sample plus this ready-made prompt (copy below).
3. Ask the AI for: top 3 time drains, 2 tasks to delegate or automate, and a 1-week experiment to reclaim 3–5 hours.
4. Pick one experiment and schedule it on your calendar as a non-negotiable block.
5. Run the experiment for a week, then re-export and repeat the analysis.

Example (what to expect)

• AI might identify frequent short meetings as a time drain and suggest batching them into two blocks per week.
• It may spot recurring admin tasks that can be automated (invoicing, calendar invites).
• It will propose a simple experiment like “Reduce meeting time by 25% and move two meetings to email.”

Common mistakes & fixes

• Too much raw data: sample 1 week first, then scale up.
• Vague categories: standardize task names (e.g., “Email” not “Misc”).
• Privacy worry: anonymize client names before sharing with AI.

Action plan (next 7 days)

1. Day 1: Export 1 week and run the AI prompt below.
2. Day 2: Pick one AI-suggested experiment; block time on your calendar.
3. Days 3–7: Follow the experiment and note results each day.
4. End of week: Re-run the AI with updated data and iterate.

Copy-paste AI prompt (use as-is)

Here are 10 rows of my time-tracking (columns: date, project, task, duration_hours, billable, notes). Please analyze and give me: 1) top 3 patterns or time drains, 2) two practical tasks to delegate or automate, 3) one 7-day experiment that can reclaim 3–5 hours, and 4) a simple metric to track to see if it worked. Be specific and action-focused.

Keep it simple: AI helps you find options — you decide which to try.

Nice — asking whether AI can both analyze performance and suggest copy improvements is exactly the right question. Practical AI can do this fast, but it works best when you give it the right data and then test its suggestions.

Quick yes/no checklist — do / don’t

• Do provide the post text, platform, and clear performance metrics (impressions, CTR, likes, comments, shares, saves).
• Do tell the AI about your audience and the desired outcome (brand awareness, signups, sales).
• Do A/B test at least 2 variations and measure for 3–7 days.
• Don’t ask AI to “fix” strategy without hard numbers — it advises, you validate.
• Don’t rely on one suggestion — iterate quickly and learn.

What you’ll need

• The post text (original).
• Key metrics for the last 7–30 days (impressions, CTR, engagements).
• Platform (LinkedIn, X/Twitter, Facebook, Instagram, email subject).
• Your audience profile and goal.

Step-by-step: how to use AI for analysis and copy improvements

1. Collect the data above and paste into the AI prompt (see example prompt below).
2. Ask the AI for: headline variants, opening lines, CTA options, recommended length, hashtag suggestions, and an A/B test plan.
3. Pick 2–3 AI suggestions and create simple variations.
4. Run an A/B test on the platform (same audience, same time window).
5. Measure results and feed them back into the AI for the next round.

Copy-paste AI prompt (use as-is)

Analyze the following social post and its performance. Post: “{PASTE YOUR POST HERE}” Platform: {LinkedIn / X / Facebook / Instagram}. Metrics: impressions {number}, CTR {number%}, likes {number}, comments {number}, shares {number}. Audience: {brief}. Goal: {awareness / clicks / signups / sales}. Provide: 1) three headline/hook alternatives (brief, attention-grabbing); 2) three opening sentence variations; 3) two CTAs; 4) recommended post length and structure; 5) five relevant hashtags or keywords; 6) a simple A/B test plan with expected KPI improvements and why each change should help.

Worked example (short)

Original post: “5 tips to grow your email list fast.” Metrics: impressions 6,000; CTR 0.6%; likes 10; comments 1. Audience: small biz owners; Goal: clicks to landing page.

AI suggestions (example):

• Hooks: “Stop wasting time — grow your list with these 3 proven tweaks”, “How I added 1,000 emails in 30 days (no ads)”, “The quick checklist your signup page is missing”
• Openers: Start with a surprising stat, a short story, or a single-question lead that targets pain.
• CTAs: “Download the 1-page checklist” or “Try tip #2 and tell me the result”
• Structure: 1-sentence hook, 3 short bullets with one example each, CTA, 3 hashtags.

