You open your landing page next to your two closest competitors’.
Three tabs. Three headlines. All promising to simplify your workflow and unlock your team’s potential.
You cannot tell them apart.
Neither can your buyer, and that’s not a coincidence anymore. It’s the default output. Every one of you is drafting from the same handful of language models, trained on the same internet of software marketing, reaching for the same six adjectives.
“Write in your customer’s voice” used to be advice a good copywriter could half-follow and still win on craft alone.
That’s over. Now it’s the only move left that a model can’t fake for you.
This is how you go get the one input an AI cannot generate: what your own customers actually say, in their own words, collected and mined at a scale that beats reading reviews one at a time.
What is voice of customer research for copywriting?
Voice of customer research for copywriting is the practice of collecting your customers’ exact words, from reviews, calls, tickets, and real conversations. You write your copy from that language instead of your internal vocabulary.
You don’t paraphrase what you think they mean. You lift the phrase they actually used, the one your prospect will recognize as their own thought handed back to them.
That recognition is the entire mechanism. A headline built from your product roadmap sounds like a pitch. A headline built from a customer’s own sentence sounds like someone finally gets it.
“Centralize” and “unlock efficiency” are words your team uses in Slack. Almost no customer has ever said either one out loud about their own problem. If a phrase only lives in your marketing docs, it’s brand language, not voice of customer.
Why does copy sound the same in 2026?
For a decade, sounding like every other SaaS site was a taste problem. A sharper copywriter could out-write it with a better headline.
That’s not the game anymore.
Ask three different marketers for a “punchy SaaS hero headline” and a model hands each of them a version of the same fifteen sentence shapes, because they’re all drawing from the same training data and the same handful of tools. The copy doesn’t just sound similar. It’s converging.
Undifferentiated copy is turning into a measurable discount, not just a vibe. A November 2025 study of a year of real website sales found human-written sites sold for 39% more than sites carrying AI-generated content, per Originality.AI’s analysis of Motion Invest marketplace data.
Sites with AI content also sat unsold 54% longer. Buyers are already pricing sameness as risk. Readers are ahead of buyers.
It’s not only a human-trust problem, either.
Here’s the part most marketing teams miss: the machines reading your copy penalize genericness too.
Researchers at Princeton and IIT Delhi tested nine ways to get a page cited inside an AI-generated answer. The two strongest levers, by a wide margin, were citing a real source and quoting someone directly, each worth a 30 to 40% lift in visibility, according to their GEO study presented at KDD 2024.
So the fix for both problems is the same one: say something only you can say.
Your customers’ exact words are the one input a model has never seen. They live in your support inbox, your call recordings, and your reviews, not anywhere near its training set.
Where the words you need already live
You don’t need a new research budget. You need to go collect what customers have already handed you, for free, in their own language.
- Support tickets and chat transcripts. The rawest phrasing you own. Nobody softens their words when they’re stuck on hold three days before a deadline.
- Public reviews and forums. G2, Capterra, app store reviews, and the threads where your buyer already vents about the problem, unprompted, in front of strangers.
- Sales call recordings. The words a prospect uses before your rep reframes the problem into your language. That first, unfiltered framing is gold.
- Live customer conversations. The richest source by far, and historically the hardest one to run at any real volume.
Review mining specifically, reading through a category’s reviews to pull the exact phrases customers use, is a technique conversion copywriters have leaned on for years to find language a brand would never invent on its own, as Copyhackers’ breakdown of the method shows.
How do you mine voice of customer research at scale?
Knowing where the words live doesn’t help if it takes a week to dig them out. Here’s the fast version.
Step 1: Build one pile. Dump every raw quote, ticket line, review sentence, and call snippet into a single doc as you find it. Don’t edit yet. Don’t judge it yet. Just collect.
Step 2: Tag by pattern, not by source. Sort each line into one of four buckets: the pain (what’s broken), the trigger (what made them finally act), the objection (what almost stopped them), and the win (what changed after).
Frequency is your signal. If six people describe the same moment the same way, that’s your headline, not a coincidence.
