You have spent hours staring at a blank page trying to write a headline.
Meanwhile, the best headline you will ever run is sitting in your inbox. Unread. Unused. Written by a customer who was not even trying.
You will find it in the enquiry email that says "I just need someone who will call me back." Or the review that says "I wish I had done this two years ago." Or the sales call where a buyer says "my last agency sent me reports I could not understand."
Most businesses file that under feedback. Treat it as copy.
Your customers already wrote your best marketing. Most businesses just never read it.
The method is simple. Collect what customers wrote and said, then strip out the personal details. Get an AI tool to sort the pile into recurring objections, questions and phrases, and put the exact words into your headlines and ads. AI does the sorting. The words stay theirs.
Why your words lose to theirs
Every business describes itself in its own language. Polished phrases like "tailored solutions" and "end to end support."
Your buyer does not think in those words. They think in problems. They lie awake with one specific frustration, and they describe it in plain, slightly messy language.
Put their exact words in your ad and they feel understood before they have read a single claim. Put your words in it and they have to translate. People scrolling past do not translate. They leave.
Direct response writers have been saying it for generations: enter the conversation already happening in the customer's head. The fastest way to learn it is to listen. If your ads keep underperforming despite decent targeting, there is a good chance your message is the problem, and customer language is the quickest fix for it.
You do not need a better copywriter. You need a better ear. The market has been telling you what to say for years.
Where the gold is buried
You are sitting on more customer language than you realise. It is scattered across a dozen places nobody connects.
- Enquiry emails and form messages. The first thing a buyer writes, before you have shaped their thinking.
- Sales call notes and recordings. The questions asked before buying, and the hesitations right before a yes or a no.
- Reviews. Yours and your competitors'. The praise tells you what matters. The complaints about competitors tell you what to promise.
- Chat logs and support tickets. What confused people, and what nearly made them ask for a refund.
- Lost deals. The polite reasons people give for going elsewhere. Every one is an objection your marketing never answered.
One message looks like noise. A few hundred together show patterns: the same fear, phrased the same way, by strangers.
Reviews deserve extra attention because they do double duty. They are raw copy for you and, increasingly, source material for the AI assistants recommending businesses to your future customers. We covered that side in AI reads your reviews before it recommends you.
Where AI earns its keep
Reading a few hundred emails and transcripts and sorting them by theme used to take days a small business did not have.
Now it takes an afternoon.
One thing to get straight first. You are asking the AI to sort, group and count what real people already said, and nothing else. It writes no copy. That keeps the raw material honest, because every word came from a customer.
Give it a batch of anonymised messages and four jobs.
- Group the objections. Price, timing, trust, a bad experience last time, not sure it applies to them. Ask which come up most often.
- Pull the recurring phrases. The exact wording customers repeat, quoted word for word.
- List the questions asked before buying. Sorted by how often they appear and how early in the conversation.
- Separate the before and after. What life looked like before they found you, and how they describe it afterwards.
Then insist on one rule. Every theme it reports must come with the original quotes that support it. If it cannot show you the customer's words, you cannot trust the pattern.
The afternoon, step by step
- Gather the raw material. Export your last fifty to two hundred enquiry emails, form submissions, reviews and call notes into one document. Add a handful of your competitors' reviews, the glowing ones and the angry ones.
- Anonymise everything. Names, phone numbers, emails and addresses all come out. More on this below, because it is not optional.
- Split it into batches. Twenty or thirty messages at a time. A small batch is easy to check against the originals. One giant paste is not.
- Give the tool a tight brief. Something like: "Here are customer messages. Do not rewrite or summarise them. Group them into objections, questions asked before buying, and descriptions of the problem. For each group, count how many messages mention it and quote the three clearest original lines."
- Merge the batches. Ask for one combined list, ranked by frequency, with quotes still attached.
- Read the quotes yourself. Every one. You will catch the tool misreading sarcasm, or inventing a tidy theme the messages do not support.
You end the afternoon with a ranked list of what your customers worry about and ask about, in their words. Most businesses have never had that document. Every piece of copy you write next starts from it.
Clean it before you paste it
Do not skip this.