Mistakes to avoid & quick fixes

• Don’t write long blocks — break into short lines for social feeds. Fix: use bullets or short sentences.
• Don’t skip the CTA. Fix: be explicit and use action words.
• Don’t ignore platform norms (hashtags on Instagram, short hooks on X). Fix: tailor length and format.

7-day action plan

1. Day 1: Gather one post + metrics and run the AI prompt.
2. Day 2: Create 2–3 AI-inspired variations.
3. Days 3–6: Run A/B test and monitor results.
4. Day 7: Pick the winner, apply learnings to next post, repeat.

AI won’t replace your judgement but it will speed up testing and suggest high-probability improvements. Small, regular experiments win — aim for steady lifts, not overnight miracles.

Quick win: In under 5 minutes paste your single target keyword into the prompt below and ask for a concise SEO brief — you’ll get a one-page outline a writer can use straight away.

Nice point about testing two title variants — that small A/B is one of the fastest ways to lift CTR. I’ll add a compact, practical workflow you can run now and a ready-to-use prompt to copy-paste.

What you’ll need:

• Your single target keyword (one phrase).
• One top competitor URL (the page you want to beat).
• One-sentence audience description (who they are and what they need).
• Desired CTA (lead form, download, sign-up, purchase).

Step-by-step (do this now):

1. Open your AI tool and paste the prompt below. Replace placeholders: [KEYWORD], [COMPETITOR_URL], [AUDIENCE], [CTA].
2. Ask for two outputs: a “one-page writer brief” and a “publish checklist”.
3. Scan the brief, pick two title variants (benefit-driven + question-driven). Save both for a 7–14 day CTR test.
4. Edit tone/CTA to match your brand. Hand the one-page brief to your writer or draft directly.
5. Publish, submit to index, and monitor ranking + CTR for one week. Adjust title if CTR is low.

Copy-paste AI prompt (use as-is)

“Create a concise SEO brief for the target keyword: [KEYWORD]. Competitor: [COMPETITOR_URL]. Audience summary: [AUDIENCE]. Desired CTA: [CTA]. Include: 1) Best page title (<=60 chars) and meta description (<=155 chars) optimized for CTR; 2) Primary search intent and one-sentence user problem to solve; 3) H1 and H2/H3 outline with approx. word counts per section and total suggested word count; 4) 5 semantic/LSI keywords; 5) 5 FAQs to include as on-page Q&A; 6) 3 suggested internal links (anchor text + page type) and one backlink target idea; 7) 3 quick optimization notes (readability, schema type, image guidance). Keep it short and actionable.”

Example (what you’ll get):

• Title: Best Ergonomic Office Chairs 2025 (<=60 chars)
• Meta: Find comfortable, supportive ergonomic chairs with buying tips & budget picks. (<=155 chars)
• Outline: H1, H2: Top picks (600 words), How to choose (400), Setup tips (300), FAQs (400) — Total 1,700 words.
• Semantic keywords: lumbar support, adjustable armrests, seat depth, office posture, warranty.

Common mistakes & fixes:

• Brief misses user intent — Fix: force the AI to state the primary user problem and intent in the brief.
• Vague headings — Fix: require approximate word counts per H2/H3 so the writer knows depth.
• No publish checklist — Fix: ask for a separate “publish checklist” including schema, image alt text, and internal links.

1-week action plan:

1. Day 1: Run prompt, finalize brief, pick 2 titles.
2. Days 2–3: Draft using H2s, FAQs, schema note.
3. Day 4: On-page check (meta, headings, internal links, images).
4. Day 5: Publish & submit to index; start outreach for 3 backlink targets.
5. Days 6–7: Monitor CTR and ranking; swap title if CTR is under target.

Small, practical steps beat big plans. Want to try one keyword now? Paste it and I’ll shape the brief with you.

— Jeff

Let’s make this boringly reliable. You’ve got the core right: small schema, three-step chain, human review. Now let’s upgrade it to a repeatable, 20-minute weekly routine that keeps improving itself.

