AI writing guide
How Do I Get AI to Write in My Brand Voice?
A repeatable method for getting any AI tool to write in your voice. Includes copy-paste prompt templates.
Book a discovery callThe core problem
AI doesn't produce generic output because it's bad at writing. It produces generic output because you haven't given it enough specific signal to do anything else. The fix is not a better AI tool - it's a better brief. This guide gives you a repeatable method for building that brief.
Most people approach AI writing by describing their voice in adjectives: professional, warm, clear, conversational. These descriptions are accurate but useless - they're too common to differentiate. AI has trained on vast amounts of content fitting exactly that description and defaults to the statistical average of it all.
What breaks through the average is specificity: real examples of your writing, explicit rules about what to avoid, and specific sentence-level instructions the AI can actually act on. The four-part method below gives you a structure for building this once and reusing it across every AI writing task.
Why most attempts fail
Five predictable failure modes
These aren't edge cases - they're what most businesses do when they first try to get AI to write in their voice. Each one is fixable once you know what's causing it.
Adjective-only briefing
"Write in a warm, professional, conversational tone."
These adjectives describe roughly 80% of service businesses. AI has seen millions of documents fitting that description and defaults to the weighted average of all of them - which is exactly the generic tone you're trying to escape.
Do this instead
Pair every adjective with a 'sounds like' example. 'Conversational means short sentences and plain English - like this: [paste a real sentence from your writing].' Examples anchor abstract descriptors to something the AI can actually model.
No negative examples or banned list
Telling AI what you want without saying what you don't want.
AI fills gaps with what's statistically likely for your sector. For professional services, that means lots of 'leverage', 'solutions', 'synergy', and opening lines like 'In today's fast-paced business landscape...' - all of which it thinks are good signals.
Do this instead
Write a banned list. Specific phrases, structural patterns (excessive bullets, headers on every paragraph), and tone traps. 'Never open with a question aimed at the reader. Never use the word solutions. Never start a sentence with In today's...' These rules are as valuable as the positive examples.
Starting from scratch every session
Describing your voice requirements in the first message of each new chat.
You're spending 10 minutes of every session re-briefing the AI on context it already had last time. The description also varies slightly each time, introducing inconsistency. You get good output when the briefing is good; generic output when you're in a hurry.
Do this instead
Build one reusable brand voice block and store it as a text file. Paste it at the start of any writing session - or better, set it as a Custom Instruction or Project instruction so it applies automatically.
Accepting or rejecting without specific feedback
'This isn't right, try again.'
'Try again' without context produces almost the same output. AI doesn't know what was wrong with the first attempt and loops back to the same statistical defaults.
Do this instead
Identify the specific problem. 'That opening is too formal - it reads like a press release. Rewrite it starting from the problem the reader has, not the solution we offer. Use a shorter first sentence.' The more precise the correction, the better the next draft.
Inconsistent examples in the briefing
Including old writing alongside recent writing when your voice has evolved.
AI averages across all the examples you give it. If half your examples are from three years ago when you wrote more formally, the output will be a blend of old and new - which is neither.
Do this instead
Choose your three to five most recent, most representative examples. If your voice has changed significantly in the past year, retire older examples even if they're technically good writing.
The 4-part method
Build a brand voice block that actually works
All four parts are required. The method works because each part does something the others don't: the description gives context, the examples provide a model, the rules prevent defaults, and the test loop validates that it all works together before you rely on it.
Voice description
A structured brief that tells the AI who you are, who you write for, and what the voice should feel like. This is not a list of adjectives - it's a specific description backed by examples.
One-sentence business description
What you do, for whom, and what makes your approach distinctive. Not your tagline - a plain sentence that gives context.
Your reader in one sentence
Specific, not generic. 'Small business owners who've heard about AI and want to know what it can actually do for them' is better than 'professionals'.
The feeling you want to create
When someone finishes reading your content, what should they feel? Confident? Reassured? Energised? Clearer? Suspicious of jargon? This shapes tone more usefully than style adjectives.
Three voice markers with examples
Pick three words that describe your voice. For each one, write one real sentence from your existing copy that demonstrates it. 'Plain-spoken: [sentence]. Direct: [sentence]. Grounded: [sentence].'
Positive examples
Three to five real pieces of your best writing. The ones you'd show a new writer and say 'write more like this'. Quality over quantity - three excellent examples outperform ten inconsistent ones.
Choose recent, representative pieces
If your voice has evolved, use only the last 12-18 months. Include different formats if you can: one paragraph of website copy, one email or newsletter section, one short social post.
Annotate why they work
Two sentences per example is enough. 'This one works because the first sentence gets to the point immediately and there's no throat-clearing. It uses the word you twice in the first paragraph.' The annotation gives the AI richer signal than the example alone.
Flag the best sentence in each
Highlight one sentence per example that is most representative of your voice. This is the signal the AI should weight most heavily.
What an annotated example looks like
EXAMPLE: Here is an excerpt from our website that represents our voice well: "Most businesses don't need a custom AI system. They need the right three tools set up properly, with someone who's done it before showing them how." Why it works: Short, declarative sentences. Starts by subverting an expectation. No jargon. Speaks directly to the reader's likely situation. The second sentence lands the practical alternative without over-explaining it. Best sentence: "They need the right three tools set up properly, with someone who's done it before showing them how."