Step 3: Keep the rough edges. The line that isn’t grammatically perfect is usually the one that converts, because it sounds like a person, not a brochure.
Search “[your category] + ‘wish it’ OR ‘I hate that’ OR ‘finally’ site:g2.com” (swap the site for Capterra, Reddit-adjacent forums, or the App Store). You’ll surface the exact sentences people use to describe frustration and relief, minus the five-star filler.
The manual version doesn’t scale, so most people quit at ten reviews
Here’s the honest problem with everything above: it works, and almost nobody keeps doing it past the first afternoon.
Reading a hundred reviews by hand is slow. Reading a thousand is a part-time job nobody signed up for. So most teams grab the first ten quotes that feel usable, write the campaign, and move on, long before the real pattern shows up.
The words that would have made the copy actually convert are usually past quote fifty, not quote ten.
Live conversations solve the depth problem, since a real back-and-forth surfaces language and context no static review ever will. They’ve always had the opposite problem: they don’t scale past a handful a week for one person with a calendar.
That’s the exact gap AI-moderated conversations are built to close. hollie can run that conversation with dozens of your customers in parallel, ask the follow-up a human researcher would ask, and hand you back their actual sentences, not a summary written in someone else’s words. See how hollie runs it.
Research on narrative persuasion links easier processing to stronger persuasion, an effect called processing fluency, per a 2021 study in Frontiers in Communication. A sentence your reader already thought takes zero effort to process.
Turn the raw quotes into copy that doesn’t sound like anyone else’s
A pile of quotes isn’t a headline yet. The move is translation, not transcription: find the line that captures the pattern, then trim it until it reads like copy instead of a transcript.
Here’s the difference in practice:
- Generic: “Streamline your onboarding workflow.”
- From the pile: a support ticket that reads, “I don’t want to babysit five different tools just to get one new hire set up.”
- The headline: “Stop babysitting five tools to onboard one person.”
Same idea. One of them could run on any SaaS site built this year. The other could only run on yours, because nobody else has that exact ticket.
The instinct is to clean a customer’s phrasing up until it sounds professional. Resist it. The specific, slightly odd word choice is what makes the line unmimicable in the first place. Polish the grammar if you must, never the personality.
Want to keep the pile growing instead of freezing after one round?
Once you’ve written from a customer’s exact words, ask sharper questions on your next call. You’re not fishing for compliments, you’re fishing for their next unscripted sentence.
And when a quote is strong enough to carry a whole page, not just a headline, turning that one conversation into a full case study is the next step.
Frequently asked questions
What’s the difference between voice of customer research and general customer research?
General customer research asks what to build. Voice of customer research asks how to say it. The output isn’t a feature list, it’s a document of exact phrases, sorted by pain, trigger, objection, and win, ready to drop straight into a headline or an email subject line.
How many customer quotes do you need before you can write from them?
Fewer than most people assume. A pattern showing up in five to ten independent sources, in roughly the same words, is usually enough to trust it. What matters more than raw count is independence: five strangers saying the same thing unprompted beats fifty quotes pulled from one enthusiastic customer.
Where do you find voice of customer data if you don’t have customers yet?
Mine reviews for the closest adjacent category, plus any public forum where your future buyer already complains about the problem you solve. Their language about the pain rarely depends on which product they eventually pick, so it’s usable before you’ve made a single sale.
Should you ever polish a customer’s exact words?
Only for grammar that would confuse a reader, never for tone. The slightly rough, specific phrasing is the part doing the persuading. Smooth it into brand-safe language and you’ve deleted the one thing your competitor’s AI couldn’t have written.
Stop guessing at your customer’s language. Go collect it, mine it, and write from it.
The Bottom Line
Your competitors are all drafting from the same models now, so generic copy is easier to spot and easier to skip than ever. The only language a model can’t invent for you is the exact sentence your customer already used.
Collect it from tickets, reviews, and real conversations, tag it by pattern, and write from the rough version, not the polished one. When you can’t run enough of those conversations yourself, hollie can have them at scale and bring back the actual words. Try holito.
