To QA AI client work without redoing it, check the work against written client rules, not your gut. Add 1 review step before anything ships. The reviewer scores a short checklist, fixes small misses, and sends big misses back with the rule that broke. Then you add that rule to the client file, so the miss never repeats.
Redoing Is a Sign the Rules Were Never Written
When a reviewer rewrites AI work, they are not fixing the AI. They are applying rules that only live in their head.
"This client hates exclamation points." "We never mention price in ads." "The CEO wants the case study first." None of that was in the brief. So the AI could not follow it. The reviewer knew it, so the reviewer rewrote it.
That is the trap. Review turns into the real work. The AI saves an hour on the draft, and your senior person spends that hour on the rewrite. Net gain is zero.
The lesson is simple. A rewrite is a missing rule in disguise. Do this. Pull the last 5 rewrites your team made. Write down the rule behind each edit.
Why More Review Makes It Worse
The usual fix is more eyes. Add a second reviewer. Have the director check everything. It feels safe.
It backfires. Each reviewer brings their own taste. Feedback starts to conflict. Your best people become the bottleneck, and work stacks up behind them.
This is not your team's fault. AI tools start blank on every task. Your reviewers are doing the job a brief should have done. They are filling the gap with memory and instinct, and that does not scale past a handful of accounts.
The answer is not more review. It is better review. Do this. Count how many people touch 1 piece of client work before it ships. If it is more than 2, you have a loop problem.
The Approval Loop in 5 Steps
A good loop has 1 job. Catch misses fast and make sure they never come back.
- Write the rules. Each client gets 1 file with goals, voice, banned claims, audience and who approves what. No rule, no fair review.
- Generate with the rules attached. The AI gets the full client file every time. Nobody pastes context from memory.
- Review against the checklist. The reviewer scores the work against the 8 checks below. Taste is not on the list.
- Route the result. Pass it, fix it, or send it back. If a fix takes less time than explaining it, fix it. If not, send it back with the exact rule that broke.
- Feed the rule back. Every send-back adds or sharpens a line in the client file. The next draft starts smarter.
Step 5 is the one most agencies skip. It is also the one that makes the loop shrink. Without it, you review the same miss forever.
Do this. Pick 1 client and run all 5 steps on the next deliverable. Time the review before and after.
The Checklist
Your reviewer should not ask "Do I like it?" They should ask 8 plain questions. Each one ties to a line in the client file.
| Check | The question | If it fails |
|---|---|---|
| Goal | Does this serve the client's current goal? | Send back. The goal in the file may be stale. |
| Facts | Is every number, name and date true and sourced? | Fix if small. Send back if the AI invented anything. |
| Voice | Does it sound like the client, per their voice rules? | Fix small slips. Add a rule if it repeats. |
| Claims | Does it avoid every banned claim and legal risk? | Send back. Add the claim to the banned list. |
| Audience | Is it written for the right buyer at the right stage? | Send back with the audience line that was missed. |
| Format | Does it match the spec, length and channel? | Fix it. Add the spec to the file. |
| Links | Are links, tracking and calls to action correct? | Fix it. Check the template. |
| Approval | Does anything here need client sign-off first? | Hold it. Route to the named approver. |
Keep the checklist short on purpose. 8 checks can be scored in minutes. A 30-point list gets skipped by Friday.
Do this. Print the table. Use it on the next 10 pieces of client work. Note which check fails most.
Who Reviews What
Not every piece of work needs the same review. Match the review to the risk.
- Low risk. Internal drafts, research and first outlines. The operator spot checks and moves on.
- Client facing. Posts, emails, reports and ad copy. The operator runs the full checklist before anything goes out.
- High risk. Anything about money, legal claims, pricing or a client's own customers. The operator checks it, then the named client approver signs off.
This is how 1 person carries more accounts without lowering the bar. They spend real review time on the work that can hurt a client. They do not spend it on a first outline.
The rule. Review effort follows risk, not seniority. Do this. Tag your next 20 deliverables low, client facing or high risk. You will see where review time is wasted.
Make the Loop Shrink Over Time
A good approval loop gets faster every month. A bad one stays the same size.
Track 1 thing. The reason for every send-back. When the same reason shows up twice, a rule is missing from the client file. Write it. Then that miss stops.
This is the core of Portable Delivery Intelligence. Each client's strategy, rules, priorities and approvals live in 1 place and connect to the AI tools your team already uses. Every fix from review goes back into that place. So the next draft from ChatGPT, Claude or Gemini starts with the lesson already learned.
That is what moves capacity. Most agencies cap out at 4 to 8 accounts per account manager. Our per-operator target with the system is 18 to 25. Here is how we think about agency capacity.
We run our own delivery this way. In our internal mapping, 1 operator plus the system carries the work a 25-person marketing team used to carry. You can see the proof here.
Do this. Start a send-back log today. 3 columns. Client, check that failed, rule added.
FAQ
How do you QA AI-generated client work?
Check it against written client rules with a short checklist. Fix small misses, send big ones back with the rule that broke, and add that rule to the client file.
Who should review AI work in an agency?
The operator who owns the account reviews everything client facing. High-risk work like pricing or legal claims also goes to a named client approver.
How do we stop rewriting AI drafts?
Write down the rule behind every rewrite and add it to the client file. Rewrites happen when rules live in someone's head instead of the brief.
Does Portable Delivery Intelligence replace the review step?
No. It makes review faster. Each client's rules and approvals sit in 1 place, so the AI starts with them and the reviewer checks against them.
Bring 1 client account. We map where your team's time leaks and how many more accounts each person could carry. You keep the map either way. Book your map here.
