Voice drift is quiet. No single post sounds wrong; the aggregate slowly stops sounding like you. Especially once AI is drafting and multiple hands are publishing. You can't fix what you only feel, so once a month, measure it. This is the proof-point system behind the Outcomes pillar.
Step 1: Pull your sample (10 minutes)
Collect 10 published pieces from the last 30 days. Spread them deliberately:
- 3 social posts (mix platforms)
- 2 emails (newsletter, sequence, or one-to-one templates)
- 2 long-form pieces (blog, article, video script)
- 2 support or sales touchpoints (a reply, a proposal paragraph, a FAQ answer)
- 1 wildcard: whatever channel you trust least this month
Pull them into one doc, in random order, stripped of dates. If you have a team, have someone else pull the sample so you can't cherry-pick.
Step 2: Score each piece (25 minutes)
Four dimensions, each 1 to 5. Keep your Voice Charter and banned list open while you score.
Dimension 1: Charter match. Read the charter, then the piece. Does it sound like the person the charter describes?
- 5: Unmistakably us
- 3: Could be us, could be a competent competitor
- 1: Could be anyone with a ChatGPT subscription
Dimension 2: Banned words and constructions. Search the piece for every banned word, plus your banned constructions (engagement bait, fake-profound closers, whatever's on your list).
- 5: Zero hits
- 3: One hit
- 1: Two or more hits
Dimension 3: Structure. Does the piece follow your rules for its channel: length, opening style, CTA style, formatting?
- 5: Follows all channel rules
- 3: Right shape, wrong details (CTA off-style, opening generic)
- 1: A LinkedIn post wearing an email's clothes, or vice versa
Dimension 4: Would they recognize us? Cover the byline and logo. Would a regular reader or good client guess it's you? This is the gut-check dimension, and it's allowed to disagree with the other three.
- 5: Instantly
- 3: Maybe, after a paragraph
- 1: Never
Log everything in the tracking table (below). Each piece scores 4 to 20.
Step 3: Tally and read your band (5 minutes)
Add all 10 piece scores. Maximum is 200.
170 to 200: Holding. Your system works. Do this: find your single lowest-scoring piece, identify which prompt or person produced it, and make one adjustment. Don't renovate what's working.
140 to 169: Drifting. Consistent-ish, with leaks. Do this: look at scores by dimension, not by piece. One dimension is usually dragging everything. Banned words leaking means the list isn't in your prompts. Structure slipping means channel rules exist in your head but not in your Channel-Switch prompts. Fix the system for that one dimension this month, nothing else.
110 to 139: Fraying. A regular reader would notice. Do this: sort pieces by score and look for the pattern. Usually it's one channel or one author (human or prompt) producing all the low scores. Re-onboard that channel: updated Brand Voice JSON in every prompt, red-pen the next five pieces before publishing.
Under 110: Reset. The voice system isn't being used, or doesn't exist in usable form. Do this: stop auditing, start rebuilding. Confirm the Voice Charter still describes you, rebuild the Brand Voice JSON, and put it into every drafting workflow. Score again in 30 days; expect to jump a full band.
Step 4: Log it (5 minutes)
Per-piece table (one per month):
| # | Piece / channel | Charter (1-5) | Banned (1-5) | Structure (1-5) | Recognize (1-5) | Total | Note |
|---|---|---|---|---|---|---|---|
| 1 | |||||||
| 2 | |||||||
| ... | |||||||
| 10 |
Month-over-month tracker (one row per month):
| Month | Total (/200) | Band | Weakest dimension | Weakest channel | The one fix shipped | Next month's watch item |
|---|---|---|---|---|---|---|
The "one fix shipped" column is the contract. One fix per month, actually implemented, beats five fixes planned.
Rules that keep the score honest
- Score published pieces only. Drafts flatter you.
- Same scorer each month, or two scorers who average. Changing scorers changes the baseline.
- Never score your favorite pieces from memory. The random pull is the point.
- A rising score with a falling "recognize" dimension is a warning: you're following rules but losing the person. Reread the charter out loud and check it's still true.
- After three months in "Holding," drop to quarterly. The scorecard is scaffolding, not a lifestyle.