How do agencies keep each client's brand voice when using AI?

Agencies keep each client's brand voice by giving AI a separate, written voice profile per client: tone rules, approved examples, banned words, product facts and the Arabic register the brand uses. Drafts are generated only from that client's material, checked against the profile, and approved by the person who owns the account before anything is published.

#Key takeaways

• Brand voice drifts when every writer uses one shared chat window with no client context, so drafts slide toward the same generic tone.
• A voice profile per client, with real approved examples, does more for consistency than any clever prompt.
• Each client's material must stay separate, so one brand's claims or phrasing never appear in another's draft.
• A named person on the account approves every piece; AI drafts, it does not publish.
• For a few clients with light content needs, a shared prompt library on an approved team account is enough.

#Why does every client start to sound the same?

A content team at a busy agency writes for a bank, a restaurant group, a property developer and a skincare brand in the same week. Writers open ChatGPT or Claude, paste a brief, and get a competent draft that sounds like everyone else. After a few rounds of edits, the bank's posts and the skincare captions share the same rhythm and the same favourite words.

Clients notice. The World Federation of Advertisers, september 2024 found that 80% of large brand owners have concerns about how their agencies use generative AI for them, with legal and reputation risk high on the list. A generic tone is the visible symptom; uncontrolled use of client material is the risk behind it.

The habit is already there. The Microsoft and LinkedIn Work Trend Index, may 2024 found that 78% of AI users bring their own AI tools to work. In an agency, that means client briefs pasted into personal accounts nobody manages.

#What goes into a client voice profile?

• who the brand is talking to, in one or two sentences, per market.
• tone rules written as pairs: 'confident, not boastful', 'warm, not chatty'.
• ten approved pieces the client signed off, across formats: a post, an ad, an email, a landing page section.
• words and claims to avoid, including regulated claims in health, finance or property.
• product and offer facts the copy may state, with the source document for each.
• Arabic register: Gulf dialect, formal Arabic or a mix, and which channels use which.
• who approves on the agency side and on the client side.

Keep it short enough that a new copywriter can read it before their first brief. If a profile needs a long meeting to explain, it is a brand book, not a working profile.

#How should an agency set this up?

1. Start with the three clients that produce the most copy and write their profiles with the account lead.
2. Store each profile and its approved examples in a separate project or workspace, one per client.
3. Brief writers to draft only inside the right client workspace, on approved accounts, never in personal tools.
4. Add a check step: the draft is compared with the profile for banned words, unapproved claims and the wrong register.
5. The account owner approves; every rejection is logged with a reason, and the reasons update the profile.
6. Review drift every quarter by reading a sample of published pieces side by side with the approved examples.

#What can go wrong?

The common failures are cross-client leakage, where a draft borrows another client's slogan or product claim; invented facts, such as an offer or a branch that does not exist; and Arabic that reads like a translated English post. Separate workspaces, a facts list with sources and a native-speaker review on Arabic copy cover most of it.

The slower failure is the profile going stale. When the client launches a new campaign line, the profile and examples need to change too, or every draft pulls the brand back to last year.

#When is a prompt library enough?

If you write for a handful of clients, with one or two writers each, a prompt library on an approved ChatGPT or Claude team plan is enough: one project per client holding the profile and approved examples. A short internal session on how to use it, plus a rule about personal accounts, covers the rest.

You need more when many writers serve many clients, when copy runs in Arabic and English, when regulated sectors are on the roster, or when clients ask how their material is handled and you want a written answer.

#Where Stennir fits

Stennir starts with AI consulting: the AI Opportunity Audit looks at how copy moves through your agency today and ranks it against other workflows, then sets the governance rules clients will ask about. If a build makes sense, AI implementation sets up per-client workspaces and checks, and AI training teaches writers to use them.

#What should an agency content lead do next?

Pick your three busiest clients and try writing their profiles; the gaps you find are the brief. Then read the AI for marketing page and AI for content production. When ready, take the free readiness assessment or book a 30-minute call.

#Common questions