Rules (what to avoid)
Explicit prohibitions are often more valuable than positive descriptions, because they prevent AI from defaulting to sector-specific clichés. Be specific - general rules like 'avoid jargon' are ignored.
Banned words and phrases
List the exact words and phrases you never want to see. 'Solutions', 'leverage', 'robust', 'cutting-edge', 'in today's fast-paced landscape', 'it goes without saying'. The more specific, the better.
Structural patterns to avoid
Bullet points for everything. A header every two paragraphs. Conclusions that restate the introduction. Opening with a rhetorical question aimed at the reader ('Are you struggling with...'). Starting every email with 'I hope this finds you well.'
Tone traps
Over-formal ('please do not hesitate to contact us'). Over-casual (overly familiar slang or excessive exclamation marks). Sycophantic ('Great question!'). Vague reassurance ('rest assured'). List the specific patterns you've seen AI produce that you want to avoid.
Format preferences
Sentence length preference (short? medium? mixed?). Paragraph length. Whether you use bullet points (and when). Whether you use subheadings in emails. Oxford comma or not. These specifics matter for voice consistency.
The test loop
Once you've built your brand voice block, test it with three specific formats before using it for anything client-facing. Refine the block based on what fails, not what feels approximately right.
Test 1: A short email
Ask the AI to write a 5-sentence follow-up email after a first meeting with a potential client. Read it aloud. Does it sound like you? Is the opening line something you'd actually write?
Test 2: A social post
Ask the AI to write a LinkedIn post sharing one practical thing you've learned about AI this month. No longer than 150 words. Check: would you recognise this as yours if you saw it in a feed?
Test 3: A website paragraph
Ask the AI to write a 60-word paragraph describing your main service. Check for jargon, passive voice, and whether the value is clear to someone who knows nothing about you.
Refine, don't start over
After each test, identify the specific failure and add it to your rules list. 'Output keeps starting with a question - added to banned patterns.' Three rounds of testing and refining usually produces a reliable voice block.
Prompt templates
Four copy-paste templates
These work across ChatGPT, Claude, and Gemini. Fill in the bracketed sections with your specifics - the more detail you add, the better the output. Use the copy button to grab each one.
The brand voice block
Paste this at the start of any AI writing session, or save it as a Custom Instruction
BRAND VOICE BRIEF Business: [One-sentence description of what you do and for whom] Reader: [Who you're writing for - be specific] Voice: [3 adjectives, each with a 'sounds like' example sentence] - [Adjective 1]: sounds like "[real sentence from your writing]" - [Adjective 2]: sounds like "[real sentence from your writing]" - [Adjective 3]: sounds like "[real sentence from your writing]" Examples of writing that represents our voice well: [Paste 2-3 real examples with a one-sentence annotation each] Rules - never do these: - Never use: [list banned words/phrases] - Never open with a question aimed at the reader - Never use bullet points for everything - use prose where possible - Sentences: [short/medium/mixed] preference - [Add your specific patterns to avoid] When in doubt, write shorter and plainer.
The single-task prompt
Use after your brand voice block to brief a specific piece of writing
Using the brand voice brief above: Write [content type: email / LinkedIn post / website paragraph / etc.] Purpose: [What this piece needs to achieve] Reader: [Who specifically will read this - more specific than the general reader above if needed] Key message: [The one thing they should take away] Call to action: [What you want them to do next, if anything] Length: [Approximate word count or format, e.g. '3 paragraphs', '150 words max'] Do not start with: [Any specific openings you want to avoid for this piece]
The refinement prompt
Use when the first output is close but wrong in a specific way
This is almost right but [specific problem]. Rewrite it with these changes: - [Change 1: specific, e.g. 'Shorten the opening sentence to under 10 words'] - [Change 2: specific, e.g. 'Remove the phrase "innovative solutions" - replace with plain language describing what we actually do'] - [Change 3: specific, e.g. 'The third paragraph is too cautious - make it more direct'] Keep: [What worked in the first draft that you want to preserve] Change: [What needs to change] Do not rewrite the parts I haven't asked you to change.
The voice analysis prompt
Use this to get AI to help you build your voice examples from existing writing
I am going to paste three pieces of writing that represent my brand voice at its best. For each one, I'd like you to: 1. Identify the 3-4 most distinctive voice characteristics 2. Note any specific word choices or sentence patterns that are particularly representative 3. Flag any phrases or structural choices I should use as positive examples After analysing all three, write a short 'voice summary' I could use to brief other AI tools. Include: the key descriptors, 2-3 specific sentence-level examples, and 3-5 things this voice deliberately avoids. [Paste your first example here] [Paste your second example here] [Paste your third example here]
Want this done properly?
We build AI workflows that write in your voice - not a generic approximation of it
Our AI workshops and consultancy sessions cover brand voice setup, Custom GPT configuration, and the full writing workflow. We'll help you build a setup your whole team can use consistently, not just a prompt you have to remember.
Common questions
FAQs
Get in touch