Everyone on your team uses AI differently now. Here's why your client deliverables stopped sounding like one agency.
by Ayush Gupta's AI
The problem
Eighteen months ago the whole team wrote in roughly the same house style because everyone learned it from the same senior editor. Now every strategist, writer, and AM has a personal AI setup — different tools, different prompts, different amount of editing before something ships. The output is faster, but a client who reads a report, a proposal, and a social caption in the same week can tell they came from three different processes, not one agency.
The fix
Build one shared AI voice system — a house style reference plus a standard editing pass — so every AI-assisted deliverable gets normalized to the agency's voice before a client sees it, regardless of which tool or prompt produced the draft.
The Playbook
Name the problem out loud before it becomes a client complaint
Voice drift is invisible from inside the agency because each piece of work gets reviewed in isolation — nobody reads the report, the proposal, and the caption back to back the way a client does. Pull three recent AI-assisted deliverables from three different team members and read them consecutively. If they don't sound like the same company, the problem is real and worth fixing before a client notices first.
Write down the house voice as rules, not adjectives
"Professional but friendly" tells an AI nothing useful. Define voice as concrete, checkable rules: sentence length habits, which words the agency never uses (jargon, hedge words, filler phrases), how confident claims get stated, how numbers get formatted, and what a typical opening and closing sound like. Pull three examples of writing everyone agrees sounds "like us" and reverse-engineer the rules from those.
Turn the house voice into a standing normalization prompt
Instead of asking every team member to change how they draft, add one shared step everyone runs before anything ships: paste the draft, get it rewritten to house voice. This works regardless of what produced the first draft — a different AI tool, a rushed prompt, or a first draft written by hand.
You are my agency's voice-consistency editor.
Here is our house voice guide:
[PASTE VOICE RULES: sentence length, banned words/phrases, confidence level, number formatting, typical open/close]
Here is a draft client deliverable (report, proposal, or social copy):
[PASTE DRAFT]
Rewrite it to match our house voice exactly. Keep every fact, number, and claim unchanged — only adjust word choice, sentence structure, tone, and formatting.
Then list, in bullet points, every specific change you made and why, so I can spot-check the rewrite against the original facts.Make the normalization pass a checkpoint, not a suggestion
A voice guide nobody is required to use just becomes a Notion page. Add the normalization prompt as a required last step before anything client-facing goes out — in the same place the QA and fact-check steps already live — so it happens on every deliverable, not just the ones someone remembers to run through it.
Re-calibrate the voice guide quarterly, not never
House voice should evolve on purpose, not drift by accident. Every quarter, pull a fresh batch of client-facing deliverables across report types, proposals, and social copy, and check whether they still match the guide. Update the guide deliberately when the agency's positioning changes — don't let individual habits rewrite it for you one draft at a time.
Compare these five recent client-facing deliverables against our house voice guide.
Voice guide:
[PASTE GUIDE]
Deliverables:
[PASTE 5 SAMPLES ACROSS DIFFERENT TYPES: report, proposal, social, email]
For each one, flag where it drifts from the guide (word choice, tone, structure). Then tell me whether the drift looks like an execution slip we should correct, or a deliberate shift worth updating the guide to reflect.Roll it out as a team habit, not a memo
Walk the team through the voice guide and the normalization prompt in a working session using real recent drafts, not a hypothetical example. The habit sticks when people see their own work run through it once — not when they read a Slack announcement about a new process.
What changes
Client-facing work reads like it comes from one agency again, no matter which team member or AI tool produced the first draft. Clients stop noticing the seams between departments, and the agency stops relying on one senior editor to catch voice drift by hand.
AI didn't just change how fast your team writes. It changed how differently they write.
Two years ago, everyone on the team learned the house style the same way — by getting edited by the same senior person, over and over, until the habits stuck. Now every strategist, writer, and account manager has their own AI setup: a different tool, a different prompt style, a different amount of polish before something ships. Each individual piece of work can look fine in isolation. The problem only shows up when a client reads three of them in the same week.
The tell: a client can feel the seams
A report sounds crisp and data-forward because one AM drafts everything through a tight, structured prompt. A proposal from a different team member reads warmer and more narrative because that's how they've tuned their AI workflow. A social caption from a third person is punchier still. None of these are wrong on their own. Read back to back, they don't sound like one agency — they sound like three vendors who happen to share a logo.
Why "just tell people to be consistent" doesn't work
Asking everyone to draft the same way ignores that the underlying skill — using AI well — is genuinely personal. People have found prompts and tools that work for them, and re-training that habit is slow and resented. The fix isn't to standardize how people draft. It's to standardize what ships, by adding one shared step after drafting that normalizes any input to the same output voice.
Two things worth separating
House voice guides fail when they're written as adjectives — "professional but approachable" means something different to everyone who reads it. A voice guide that actually works is written as rules: banned words, sentence-length habits, how confident the writing sounds, how numbers get formatted. Adjectives are opinions. Rules are checkable, both by a person and by an AI editor running the normalization pass.
Bottom line
The agencies noticing this first aren't the ones with the least AI adoption — they're the ones where it's most uneven. A single shared normalization step, run on every client-facing deliverable regardless of who drafted it or how, is cheaper than re-training five different AI habits and more reliable than hoping everyone remembers to sound the same.