Different deliverables have different failure modes. A report fails differently than an email, which fails differently than a landing page. This lesson provides a pre-built checklist for each common type you'll ship.

Client performance report

☐ Data accuracy - 3-5 metrics spot-checked against source platforms

☐ Date ranges match what was requested (not default)

☐ Comparison periods correct (vs last period + YoY if relevant)

☐ Client name correct throughout - no leftover references to other clients

☐ Brand correct - colours, logo, fonts match the brand guide

☐ Recommendations make sense for THIS client's situation

☐ Recommendations are actionable (not "improve performance")

☐ Math checks - ROAS = revenue/spend, percentages add up, totals match breakdowns

☐ Mobile responsive - open on phone and read through

☐ Encrypted if client-facing (password protected URL)

☐ Charts render correctly (no broken images)

Audit report

☐ All expected platforms included (no channel missed)

☐ Findings are real - cross-reference against the actual platform

☐ Recommendations are specific, not generic ("increase budget" is not a recommendation)

☐ Scoring (if any) matches the findings (a "red" finding gets a low score)

☐ Competitor data verified - URLs + company names correct

☐ No false positives (flagging issues that aren't actually issues)

☐ No false negatives (missing issues that ARE issues)

☐ Next steps are prioritised (most impactful first)

Landing page / HTML deliverable

☐ Every link works - click each one

☐ Every button + CTA works

☐ Form submits (test with fake data if needed)

☐ Copy sounds like the client's voice (not generic AI voice)

☐ Claims are accurate (no fabricated statistics or testimonials)

☐ Images load + are the right images

☐ Image resolution is adequate (no pixelated hero images)

☐ Mobile responsive - check at 375px, 768px, 1440px

☐ Brand correct - right logo, right colours, right fonts

☐ No Webprofits branding on client pages (or vice versa)

☐ Page loads in under 3 seconds ☐ Tracking / analytics installed if expected

Email (copy only - not sending)

☐ Subject line accurate to the content

☐ Preview text complements subject (not duplicate)

☐ Tone matches client voice guide

☐ No fabricated social proof, stats, or testimonials

☐ Links work + go to the right destination

☐ Personalisation placeholders render correctly ({{first_name}} etc)

☐ Sender name + reply-to correct

☐ Unsubscribe link present (legally required)

☐ Images have alt text

☐ Mobile preview - does it look good on phone?

☐ Dark mode preview (some clients have weird dark-mode rendering)

Email (when triggering a send) - extra paranoid

On top of the copy review

☐ Confirmed sending to the right segment (not all subscribers)

☐ Confirmed NOT a test segment rendering (you're about to send for real)

☐ Confirmed send time is correct timezone

☐ Confirmed NOT in the middle of a broken campaign (if something's wrong, hitting send compounds the damage) ☐ Claude's permission prompt explicitly shows "send to X people" - confirm the number is right

☐ Budget / value estimate - if this is a revenue email, you're okay if it underperforms

Read every field carefully. Sends can't be undone.

Proposal

☐ Client name + details correct throughout

☐ No copy-paste from another proposal (references to wrong clients)

☐ Pricing matches what was agreed

☐ Every number in the pricing checks out

☐ Scope matches discussed scope (not scope creep + not reduced)

☐ Contact details correct - names, emails, phone numbers

☐ Signature block correct

☐ Terms + conditions appropriate for this engagement

☐ Timeline realistic

☐ No placeholder text left in (lorem ipsum, "client name here")

Ad creative (before launch)

☐ Ad copy matches client brand voice

☐ Claims are accurate (no fabricated data)

☐ Images are the right images + right resolution

☐ Call to action aligned with campaign goal

☐ Landing page URL correct + works ☐ UTM parameters correct for tracking

☐ Targeting appropriate (not too narrow, not too broad)

☐ Budget + schedule set correctly

☐ Preview on mobile + desktop

☐ Legal review if regulated industry

☐ Client approval obtained (if that's the process)

Using Claude to run the checklist

Instead of running checklists manually

Apply the [report / audit / landing page / email / proposal / ad] QA checklist to this deliverable. Go through each item, flag anything that doesn't pass, and show me the specific issue.

Claude reviews against the checklist. Finds stuff you might miss. You still do the final eyeballs - Claude's review catches obvious issues, yours catches subtle ones.

Building checklists into skills

Every time you build a skill that produces a deliverable type, include a verification checklist in the skill:

[Report type] Skill

Output

[How to build the report]

Verification (run before delivery)

• Data spot-check - 3 numbers traced to source
• Date range confirmed
• Client name correct throughout
• Brand compliance
• Mobile check

Every invocation of the skill ends with the checklist. Makes QA automatic.

Action items

☐ Save these checklists somewhere you'll reference - ideally as a skill or a file in _shared/

☐ Apply the right checklist to your next deliverable

☐ Add verification sections to every skill you build that produces deliverables

☐ Be extra paranoid on emails + ad launches - can't be undone

Next lesson: Using Claude to QA itself.

Exercises

1. Review the concepts covered in this lesson: Checklists per deliverable.
2. Write down your key takeaway from this lesson.
3. Practice running any commands or prompts mentioned above inside your terminal.