Messaging From Customer Language: Turning Reviews and Threads Into Copy

What does voice of customer copywriting mean?

You find the message instead of inventing it. The steps are to collect language, tag it, group it into themes, count how many different customers say each thing, write from the result and test the draft.

collect phrases → tag them → cluster into themes → count customers → draft → test with real readers

• Reviews and community threads hold language that customers wrote without a marketer in mind. Interviews, calls and tickets add depth.
• Pull phrases word for word, tag each one and count distinct customers, not mentions.
• The worked example below runs three statements through every step. It ends with one medium theme and four thin ones, which shows that a small set yields hypotheses, not headlines.
• Test drafts with five to ten readers who match your buyer, and ask permission before quoting anyone.

Why does copy written from the inside miss?

Copy written from the inside misses because a team describes what it built in the vocabulary it uses in meetings, and a stranger may not share that vocabulary. AskTheCustomer cites a May 2018 TrustRadius poll of 769 software buyers in which about 3 in 4 called vendor marketing language jargon or fluff. That poll is old, and this page read it only through AskTheCustomer, so treat it as a pointer rather than a current benchmark.

Voice-of-customer research is wider than copy. Copyflight quotes CXL's Peep Laja defining it as discovering what customers want and need through qualitative and quantitative research. Copywriting uses the qualitative slice: the wording.

AskTheCustomer separates this from a satisfaction score, since a score says how people feel and a phrase says how they talk. Daniel Doan argues that copy echoing a reader's own thoughts builds trust. That is his opinion, and none of the five pages read here offers controlled evidence that customer-language copy lifts conversion.

Where does customer language come from?

Customer language comes from two groups of sources: records you already hold, such as interviews, calls and tickets, and public places where customers talk without you, such as reviews and community threads. The table brings together what the five guides say about each source and the caution each one attaches.

The guides disagree on interviews. Doan puts them first and says spoken answers are more spontaneous than written ones. PhraseMine says interviews trigger a polished performance, and that surveys filter language.

PhraseMine sells a tool that does Reddit research, so it has a reason to favour anonymous sources. Doan sells copywriting services, so he has a reason to favour methods he performs. Kanso's reading is to use both kinds and keep them apart, which AskTheCustomer also advises for current customers, prospects and the open market.

What should you read in reviews and threads?

Read the middle of the review range and the replies in communities where your buyer talks to peers, because that is where the explanations sit. PhraseMine's view is that five-star reviews tend to be vague, one-star reviews tend to be rants, and three-star reviews say what worked, what did not and what the reviewer wished were different. AskTheCustomer, citing Wynter, gives the same warning from the other direction: extreme reviews can be skewed.

For threads, PhraseMine suggests searching Google with the site name Reddit plus a phrase of struggle, such as "does anyone else struggle with" or "I'm so frustrated with". It also recommends checking adjacent communities, since a fitness product's buyers may talk in running and weight-loss communities as well as fitness ones. Doan makes the case for threads in terms of audience: people there do not know a brand is reading.

Two habits keep this from turning into scrolling. Work in a fixed session: PhraseMine suggests a timer and one hour, and says that yields 15 to 25 usable phrases, which is a vendor's estimate that Kanso has not tested. Then return briefly each week, which PhraseMine puts at 20 minutes in saved communities.

If you want a routine for finding the conversations in the first place, Finding Buyer Conversations on Reddit Without Spamming lays out a weekly 30-minute one.

How do you build a phrase bank, step by step?

You build it in six steps, and the first three produce the bank: pick the page, gather the evidence, and pull phrases into a table with a source and a theme on every row. The sequence follows AskTheCustomer's five steps, with Kanso splitting the last one into drafting and testing.

1. Pick one asset, one reader and one question. A homepage headline, a launch email or a pricing page is enough, with one named reader and a question such as what makes someone start looking for this. The question decides which phrases are worth keeping, and a brief that records it, such as the content brief template with a filled example, keeps the page tied to one reader.
2. Gather the evidence and keep each kind apart. Put current customers, prospects and the open market in separate tabs, so a difference between them stays visible.
3. Pull phrases word for word. AskTheCustomer lists six kinds of language to look for: the problem in their life, what they have tried, what made them start looking, what they hoped for, what they valued and what made them hesitate. Do not tidy the wording, because the clumsy version is often the useful one.
4. Group by theme and count customers. The next two sections show how, on a small set.
5. Write from the themes. Put the problem in the headline, the customer's vocabulary in the explanation and the top worry in a short objection line.
6. Test. Show the draft to five to ten people who match your reader and ask what the company does, in their own words.

The table below merges the phrase bank columns in AskTheCustomer's template with the extra fields PhraseMine suggests. The choice of seven columns is Kanso's, and PhraseMine also groups phrases by Eugene Schwartz's awareness stages, which is optional here.

Worked example, part 1: how do you pull and tag phrases?

You pull phrases by reading each statement once, copying the fragments that carry a problem, an alternative, a value or a hesitation, and tagging each with its source and kind. The example below does that for three statements, and it is hypothetical on purpose.

The statements are the ones Daniel Doan invents for a banana-snack brand in his guide: a review on the brand's own site, a subreddit comment about sourcing, and a comment from someone who struggled to eat healthier. They are not real customers, and Kanso did not collect them.

