Reviewing AI Output Before It Ships
Armbrain and Claude draft things. Your job is quality control before it reaches the client. The CMO trusts that anything you approve is ready to send.
Why This Role Exists
AI is fast. It can draft an email in seconds, pull together a meeting brief in under a minute, and generate a campaign summary before your coffee gets cold. But speed without judgment is dangerous.
AI makes confident-sounding mistakes. It references meetings that didn't happen. It drifts away from a client's brand voice over time. It fills gaps in its knowledge with plausible-sounding filler. Your job is to catch all of that before anyone outside the team sees it.
Common AI Mistakes to Watch For
| Mistake | What It Looks Like | How to Catch It |
|---|---|---|
| Made-up details | "As we discussed in last Tuesday's call..." (there was no call) | Cross-check dates against the calendar |
| Wrong tone | Too formal for a casual client, too casual for a corporate one | Read it out loud - does it sound like your CMO talking to this person? |
| Generic filler | "We're excited to leverage synergies to drive results" | If it sounds like it could be about any client, it's filler |
| Outdated info | References an old campaign, a former team member, or last quarter's numbers | Check against recent meeting notes and Armbrain |
| Hallucinated metrics | "Your traffic increased 23% last month" (it didn't) | Verify any specific numbers against actual data |
| Name confusion | Using the wrong stakeholder name or mixing up client details | Confirm names and roles against the client mind |
The "Would the CMO Actually Say This?" Test
Before approving anything, read it through this filter: if your CMO walked into a room and said these exact words to this client, would it sound natural?
If the answer is no, something needs to change. Common failures:
- Too stiff. Your CMO calls the client "hey Sarah" and the draft says "Dear Ms. Chen." Fix the register.
- Too chummy. The client is formal and the draft opens with "What's up, team!" Match the relationship.
- Too long. Your CMO sends three-sentence emails. The draft is four paragraphs. Cut it.
- Wrong priorities. The draft leads with metrics when this client cares about creative direction. Reorder.
You know your CMO's voice because you hear it every day. Trust that knowledge.
Brand Voice Drift
This is subtle and dangerous. When AI writes for a client over weeks and months, it slowly drifts away from the client's actual brand voice. It starts sounding more like "generic marketing AI" and less like the client.
Watch for:
- The client uses casual language but the AI keeps slipping in corporate jargon
- The client's brand is warm and personal but the AI produces distant, third-person copy
- Industry-specific terms get replaced with generic equivalents
- The client's preferred phrases ("our community" vs "our customers") get swapped
When you notice drift, pull the brand voice to compare:
"Switch to Acme Corp""Show me Acme's brand voice"
Compare the stored voice guidelines against what the AI just produced. If they don't match, flag it. Over time, you'll develop an instinct for each client's voice and catch drift before it compounds.
What to Check in Each Type of Output
Email drafts:
- Correct recipient name and title
- Tone matches the relationship
- Any referenced meetings or calls actually happened
- Action items are accurate and attributed to the right person
- Attachments mentioned actually exist
Meeting briefs:
- Attendee list is current (nobody who left the company, nobody missing)
- Open items are still open (not already resolved)
- Background context is accurate, not fabricated
- Recommendations align with what the CMO has actually been discussing
Client reports:
- All metrics are verified against source data
- Comparisons reference the correct time periods
- Recommendations follow logically from the data shown
- Client-specific terminology is used correctly
Content drafts:
- Brand voice alignment (check against stored guidelines)
- Facts and statistics are verifiable
- No placeholder text left in ("insert X here", "[TBD]")
- Formatting matches what the client expects
When to Fix It Yourself vs. Flag It
Fix it yourself:
- Typos, grammar, formatting issues
- Minor tone adjustments (too formal, too casual)
- Obvious factual corrections (wrong date, misspelled name)
- Removing generic filler and tightening language
Flag for the CMO:
- Strategic recommendations you're not sure about
- References to conversations or decisions you can't verify
- Significant tone mismatches that might reflect a relationship nuance you don't know about
- Anything where you're not confident in the correction
When you flag something, be specific: "The draft references a pricing discussion from March 15 but I don't see that on the calendar. Did this happen in a call I don't have notes for, or did the AI make it up?"
Building the Trust Loop
The goal is a simple workflow: AI drafts, you review, CMO sends. Over time, the CMO should be able to trust that if something made it past you, it's ready. That trust is earned by catching things early and being honest when you're not sure.
Store patterns you notice so the system improves:
"Quick note: AI keeps using 'leverage' in Acme drafts. Sarah has said twice she hates that word. Flag for brand voice update."Related Guides
- End of Week: Building the Weekly Summary - reviewing the weekly summary before it goes out
- Tracking Action Items Across All Clients - verifying action items referenced in drafts
- Prepping Your CMO for Meetings - reviewing meeting briefs before the CMO sees them