Why this works: a tight schema + a short chain + a simple review lane gives you consistency. Add a calibration set and a tiny dictionary, and accuracy climbs without changing how you take notes.

What you’ll need (keep it light):

• 30 representative notes (typed or OCR).
• A locked 6-field schema: Type, Date, VendorOrTopic, Amount, Currency, ActionNeeded (+ ActionText and Confidence if helpful).
• A spreadsheet with columns that match your schema plus ReviewNotes and Reviewed (true/false).
• A living vendor/topic dictionary (two columns: RawName, CanonicalName).
• 10 corrected examples to use as a mini calibration set (you’ll build this on day one).

Your upgraded chain (4 tiny steps)

1. Split multi-topic notes into single-topic chunks (1 chunk = 1 record).
2. Classify each chunk (Receipt, Meeting, Invoice, Idea).
3. Extract & normalize into your locked fields (no guessing; standard formats).
4. Verify critical fields (date, amount, currency), set Confidence, and add a short Reason if Confidence < 80.

Insider tricks that stabilize accuracy fast

• NDJSON output: ask for one JSON object per line. It pastes cleanly into sheets and imports easily.
• Dictionary assist: pass a small list of common vendors/topics and map near matches (spacing/case/punctuation differences allowed).
• No-guess rule: blanks + lower Confidence beat wrong data. Review only the few uncertain rows.
• Stopwords for VendorOrTopic: ignore terms like Inc, LLC, Ltd, The when matching.

Copy-paste prompts (use as-is)

1) Splitter (multi-topic notes → chunks)

Break this note into the smallest useful single-topic chunks. Don’t rewrite content; just trim obvious noise. Classify each chunk as one of [Receipt, Meeting, Invoice, Idea]. Return ONLY JSON with an array named Items. Example format: {“Items”:[{“Text”:”…”,”SuggestedType”:”Receipt”}]} Note: keep original order. If no clear chunks, return {“Items”:[]} Note: here is the note: “{paste full note here}”

2) Extractor (one chunk → one record, NDJSON)

You extract fields to a locked schema. Return ONE JSON object on ONE line (NDJSON) with keys exactly: Type, Date, VendorOrTopic, Amount, Currency, ActionNeeded, ActionText, Confidence. Rules: 1) Type = one of [Receipt, Meeting, Invoice, Idea] (use the suggested type if provided). 2) Date = YYYY-MM-DD or “” if ambiguous. 3) Amount = number (no commas) or 0 if not present. 4) Currency = 3-letter code (USD/EUR/GBP etc) or “”. 5) ActionNeeded = true/false. 6) ActionText max 120 chars. 7) No guessing: if unsure, leave field blank and set Confidence < 80. 8) VendorOrTopic: map to the closest CanonicalName from the dictionary when similar (ignore case/spacing/punctuation and common suffixes like Inc, LLC, Ltd, The); if no good match, use the best clean literal from the text. Input chunk: “{paste chunk text here}” Optional dictionary (RawName→CanonicalName, small list): {paste pairs here}

3) Verifier (sanity check)

Compare the JSON to the original chunk. If Date, Amount, or Currency is missing, ambiguous, or unlikely, set Confidence under 80 and add a short Reason field (max 80 chars). If all good, keep Reason out. Prefer YYYY-MM-DD. If a numeric date is ambiguous (e.g., 4/7/25 could be MM/DD or DD/MM), leave Date blank and lower Confidence. Return ONLY the updated JSON on one line. Original: “{paste chunk text here}” JSON: {paste JSON here}

Optional: one-shot batch (paste multiple chunks)

Process each chunk separately and return one JSON object per line (NDJSON). Apply the same schema and rules as above. If a chunk is not actionable, still return a JSON line with blanks and lower Confidence.

Worked example (multi-topic note → 2 records)

Raw note: “4/7/25 Lunch w/ client at Joes Diner — $42.10 — follow up on proposal. Also: Invoice 1782 from ACME C0rp due 12-3-24 for EUR 3,950 — schedule remittance.”

