By the time content feels generic, it's usually been drifting for months, and the fix isn't a better prompt for tomorrow's post. It's a sequence: find out how bad it is, rebuild the system, prove the fix, then relaunch. This checklist is that sequence, ten steps over two weeks, built for the Iteration pillar. Run the Voice Audit Checklist as step 1 below, and once you're stable, keep drift from coming back with the Content Testing Framework.
Each step has a time estimate and a done-criteria: a specific thing that's true when the step is actually finished, not just attempted.
Week 1: Audit and rebuild
Step 1: Run a full voice audit. (15 minutes) Pull five recent pieces across channels and score them against the Voice Audit Checklist's ten checks. Done when: you have a number out of 50 and know your two weakest checks.
Step 2: Name the drift, specifically. (20 minutes) Read the five pieces again and write down, in plain language, what generic actually looks like in your content. Not "it's not us," the actual pattern: "every post opens with a general statement," "we stopped using specific numbers," whatever it is. Done when: you can describe the drift in one sentence someone outside your business would understand.
Step 3: Find your last on-voice piece. (15 minutes) Go back through your archive until you find something that still sounds unmistakably like you. This becomes your anchor for the rest of the recovery. Done when: you have one piece you'd point to and say "write like this."
Step 4: Rewrite your Voice Charter. (30 minutes) Don't assume the old one still holds; drift often starts because the charter got vague or was never written down. Rewrite it in 25 words or fewer using the anchor piece from step 3 as your reference. Done when: you can read the charter aloud and it sounds like a person, not a mission statement.
Step 5: Rebuild your banned list from the actual drift. (20 minutes) Go back to step 2's pattern and turn it into specific banned words and constructions, not a generic AI-slop list copied from somewhere else. If your drift is "every opener is a general statement," your banned list needs an actual rule about openers, not just banned words. Done when: the list has at least one item that directly targets your specific drift pattern, not just the usual suspects.
Step 6: Update your prompts with the new charter and list. (30 minutes) Go through every AI prompt you use regularly and paste the rebuilt charter and banned list at the top of each one. Old prompts with the stale charter are the single most common cause of drift returning within a month. Done when: every prompt you use weekly has the current charter in it, and you've deleted or archived the old versions so you can't grab them by mistake.
Week 2: Test and relaunch
Step 7: Run one prompt jam before writing anything real. (15 minutes) Use the Content Testing Framework's format: generate two versions of one low-stakes piece with your rebuilt prompt, score both, confirm the new prompt actually beats the old habit before you trust it on real content. Done when: the rebuilt-prompt version scores higher on charter match than what you were producing before recovery.
Step 8: Write 3 real pieces with the rebuilt system. (60-90 minutes total) Not test pieces, real ones you intend to publish. Use the new charter, the new banned list, and the anchor piece as your reference every time. Done when: all three exist as finished drafts, not just outlines.
Step 9: Score the 3 pieces against the original audit. (20 minutes) Run the same ten checks from step 1 on these three new pieces. Done when: your score has moved up at least 10 points from the original audit, or you know specifically which check is still failing and why.
Step 10: Relaunch and set a recheck date. (15 minutes) Publish the three pieces. Put a recurring reminder on the calendar to rerun the Voice Audit Checklist in 30 days, not "eventually." Done when: the pieces are live and the 30-day recheck is on the calendar, not just intended.
The 2-week sequence at a glance
| Step | What | Time | Done when |
|---|---|---|---|
| 1 | Run the audit | 15 min | Score out of 50, two weakest checks named |
| 2 | Name the drift | 20 min | One-sentence description of the pattern |
| 3 | Find the anchor piece | 15 min | One piece that still sounds like you |
| 4 | Rewrite the charter | 30 min | Reads aloud like a person |
| 5 | Rebuild the banned list | 20 min | Targets your specific drift |
| 6 | Update every prompt | 30 min | Old prompts deleted or archived |
| 7 | Run one prompt jam | 15 min | New prompt beats old habit |
| 8 | Write 3 real pieces | 60-90 min | Three finished drafts |
| 9 | Re-score against original audit | 20 min | Score up 10+ points |
| 10 | Relaunch, set recheck | 15 min | Live, and 30-day recheck scheduled |
What to do if step 9 doesn't move the score
If three real pieces don't clear a 10-point jump, the problem usually lives in step 2 or step 5: the drift wasn't named specifically enough, so the banned list and charter didn't target the actual pattern. Go back to two real drifted pieces, read them line by line, and find the sentence-level habit you missed the first time. Recovery that skips the diagnosis step tends to relapse within a month; the fix is almost always more specific, not more effort.
Keeping it from happening again
The 30-day recheck from step 10 is not optional. Voice drift is quiet by nature; it never announces itself the way a broken feature does. A scheduled recheck is the only thing that catches it before six months of content go generic again.