Brand Voice Audit: How to Check If Your Copy Still Matches Your Guide

Most companies with a brand voice guide have never checked their website against it. The guide gets written, approved and shared, and then the site keeps changing underneath it: a new pricing page, a feature launch written in a hurry, a batch of help articles drafted with AI, a homepage headline swapped for a campaign and never swapped back. Six months later the guide describes one company and the website describes another, and nobody can say exactly where the gap opened.

A brand voice audit closes that gap. It takes the rules in your guide and reads your live pages against them, sentence by sentence, so every problem you find comes with the rule it breaks. That makes it different from a general copy audit, which asks whether copy is clear and persuasive, and from a consistency review, which compares pages with each other. Here the guide is the standard, and the question is whether each page still meets it. If you don't have a written guide yet, our guide to defining and documenting a brand voice is the place to start, and the end of this article covers what to do in the meantime.

Turn your guide into checks before you read anything

An audit is only as precise as the guide behind it. A principle like "warm but professional" can't be checked, because two reviewers will read the same sentence and reach different verdicts, and the audit turns into a debate about taste. So before you open a single page, go through the guide and rewrite every principle as a question that a sentence can answer yes or no. Where the guide already gives passing and failing examples, the question is usually easy to write. Where it doesn't, you have found the first gap in the guide, and it is worth fixing now rather than halfway through the audit.

To show how this works, we'll audit a fictional booking app for physiotherapy clinics. Its guide has four principles and a tone map, and each one becomes a check:

• Name the work. Does the sentence say what the app does, with an object and an action a clinic owner would recognise?
• Earn certainty. If the sentence makes a claim, is it backed by a number, an example or a named comparison?
• Use our words. Does it say patients, appointments and clinic, rather than clients, sessions and practice?
• Talk to clinic owners as peers. Does it assume they know physiotherapy while explaining anything that is software jargon?
• Match the tone map. Is the energy right for the page, with pricing and billing kept matter-of-fact?

Every check points at something visible in the sentence. "Is this confident?" describes a mood, while "is this claim backed by a number, an example or a named comparison?" is a check, and two reviewers will almost always agree on the answer. Keep the list short enough to hold in your head while you read, because an audit with fifteen checks per sentence tends to be abandoned by the third page.

Choose the pages that matter

You don't need to audit every page to learn whether your site still matches the guide, and on a large site trying to do so is a good way to never finish. A sample of ten to fifteen pages is enough for most small and mid-sized sites, as long as it is chosen deliberately rather than by clicking through the navigation from the top. Aim for a mix that covers where the voice matters most and where it is most likely to have drifted:

• The pages that make promises: the homepage, pricing and main product or service pages.
• The pages with the most traffic, which may include a blog post or help article you wouldn't have picked.
• Anything edited or published since the guide was written.
• Pages written by someone other than the guide's authors, such as an agency, a freelancer or an AI tool.
• One old page nobody owns, which is usually the furthest from the guide.

If you manage several sites, as agencies do, audit each one against its own client's guide rather than a house standard. The checks will differ from client to client, but the logic for choosing pages stays the same.

Read line by line and log every departure

Now read each page one sentence at a time and run the checks. This is slower than skimming, and deliberately so: a page can feel on-brand at reading speed while containing three or four sentences that fail the guide outright, because the surrounding copy carries them. When a sentence fails, write down the page, the sentence, the rule it breaks and a short note on what it should be doing instead. Resist the urge to rewrite as you go. The audit is a diagnosis, and rewriting mid-audit tends to fix the first pages beautifully while the last ones get skimmed.

Here is what that looked like on our fictional booking app. The homepage headline was the first failure:

Before: The most powerful practice management solution for modern clinics.

It fails three checks at once. "Most powerful" is a superlative with no comparison, "solution" names nothing a clinic owner could picture, and "practice" is a word the guide swaps for "clinic". The rewrite names the work and lets the specifics carry the confidence:

After: Patients book online, reminders go out by text, and your treatment notes sit next to every appointment.

The features page had a quieter problem. Nothing in this sentence is false, but it describes the technology rather than what changes for the clinic, and "maximum efficiency" is a claim with nothing behind it:

Before: Our AI-driven scheduling engine optimises your calendar for maximum efficiency.

After: When a patient cancels, we text the next three people on your waitlist and give the slot to whoever replies first.