• Splitter finds two chunks: a Receipt and an Invoice.
• Extractor (with dictionary mapping Joes Diner → Joe’s Diner, ACME C0rp → ACME Corp) returns two NDJSON lines:

• {“Type”:”Receipt”,”Date”:”2025-04-07″,”VendorOrTopic”:”Joe’s Diner”,”Amount”:42.10,”Currency”:”USD”,”ActionNeeded”:true,”ActionText”:”Follow up on proposal”,”Confidence”:90}
• {“Type”:”Invoice”,”Date”:”2024-12-03″,”VendorOrTopic”:”ACME Corp”,”Amount”:3950,”Currency”:”EUR”,”ActionNeeded”:true,”ActionText”:”Schedule remittance”,”Confidence”:86}

Common mistakes & quick fixes

• Ambiguous dates (4/7/25): don’t guess. Leave Date blank, set Confidence < 80, and add a Reason. Fix in review.
• Currency drift: require a 3-letter code. If missing, leave blank and lower Confidence. Add common defaults to your dictionary notes.
• Vendor variations: feed the dictionary, ignore Inc/LLC/Ltd, and map near matches. Add new canonical names when you correct rows.
• Overfitting the prompt: keep schema stable for two weeks before adding new fields.
• Schema breakage: insist on “one JSON object on one line; no extra text.” If it breaks, re-run just those rows.

What to expect

• Iteration 1: 70–85% field accuracy; 25–40% of rows need review.
• Iteration 2–3 (with dictionary + 10 calibration examples): 88–93% accuracy; review falls below 20%.
• Steady state: 50 notes in 15–25 minutes weekly, with a short review pass.

60-minute setup sprint

1. Lock your schema and create sheet columns to match.
2. Pick 30 notes; run the Splitter; paste chunks into a new tab.
3. Run Extractor → Verifier on 15 chunks; paste NDJSON lines into the sheet.
4. Filter Confidence < 80; correct those rows; add 10 corrected cases to your calibration set.
5. Build or update the vendor/topic dictionary with names you corrected.

Weekly routine (20–30 minutes)

1. Run Splitter → Extractor → Verifier on the week’s notes.
2. Paste NDJSON into your sheet; filter Confidence < 80.
3. Correct flagged rows; add new canonical names to the dictionary.
4. Save 3–5 fresh corrected examples to keep your calibration set current.

Final nudge: Don’t chase perfection. Lock the schema, enforce the no-guess rule, and make tiny weekly improvements to your dictionary and calibration set. Predictable beats perfect — and it compounds.

Great call-out: locking attribution and running a 10–20% test first is the difference between confident scaling and expensive guesswork. Let’s add the guardrails and operating rhythm that turn an AI plan into reliable results.

Big idea: pair AI’s speed with simple rules — learning budgets, pacing, and weekly reallocation — so your plan survives the real world.

What you’ll bring

• Goal and target (CPA or ROAS)
• Total budget and a 15% test slice over 21–30 days
• Channels you’re open to (search, social, video, display, email/CRM)
• Recent benchmarks if you have them (CPM, CPC, CVR, CPA)
• Attribution choice (start with last-click if unsure)

The playbook — step by step

1. Pick channels that fit your goal.
   • Awareness: Video/Display 50–60%, Social 25–35%, Search 10–20%.
   • Leads (B2B/B2C): Search 40–50%, Social 25–35%, LinkedIn (B2B) 10–20%, Retargeting 10–15%.
   • Sales (ecom): Search & Shopping 35–45%, Meta 30–40%, Retargeting 10–15%, Video 5–10%.
2. Set learning budgets per channel. Simple rule: spend enough to hit 20 conversions in the test window. Minimum test spend ≈ Target CPA × 20 per channel. If that’s too high, test fewer channels now and add later.
3. Add guardrails before you launch.
   • Daily pacing: about 1/30 of monthly budget per day, allow ±20% wiggle room.
   • Bid targets: use tCPA or manual bid caps aligned to your target CPA/ROAS.
   • Frequency caps (video/display): 2–3/day to avoid fatigue.
   • Creative rotation: 3–5 active variants per channel; pause any with CTR in the bottom 25% after 3–5 days.
   • Search hygiene: separate branded vs non-branded; don’t let brand mask generic performance.
   • Tracking: identical conversion definitions and UTMs across all channels.
4. Run a 15% budget test for 21–30 days. Expect 5–7 days of “learning.” Judge early on leading indicators (CPM, CTR, CPC); judge scaling after you have conversion volume.
5. Use the Budget Thermostat (weekly). Move money gently:
   • If a channel’s CPA is ≤ target and has 20+ conversions, shift +10–15% into it.
   • If CPA is > target by 20%+ after 20 conversions, shift −10–15% out (or fix creative/targeting first).
   • Never move more than 20% of total budget in a single week. Stability beats whiplash.
6. Pressure-test with scenarios. Ask AI for best/base/worst cases (±20% on CPC/CVR). You’ll see how fragile or robust your plan is before you spend.