Customer emails and call notes are full of personal details. Names, phone numbers, addresses, account details, sometimes health or financial information. None of that belongs in a public AI tool. Australia's privacy regulator, the OAIC, recommends that organisations do not enter personal information into publicly available generative AI tools, and sensitive information least of all.
So strip it out first. Swap names and contact details for placeholders like [customer] and [suburb], and keep only the part of each message that carries the language you need. Check the data settings and terms of whatever tool you use, and stick to the promises your own privacy policy makes to customers.
You want the words. You do not need the people attached to them.
Human judgement decides what gets used
AI can tell you a phrase appears constantly. It cannot tell you whether that phrase belongs in your headline.
Some patterns are gold. Others are a complaint you should fix in the business instead of spinning in an ad. And a few phrases are memorable but attract the wrong buyer.
That call is yours. Pick the language that matches the customers you want more of. If a phrase came from a review or a named testimonial, get permission before you quote the person directly.
Your copy stops sounding like everyone else's and starts sounding like your market. The model sorts and you write, which is how you use AI without sounding like a robot.
A worked example: the accountant who stopped saying "peace of mind"
Say you run a small accounting practice in Melbourne. Your website headline is "Tailored accounting solutions for growing businesses." Your ads say "Peace of mind at tax time." Both are fine. Nobody has ever thought either sentence in their life.
You pull a hundred and twenty enquiry emails and reviews, anonymise them, and run the exercise. Say the top objection, by a long way, turns out to be fear of being judged: people apologising for the state of their records before they have even met you. The phrase that keeps coming back is some version of "my books are a mess." And the question most people ask before buying is whether you can fix previous years as well as this one.
So the headline becomes: "Books a mess? We have seen worse. We will sort out this year and the years you have been avoiding." An ad opens with that same worry and answers it with a promise of no judgement. The FAQ leads with "Can you fix my previous years' tax returns?"
Nothing in that copy was invented. It was all sitting in the inbox. The practice is made up, so run the exercise on your own messages to find your version.
Putting their words to work
Headlines. Take the most repeated frustration and use it close to word for word. A headline that sounds like the reader's own thought stops the scroll.
FAQs. Use the real questions asked before buying, in the order buyers ask them. Answer them plainly, including the awkward ones about price and commitment.
Ads. Build one ad around each major objection. Name the fear, then answer it with proof. Run them against each other and let the market pick. Real customer voices on camera carry this even further, which is why real customers outsell influencers.
Offers. If the most common hesitation is risk, reduce the risk. If it is speed, lead with speed. Your objection list is a map of what the offer is missing.
Do this this week
Pull fifty recent enquiry emails, reviews and call notes. Strip every personal detail. Ask an AI tool to group the objections, list the questions asked before buying and quote the phrases that repeat, with the original lines attached.
Read the quotes yourself. Pick the three strongest phrases.
Put one in a headline, one in an ad and one at the top of your FAQ.
Then book a slot each quarter to repeat it with fresh messages, because the language shifts as your market does. Your customers have been telling you what to say all along. If you would like help turning that raw material into a full campaign, that is the heart of how we approach digital marketing at LIQID NOISE: listen first, then write.
Frequently asked questions
What is voice of customer research?
Voice of customer research means collecting the exact words customers use to describe their problems and questions, then using that language in your marketing. Sources include enquiry emails, reviews, sales call notes, support tickets and lost deals. The aim is copy that sounds like your buyer's own thoughts, so they feel understood before you make a single claim.
Can I use ChatGPT to analyse customer reviews and emails?
Yes, as long as you remove personal details first and check the tool's data settings. Ask it to group objections, list the questions asked before buying and quote recurring phrases word for word. It should not rewrite them. Insist every theme comes with original quotes attached, and read those quotes yourself before trusting the pattern.
Is it legal to use customer reviews in my ads?
Using patterns and common phrases from reviews to shape your copy is standard practice. Quoting a named person directly is different: get their permission first. In Australia the ACCC says it is against the law to create fake or misleading reviews, and it warns businesses against editing reviews to remove the negative parts. Quote people as they wrote it.
How many customer messages do I need to find useful patterns?
Fifty is enough to start seeing repeated objections and phrases. A hundred or more gives you clearer rankings. Mix your sources, because an enquiry, a review and a call note each capture a different stage of the buying decision. Repeat the exercise each quarter so your copy keeps up with how customers talk.