Brand Voice Analysis for Content: How to Do It With AI
Brand voice analysis for content is the work of reading what your brand has already published, describing how it sounds as rules a writer can follow, and checking new content against those rules. The output is a short voice guide: a personality, a writing style, words and habits to avoid, and real examples. An AI can draft that guide from your website in minutes, but a person should confirm it before any writer, human or AI, uses it.
What is brand voice analysis?
Your brand voice is how your company sounds whatever the subject: confident or careful, plain or playful, "we" or "the company". Tone is how that voice shifts by situation, warmer in a welcome email and drier in a pricing page. Brand voice analysis turns that impression into something you can write down and test.
It looks at a handful of things a reader notices even if they could not name them:
Person. Do you write as "we" and speak to "you", or in the third person?
Formality. Contractions or none, slang or none, how sign-offs read.
Sentence shape. Short and direct, or longer and explanatory.
Vocabulary. The words you always use, and the ones you never do.
Claims. Whether you lead with numbers, stories or plain statements.
The point is not a mood board. It is a set of rules specific enough that two writers given the same brief produce copy that sounds like the same company.
Why does content need a voice analysis before AI writes it?
Because a language model has a default voice, and it is nobody's. Ask a general-purpose writer for a blog post with no voice guide and you get polished, interchangeable copy full of the same phrases every other brand is publishing.
A voice analysis gives the model something better to imitate than its own habits. It also makes review faster. When the rules are written down, an editor can point to the rule a draft broke instead of saying "this doesn't sound like us".
How do you run a brand voice analysis?
Work from evidence, not from how you would like to sound.
Collect samples. Pick pages and pieces that the team agrees sound right: the homepage, key product pages, a few strong blog posts, your best emails. Leave out legal pages, guest posts and anything ghostwritten in a different voice.
Start from your guidelines if you have them. A published brand guideline is a decision someone already made. Where it and your website disagree, the guideline wins.
Read for patterns. Go through the samples with the list above: person, formality, sentence shape, vocabulary, how claims are made. Note what repeats, and note what never appears.
Write the guide. Keep it to a page: a personality in two or three sentences, a writing style, a list of things to avoid, and four or five examples lifted from your own material.
Test it. Give the guide and one real brief to a writer, or an AI, and read the result next to your best existing piece. Fix the guide where the draft drifted.
Confirm and own it. One named person confirms the guide is right and owns changes to it. Revisit it when your positioning changes, not every week.
What should a brand voice guide include?
The guide is only useful if each part can be applied to a draft. This is a workable shape:
Part: Personality, What it says: Who the brand sounds like, Example of a useful line: "A senior colleague who explains, never lectures"
Part: Writing style, What it says: Person, formality, sentence length, Example of a useful line: "We and you. Short sentences. No exclamation marks."
Part: Vocabulary, What it says: Words to use and words to avoid, Example of a useful line: "Say customers, not users. Never say synergy."
Part: Things to avoid, What it says: Habits that break the voice, Example of a useful line: "No rhetorical questions in headlines"
Part: Examples, What it says: Real sentences from your own material, Example of a useful line: Two or three lines from your best page
Part: Tone by context, What it says: How the voice flexes, Example of a useful line: "Warmer in onboarding emails, plainer on pricing"
Specific beats aspirational. "Bold and innovative" tells a writer nothing; "lead each section with the answer, then the reason" tells them exactly what to do.
What does an AI do with a brand voice analysis?
An AI can help at three points, and each needs a different kind of trust.
It can draft the analysis. Given your crawled pages, a model can propose a personality, style, avoid list and examples, with a confidence figure. That is a useful first draft and a poor final answer, because a voice read off a website is a guess about how you want to sound. Treat it as a proposal until a person confirms it.
It can apply the voice when drafting. Once confirmed, the guide goes into every writing prompt, so issue twelve of the newsletter sounds like issue one and a new hire's first blog post sounds like the rest.
It can check drafts against the voice. Some rules can be checked in code: whether the draft writes in the right person, whether it uses contractions when your voice uses none, whether it contains a banned word. Others, such as "explains, never lectures", need a reviewer's judgement. The best setups check the first kind in code every time, and send the second kind to a reviewer.
How does Nagent analyse and apply your brand voice?
On Nagent, your brand lives on one screen, Brand Lock, in the Knowledge Hub. Its own description is the design principle: "Every image, copy block and asset your agents make reads from this row."
Extract a draft voice, then confirm it
The Brand voice panel describes itself as "the written voice used by copywriting and agents. Extracted from your website, applied only after you confirm." Once your website has been crawled, select Extract from website. Nagent reads up to eight of your crawled pages and writes a draft; if nothing has been crawled, it says so and spends nothing.
The draft appears as Extracted draft, awaiting review, with a personality, a writing style, what to avoid, examples taken from your own site and a confidence figure. Confirm makes it your voice; Reject discards it, and you can extract again at any time. Agents read the confirmed voice only, never the draft. Until you confirm one, they write in neutral copy rather than borrowing someone else's identity.
Upload your guidelines so they outrank the website
If you have a published brand guideline, upload it in Brand guideline document as PDF, Markdown or plain text, up to 10 MB. The screen puts it simply: "Your published guideline outranks anything we read off the website." The full brand document also holds a Communication tab for tone of voice and linguistic rules, and a Forbidden tab for patterns your brand never uses.
How CREA's team writes in the voice
CREA, the AI Content Chief of Staff, and the content team draft from the brand book's voice, alongside your products and Knowledge Hub passages. With no brand book yet, drafts use a plain neutral voice: "Write professionally and concisely. Active voice. Specific outcomes." Before a post can publish, a check in code holds it to your voice: it must write in the person your voice sets, and without contractions if your voice uses none.
How RUPA reads it for images
Voice is not only words. RUPA, the image studio on CREA's team, adds your brand guidance and logo to a brand-locked render. Before rendering, she checks the brief against the words your brand book bans and lists any she finds. Rules that cannot be checked from text, such as logo placement, are listed as Needs your eye rather than ticked.
Changes to the brand reach agents within a few minutes. Extracting or confirming a voice needs workspace settings permission, which a workspace admin holds.
What mistakes make a brand voice analysis useless?
Analysing aspirations. If the guide describes a voice you have never published, writers have nothing real to imitate.
Too many adjectives. Five personality words with no examples cannot be applied to a sentence.
No owner. A guide nobody confirms drifts with every new tool and every new hire.
Letting the extracted draft go live unread. A model's reading of your site can pick up a guest post's quirks or an old campaign's tone.
Never testing it. A guide that has not been run against a real brief is a theory.
Frequently Asked Questions
What is brand voice analysis?
It is the process of studying your published content to describe how your brand sounds, as rules covering personality, writing style, vocabulary and things to avoid, so every writer can produce copy that sounds like the same company.
Can AI analyse brand voice?
Yes, as a first draft. A model can read your website and propose a personality, style, avoid list and examples. A person should still confirm it before writers use it. On Nagent, the extracted voice waits as a draft until someone selects Confirm.
How many samples do you need for a brand voice analysis?
Fewer than you might think, if they are the right ones. A handful of pages the team agrees sound right is better than a large pile that mixes voices. Nagent's extraction reads up to eight crawled pages.
How do you keep AI content on brand?
Confirm a written voice guide, put it in every drafting prompt, check what code can check (person, contractions, banned words) on every draft, and have a person review the rest before anything publishes.
What should you do next?
Pull five pieces your team agrees sound right and write the one-page guide from them. To see a voice drafted from your own website and applied by an AI content team, read the Brand Lock guide, or book a demo and we will extract and review a voice with you.