Copy-paste prompts you can use today

1) Build the test plan with guardrails

“I have a total budget of $[TOTAL_BUDGET] over [TIME_FRAME] with the goal of [GOAL]. Assume attribution = [LAST-CLICK/TIME-DECAY/DATA-DRIVEN]. Target = [TARGET_CPA or TARGET_ROAS]. Channels to consider: [LIST_CHANNELS]. Recent benchmarks: CPM [X], CPC [Y], CVR [Z%], CPA [W] (fill blanks if needed). Please create a 15% test plan for [21–30] days that includes: 1) Channel allocations with % and $; 2) Expected ranges for CPM, CPC, CTR, CVR, CPA per channel; 3) Learning budget minimums using ‘20 conversions per channel’; 4) Guardrails (daily pacing, bid/tCPA, frequency caps, creative rotation); 5) Three scenarios (best/base/worst at ±20% on CPC and CVR) with expected conversions and CPA. Return totals that match the test budget and list assumptions clearly as bullet points.”

2) Week-1 recalibration

“Here are my week-1 results by channel: [CHANNEL: Spend, Impressions, Clicks, CTR, CPC, Conversions, CVR, CPA]. Target CPA = [X]. Apply the Budget Thermostat: increase up to 15% for channels at/under target with ≥[20] conversions; decrease up to 15% for channels 20% above target; keep minimum learning budgets intact. Provide a revised 2-week plan with new allocations, hypotheses to test, which creatives to pause/scale, and a simple stop-loss rule per channel.”

3) Creative angles that match the funnel

“Based on a [GOAL] campaign for [AUDIENCE] with [PRODUCT/OFFER], give me 5 ad angles per channel (Search, Meta, LinkedIn, Display/Video) aligned to Top/Mid/Bottom funnel. For each angle, provide: primary text, headline, CTA, and the key objection it tackles. Keep copy tight and suggest 2 variations per angle for testing.”

Example (simple numbers)

• Budget: $10,000 over 30 days. Goal: leads. Target CPA: $100.
• Test: 15% = $1,500 over 21 days.
• AI proposes: Search 50% ($750), Meta 30% ($450), LinkedIn 15% ($225), Retargeting 5% ($75).
• Learning budgets check: each channel aims for 20 leads → $2,000 ideal; we’re below that, so we accept directional learning and plan a second wave focusing on top performers.
• Guardrails: daily pacing ≈ $50, frequency caps 2/day on retargeting, 4 search ads per ad group, pause creatives if CTR is bottom quartile after day 5.
• Thermostat at week 2: Meta hits $95 CPA with 22 leads → +10%; LinkedIn at $140 with 12 leads → −10% and refresh creative; Search at $105 with 28 leads → hold and tighten negatives.

Frequent pitfalls and quick fixes

• Scaling too fast during learning — Fix: wait for ~20 conversions/channel or 14 days before big shifts.
• Audience fatigue — Fix: cap frequency, rotate creatives weekly, expand lookalikes/interests.
• Mixed search intent — Fix: split brand vs non-brand; add negatives early.
• Double counting retargeting — Fix: exclude recent purchasers/leads across platforms.
• One-size-fits-all creative — Fix: map copy to funnel stage; bottom-funnel = proof and offer.