The pricing page failed the tone map rather than any single principle. Excitement belongs on a launch announcement, not on the page where someone is deciding whether they can afford you, and "just" quietly argues with the reader about whether $49 is a lot. The fix is matter-of-fact, and it answers the next question before the reader has to ask it:

Before: Get ready to supercharge your clinic! Plans start at just $49.

After: Plans start at $49 a month per clinic, with no setup fee, and you can cancel at the end of any month.

The help centre broke the vocabulary rule three times in one short sentence, with "sessions", "users" and "client", and added a piece of software jargon in "portal". On its own that is a small problem, but the same phrasing turned up across fourteen help articles, which turns out to matter more than any individual rewrite.

Before: Sessions can be rescheduled by users via the client portal.

After: Patients can move their own appointment from the link in their reminder text.

Log the sentences you keep, too

Not every sentence that looks off-voice fails the guide. The About page of our fictional app opened with this line:

We started building this after too many Friday afternoons spent ringing patients who hadn't turned up.

It is warmer and more personal than anything on the product pages, and a reviewer reading "matter-of-fact" too literally might flag it. But it names the work, it is specific, it speaks to clinic owners as peers, and the tone map allows more warmth on the About page than on pricing, so it passes. Recording a few deliberate passes like this is worth the extra minute, because it shows the next reviewer where the guide's boundaries actually sit and stops the voice being sanded down into the blandest version that technically complies.

Keep voice problems separate from other problems

A careful read will turn up problems that have nothing to do with voice, and they need their own column. Our fictional homepage said the app was free for solo practitioners, while the pricing page started at $49 per clinic with no mention of a free plan. Rewriting either sentence in a more on-brand tone wouldn't fix that, because it is a factual contradiction and someone has to decide which version is true. The same goes for outdated feature claims, broken links and sentences that are on-voice but genuinely unclear. Log them as you find them, but label them, so the voice findings stay a clean measure of how far the site has moved from the guide.

Score the results and decide what to fix first

Once every page has been read, the log is the audit. A simple table is enough:

Prioritise by two things together: how much the page matters and how much the failure changes what a reader believes. A superlative in the homepage headline is high priority because most visitors read it and it makes a promise the product has to keep, while a vocabulary slip in one help article is low priority on its own. Then look down the "rule broken" column for patterns, because a rule that fails on one page is a line edit and a rule that fails on ten pages is a briefing problem. In our example, the fourteen help articles all came from the same template, so the right fix was one change to the template and a note to whoever writes help content, rather than fourteen separate rewrites.

If you want a single number to track over time, the share of sampled sentences that pass every check works well enough. It is crude, but it moves when the site improves, and it makes the next audit comparable to this one.

Let the audit fix the guide as well

Every audit is also a test of the guide, and the guide will usually come out of it with a few changes. When two reviewers disagreed about a sentence and neither could point to a rule, a rule is missing. When a rule was broken on page after page for a good reason, such as a technical page that genuinely needed a software term the guide discourages, an exception is missing or the rule is wrong. Write those changes into the guide before the next audit, otherwise the same arguments will happen again.

If you started without a written guide, this is also where you find out how good your working version was. A voice pieced together from your best existing pages is a reasonable place to begin, as long as it is labelled as inferred and someone with authority over the brand confirms or corrects each rule. An audit against an unconfirmed voice can only tell you how consistent the site is with itself, which is useful, but it can't tell you whether the site sounds the way the company has decided it should.

How often to run a brand voice audit

A full audit is worth running at a few predictable moments: after the guide is written or changed, before a redesign or relaunch, and when a new writer, agency or AI workflow starts producing copy for the site. Between those, a lighter quarterly pass over the highest-traffic pages and anything edited since the last audit is usually enough to catch drift before it spreads. New drafts can be checked against the same list before they go live, which is far cheaper than finding them in the next audit.

Run the audit without the spreadsheet

Revisi runs this process against your own voice guide. You add your guidance and reference copy to a site, each page is checked sentence by sentence, and every flagged line comes with the reason it was flagged, ranked so you can see which pages to fix first. If you don't have a guide yet, Revisi infers a provisional voice from your existing pages and labels it as inferred until you confirm it. The free scan reads one public page and shows the sentences that are unclear, generic or off-tone, with results in about 30 seconds.

Run a free audit