A mid-sized software company recently asked ChatGPT to describe its product in a single paragraph. The answer came back coherent, confident, and wrong. It credited the company with a feature that had been sunset two years earlier, named a founder who hadn’t worked there since 2022, and in the same breath recommended a competitor as the “more established” option. Nobody at the company had ever thought to run that query.
That blind spot is what the new brand voice audit is really about. For twenty years, the test was simple: read your own copy out loud and ask whether it sounded like you. The test that matters now is different. Pull up the chatbot your buyers already use, type your company name, and read what comes back.
The Old Test Measured Output You Controlled
The classic brand voice exercise lived inside a messaging workshop. A team pulled together landing pages, sales decks, a handful of emails, and a few recorded sales calls, then spent an afternoon arguing about whether the tone was warm enough, confident enough, or distinctive enough. The deliverable was usually a short guide with three or four adjectives and a do/don’t table.
The assumption underneath that work was that your brand voice was whatever you published. Control the copy, control the perception. A good writer and a tight guide were the whole program. For most of the last two decades, that was close enough to true that nobody questioned the setup.
The New Test Measures Output You Don’t
Buyers no longer meet your brand voice through your website first. They meet it through a chatbot’s paraphrase of your website, filtered through whatever else the model happens to have indexed about you: a seven-year-old press release, a disgruntled Reddit thread, a competitor’s comparison page. The description that lands in front of the buyer is a composite, and you didn’t write any of it.
That shift is why the audit changed. A Harvard Business Review feature on this transition frames the chatbot as the new primary interface between a brand and its customers, which turns the model’s description of you into the thing that actually shapes buying behavior. Your deck doesn’t shape it. The AI’s summary does.
The practical move is to add a second audit alongside the first. The internal one still matters. You still need the deck, the guide, the writers. The external one is new work, and it’s the one most teams haven’t started. For brands working with Houston branding agencies or other regional partners, that increasingly means looking beyond the messaging they intentionally put into the market and comparing it with the version AI platforms are giving back to customers.
Both Tests Ask “Does This Sound Like Us?” — Only One Can Surprise You
The old audit rarely produced a shock. You wrote the words, you read them back, and the worst case was that a paragraph needed a rewrite. The new audit is unpredictable in a way the old one never was, because the model’s description of you is assembled from sources you don’t curate and weighted by signals you don’t see.
A few things to actually run when you sit down to do it:
- The flat description query. Ask three or four leading chatbots to describe your company in a paragraph. Compare the answers against your own one-liner. Note every claim you didn’t make, and every claim you did make that got dropped.
- The shortlist query. Ask the same models to recommend the top options in your category. If you don’t appear, you now know. If you appear with the wrong label attached (the budget option, the enterprise-only option), you know that too.
- The objection query. Ask the model what the common complaints or weaknesses of your brand are. Models will answer this one even when the public record is thin, and the answer tells you which signals are loudest in the training data.
- The comparison query. Pit your brand against two named competitors and read how the model frames the trade-off. The adjectives it reaches for are the voice your buyer actually hears.
The Second Test Is Now the One That Decides the Sale
The behavior change under all of this is already visible in the shopping data. A July 2025 Omnisend survey of U.S. consumers found that most shoppers now use generative AI tools somewhere in their buying process, and a meaningful share rate ChatGPT above Google specifically for product research. The buyer who used to land on your homepage and form a first impression from your hero copy now forms it from a chatbot’s two-sentence summary before any click happens.
None of this retires the internal voice work. Your site still has to sound like you when the buyer finally arrives, and your sales team still has to carry the voice into the room. But the first impression, the one that decides whether the buyer arrives at all, is being made somewhere you can’t edit, by a system that reads everything written about you and compresses it into a paragraph.
The companies that figure this out early will run the two audits as one program. Write the guide, read the model’s answer, compare them. Then fix the sources until the second one starts to sound like the first.
That’s the brand voice test nobody was running a year ago. It’s the one that matters most now.
