A voice transformation is hard to describe and easy to show. A before/after gallery does the showing: real generic drafts next to their on-voice rewrites, side by side, no explanation needed. It's the fastest way to make the Outcomes pillar tangible to someone who hasn't seen the difference yet. Pair the finished gallery with the Voice Consistency Scorecard to show the score jump alongside the visual proof.
This guide walks through picking your four pieces, documenting the before honestly, generating the after with your actual voice system, writing annotations that don't oversell, and using the finished gallery in a real sales conversation.
Step 1: Pick 4 pieces (20 minutes)
Four is the right number. Fewer looks thin, more dilutes the strongest examples. Choose pieces that show range, not your single best save.
Pick one from each category:
- A generic AI first draft. The rawest example you have: unedited, obviously could-be-anyone copy.
- A rewrite you did by hand. Shows the manual version of the fix, before you had a system.
- A prompt-system rewrite. Same generic input, run through your Voice Charter and Brand Voice JSON. Shows the system working.
- A real published piece with its early draft. The most convincing one, because it's not hypothetical; it shipped.
Your fill-in:
Piece 1 (generic draft): _________________________________ Piece 2 (hand rewrite): _________________________________ Piece 3 (system rewrite): _________________________________ Piece 4 (published + draft): _________________________________
Step 2: Document the before, honestly (20 minutes)
The before has to be real. A strawman before (deliberately worse than any actual draft) makes the whole gallery untrustworthy the moment someone senses it. Use an actual first output.
For each before, capture:
- The exact prompt or brief that produced it (if you have it)
- The unedited text, word for word
- One honest note on what's wrong with it, in plain terms: "sounds like a press release," "no personality," "could be any competitor"
Resist the editor's instinct to clean it up. The generic-ness is the point.
Step 3: Generate the after with your voice system (30 minutes)
Run the same input through your actual system, the one you'd use on a real day, not a special-occasion version.
- Paste your Voice Charter and Brand Voice JSON (or banned words and approved phrases if you don't have the JSON yet) into your AI tool.
- Use the same topic or the same rough brief as the before.
- Generate the draft.
- Edit it exactly as much as you normally would, no more, no less. If you'd tighten two sentences on a real day, tighten two sentences here.
Log the edit distance: how much you changed between AI output and published after. A short edit distance is itself a proof point worth noting in your annotation.
Step 4: Write the annotation (20 minutes)
Each pair needs a short annotation that names the specific difference, not a vague "so much better." Three sentences, same format for all four.
Annotation template:
What changed: [one sentence naming the specific fix: opener, specificity, tone, structure] How: [the one thing in the voice system that produced it: charter, banned list, example injection] Why it matters: [the concrete outcome: read time, reply rate, "sounds like us" test, whatever's true and provable]
Worked example annotation:
What changed: The opener went from "In today's competitive market" to a specific number from the client's own data. How: The prompt included the Voice Charter and one instruction: lead with a concrete detail, never a general claim. Why it matters: The client said this was the first LinkedIn post from us they'd have posted themselves without editing.
Do not attach numbers you didn't measure. If you don't have a real reply-rate lift, say "the client's own words when they saw it" instead of inventing a percentage.
Step 5: Build the swipe-compare layout
Any simple two-column or slider format works: side-by-side text blocks, a slider tool, or even a single doc with a clear divider. The format matters less than the content. Keep formatting identical across all four pairs so the pattern reads instantly: before on the left or top, after on the right or bottom, annotation beneath.
Layout checklist:
☐ Same visual treatment for all 4 pairs ☐ Before and after both fully readable, no truncation ☐ Annotation directly under each pair, not on a separate page ☐ Source noted (draft date, tool used, real or anonymized client name) ☐ No pair edited after the fact to look more dramatic
Step 6: Use it in sales conversations
The gallery earns its place in a sales call at one specific moment: right after a prospect describes their current AI output as "fine" or "close, but something's off." That's the exact gap the gallery closes.
How to use it live:
- Don't lead with it. Let the prospect describe their problem first.
- Pull up the pair closest to their situation (generic-to-system if they're starting from scratch, hand-edit-to-system if they already edit by hand).
- Ask them to read the before out loud, or read it yourself. Let the contrast land before you talk.
- Point at one specific line in the annotation, not the whole gallery. One sharp detail beats four vague ones.
- Ask: "Does the after sound like something you'd publish as-is?" Their answer tells you which piece of the offer to talk about next.
Update the gallery every quarter. A gallery of only old examples starts to feel like a highlight reel from a different era of your work; refresh it with something recent enough that it still sounds like you now.