7-day operating cadence

1. Day 1: Run the test-plan prompt. Sanity-check assumptions and totals. Launch with guardrails.
2. Days 2–3: Check CPM, CTR, CPC. Kill obvious underperforming creatives. Keep budgets steady.
3. Day 5: First creative refresh on any ad below median CTR. Add negatives in search.
4. Day 7: If any channel has 20+ conversions, apply the Thermostat. If not, wait to week 2.

Bottom line: AI gives you a fast, testable plan. Your edge is the discipline — learning budgets, guardrails, and a weekly reallocation rule. Start small, measure cleanly, and let the data (not guesswork) decide where the next dollar goes.

Nice point — keeping the template tiny is the quickest win. That little change alone reduces errors and makes review manageable. Here’s a practical extension: a short, reliable prompt-chain that handles messy OCR, classifies, extracts, normalizes and flags uncertainty for human review.

What you’ll need:

1. 20–50 example notes (scanned or typed).
2. A 3–6 field template (start tiny: Date, Topic/Vendor, Amount, ActionNeeded).
3. AI tool that accepts prompts and returns JSON (or text you can paste into a sheet).
4. A spreadsheet for import and a review column for flagged rows.

Step-by-step prompt-chain (do this in order):

1. Pre-clean — run simple OCR fixes: normalize common OCR errors (O vs 0, l vs 1) and convert smart quotes. This reduces garbage early.
2. Step 1 — Classify — ask the AI to label the chunk type (receipt, meeting, invoice, note). Use this to pick the right extraction template.
3. Step 2 — Extract & Normalize — extract only the template fields, force date to YYYY-MM-DD, amount to numeric and currency as a three-letter code, limit ActionText to 120 chars.
4. Step 3 — Verify & Flag — run a short verification prompt that checks for missing or unlikely values and returns Confidence (0–100) and a Flag if Confidence < 80.
5. Human review — open the flagged rows in your sheet, correct, and add corrected examples back to your sample set.

Copy-paste AI prompt — Extraction (use as-is)

Classify this note and extract the following fields. Return only valid JSON with keys: Type, Date (YYYY-MM-DD or blank), VendorOrTopic, Amount (numeric or 0), Currency (3-letter code or blank), ActionNeeded (true/false), ActionText (120 chars max), Confidence (0-100). Normalize dates to YYYY-MM-DD, amounts to numbers. If uncertain, set Confidence under 80 and leave ambiguous fields blank. Note: return only JSON. Here is the note: “{paste note here}”

Follow-up verification prompt (optional)

Check this JSON against the original note. If any Date, Amount, or Currency seems wrong or missing, change Confidence to a value under 80 and add a short Reason field explaining why. Return only the updated JSON.

Worked example

Raw note: “4/7/25 Lunch with client at Joe’s Diner — $42.10 — follow up on proposal.”

Expected extraction (example): {“Type”:”Receipt”,”Date”:”2025-04-07″,”VendorOrTopic”:”Joe’s Diner”,”Amount”:42.10,”Currency”:”USD”,”ActionNeeded”:true,”ActionText”:”Follow up on proposal”,”Confidence”:92}

Common mistakes & fixes:

• Wrong dates — include many date examples in different formats in your sample set and force YYYY-MM-DD in the prompt.
• Currency missing — add a rule: default to your local currency if none found and mark Confidence lower.
• OCR noise — run a pre-clean with simple replacements before sending to the AI.

1-week action plan (quick wins)

1. Day 1: Pick note type and collect 20 examples.
2. Day 2: Pre-clean and split into chunks.
3. Day 3: Run classify + extract on 20 and import to sheet.
4. Day 4: Review flagged rows, correct, add 10 corrected examples back.
5. Day 5: Re-run on another 20, measure accuracy.
6. Day 6: Tweak prompts (thresholds, length limits).
7. Day 7: Schedule weekly batch and a 20–30 minute review slot.

Final reminder: start tiny, iterate fast, and keep a short review habit. Small weekly wins compound quickly and give you structured data you can actually use.