How to Write in Your Brand Voice So It Never Sounds Like AI

“Sounds like AI” is not a font problem or a tone slider. It is a pattern problem. Readers flag a post as machine-written when it hits the same beats every time: a tidy three-part rhythm, a “let’s dive in,” a closing “In conclusion,” and adjectives that could describe any company in any sector. The fix is not to write from scratch every time. It is to define your voice concretely enough that a draft can be checked against it - by you, by a teammate, or by a model - and corrected before it ships.

Here is a system that holds up when you are producing five posts a week instead of one a month.

Start with voice anchors, not adjectives

Most brand-voice docs are useless because they describe a vibe. “Confident but approachable. Bold yet warm.” Nobody can write from that, and nothing can be checked against it.

Anchors are the opposite. An anchor is a concrete, testable rule tied to a real example.

Build three kinds:

• Sentence-shape anchors. “Lead with the claim, then the evidence. Never the reverse.” “Sentences under 20 words unless there is a reason.” “One idea per paragraph.”
• Diction anchors. A short list of words you do use and a list you don’t. A founder in Bengaluru fintech might say “UPI rails,” “settlement,” “underwriting” - and never “revolutionize,” “seamless,” or “game-changer.”
• Stance anchors. What you will say plainly that competitors hedge on. “We think most ‘AI-first’ claims are marketing. Here is the test.” Voice is often a point of view, not a word choice.

The test for an anchor: could two different people apply it to the same draft and agree on whether it passed? If not, it is still a vibe. Rewrite it as a rule with an example.

Pull anchors from your own best work

Don’t invent your voice - extract it. Take five posts or emails you were proud of. For each, write down why it sounded like you: the move it made, the word it avoided, the way it opened. Those notes become your anchors. This is also the fastest way to get a voice out of a founder’s head and into something a team can reuse. (Our own guide on turning one idea into a week of posts leans on this same extract-don’t-invent principle.)

Keep a banned-phrases list - and actually enforce it

AI writing has tells. Everyone can feel them; few write them down. A banned list turns a gut reaction into a checklist.

A practical starter list, tuned for B2B:

• Empty intensifiers: “revolutionize,” “supercharge,” “unlock,” “elevate,” “seamless,” “cutting-edge,” “game-changer.”
• Filler transitions: “In today’s fast-paced world,” “Let’s dive in,” “It’s important to note that,” “In conclusion.”
• Hedge-and-both-sides: “It depends,” “There are pros and cons,” when the whole point of a post is to have a take.
• Fake-us framing: “we’re excited to announce” on something nobody is excited about.

Two rules make the list work. First, keep it specific to your market - an Indian SaaS audience rolls its eyes at different phrases than a US developer-tools crowd, so localise the list. Second, pair every banned phrase with the move you make instead. Not just “don’t say revolutionize” but “name the specific thing that changed and by how much.” A ban with no replacement just makes writing timid.

One caveat worth stating plainly: banned lists reduce the obvious tells, but they do not create a voice on their own. Strip every cliché and you still get bland. The anchors do the positive work; the banned list does the cleanup. Some teams (illustrative) find that half their “sounds like AI” flags come from just a dozen repeat-offender phrases - worth finding your dozen.

Run a self-critique loop before publishing

The last mile is a review pass that is separate from writing. When you write and edit in one motion, you grade your own homework. Split them.

A workable loop, whether a person or a model runs it:

1. Draft against the anchors.
2. Critique in a fresh pass. Score the draft on three things only: Does it break any banned phrase? Does it follow the sentence-shape anchors? Does it take a real stance, or is it safe? Be blunt - “paragraph 2 is generic, it would fit any B2B company” is a useful note.
3. Rewrite the specific failures. Not a full redo - targeted fixes.
4. Stop when it passes, not when you are tired.

The critique step is where the “AI smell” gets caught, because it is looking for pattern failures, not typos. A simple prompt to a model, or a checklist for a teammate: “Read this as a skeptical reader in our industry. Where does it sound like it was written by software? Quote the exact lines.” Then fix those lines.

A 60-second worked example

Before: “We’re excited to announce that our new dashboard will revolutionize how teams collaborate, delivering seamless workflows and unlocking productivity like never before.”

That is four banned phrases and zero information.

After: “The new dashboard cuts the steps to schedule a week of posts from nine to two. That is the whole update. If your team spends Monday mornings copy-pasting into a calendar, this is for you.”

The second version passes the anchors (claim first, plain diction, real stance), breaks no banned phrases, and - crucially - could only have been written by someone who knows this specific product. That specificity is what “not AI” actually means.

Make it repeatable, not heroic

A voice you can only hit on a good day does not survive a content calendar. The whole point of writing anchors down, keeping a banned list, and running a critique pass is that the system carries the quality when your attention is elsewhere. It is the same discipline behind scaling LinkedIn content without it flattening into sameness, and it is what separates a real voice from a tool that just fills a template.

Clanner bakes exactly this into how it drafts: your voice anchors and banned phrases feed every draft, and a self-critique pass scores each one against them before it ever reaches your calendar - so scaling your output doesn’t cost you your voice.