The phrases below are shortened restatements, because the original wording belongs to Doan. In your own bank you would paste the exact words.

The asset is a product page headline for a banana chip snack. The reader is a shopper who has bought healthy snacks before and been let down. The question the copy must answer is what made them look for another snack.

Nine rows came from three people. Voice 1 supplies rows 1 to 4, which is 4 of 9 rows, so a tally of rows would already lean toward one reviewer.

Worked example, part 2: how do you cluster and count?

You cluster by listing each theme with the rows that fall under it, then counting the different voices behind it, because rows and voices give different answers. In this set the two counts disagree for three of the five themes.

The arithmetic is short. Row 3 carries two themes, so the 9 rows produce 10 theme tags (3 + 2 + 1 + 1 + 3 = 10).

Counting rows, themes A and E tie at 3. Counting voices, theme A has 2 and theme E has 1, because all three of E's rows come from Voice 3. Theme B has 2 rows but only 1 voice, because both of its rows come from Voice 1.

AskTheCustomer states the rule this way: ten mentions from one account is one signal, and ten customers saying it in different words is a pattern. It rates a theme strong, medium or thin without giving numeric cut-offs. The ones below are Kanso's working assumption, loosely anchored to the five call transcripts AskTheCustomer says Wynter suggests as a starting point, and you should replace them with your own once you have data.

Under these cut-offs, theme A is medium and the other four are thin. Even theme A rests on two statements from one article's examples, so it is a candidate to test, not a finding. That is the main thing the example shows: a handful of statements can organize your questions, and it cannot settle your message.

Worked example, part 3: how do themes become draft copy and a test?

Themes become draft copy by putting the customer's problem in the headline, their vocabulary in the explanation and the worry in an objection line, and then checking each line against what is true of the product. The table pairs each theme with a team-language version, a customer-language draft and the check to run before publishing.

Only theme A earns a headline test, and the draft keeps the customers' own words for the problem: bland and chalky. AskTheCustomer adds a vocabulary check. If customers say report and you say analytics dashboard, use report.

Doan's own treatment of the same review is different. He reads it as a prompt to improve the product and promote the change, and he warns against claiming ethical sourcing if the company does not deliver it. That warning is why the table above makes truth the first check on every line.

Then test. AskTheCustomer suggests showing the draft to five to ten people who match the reader and asking what the company does, in their words, and using their version if it is clearer than yours. Rows D and E go into the next round of interviews or thread searches as questions.

How much customer language is enough?

No fixed number is enough, and the guides give starting points instead of thresholds. AskTheCustomer says there is no magic number, suggests five or more calls plus a sample of tickets, and says to keep reading until new sources stop adding new phrases. It also cites Wynter's advice to go deep on two or three communities.

Measured against those starting points, the worked example is small. It has 9 phrases, against the 15 to 25 PhraseMine says an hour of focused reading yields, so it is short by at least 6. It has 3 voices where AskTheCustomer suggests 5 or more calls, and it has no tickets or calls at all.

A theme that rests on one or two voices should be treated as a hypothesis. AskTheCustomer says to talk to more customers before building copy on it, and also to note what is missing, since a favourite feature nobody mentions is a finding too.

Can you use customers' words directly in copy?

You can use them directly only with the customer's permission and only if the quote is accurate, and often a paraphrase that keeps their vocabulary does the job. AskTheCustomer says never to invent or polish a quote and suggests removing names and details that could identify a person or company.

Doan is more relaxed. He says responses gathered through interviews and surveys can be used word for word, and he suggests an incentive for taking part. The two views fit together if the people you asked agreed to be used.

Public posts are different, and this part is Kanso's view rather than a published rule. A pseudonymous post from a stranger is evidence of how people talk, not a testimonial. Borrow the phrase and drop the person, and never present it as an endorsement.

How does customer language relate to a brand voice?

Customer language can ground a brand voice, but it does not replace one. A voice also has to reflect what the brand wants to be known for, and the phrase bank supplies the customer half.

In a Copyhackers tutorial presented on March 9, 2021, Joanna Wiebe argues for making your voice their voice, and says that does not mean pandering. Her objection is to writers who lay voice on thickly in an effort to sound like something. She illustrates it with GitPrime, now Flow by Pluralsight, where a line from a founder interview about engineers led to the positioning "Engineers build business"; this is her account, and Kanso has not verified it.

In Kanso's analysis, a phrase bank feeds four parts of a voice document: the words customers use for the product, the words the brand avoids, the worries to answer and examples of lines that sounded right. Positioning decides which alternative and which segment the voice speaks to, and Positioning for Early-Stage SaaS works through that with two invented companies.

What are the common mistakes, and when do other methods fit better?

The common mistakes are building a headline on one loud customer, mixing customers with prospects, reading only the happiest and angriest reviews, polishing a phrase until it sounds like the team again, copying words without their context and quoting without permission. Those six come from AskTheCustomer. The worked example adds two more.

• Counting rows instead of voices. In the example, theme E and theme A tie on rows and differ on voices.
• Writing a line the product cannot back up. Doan warns against claiming ethical sourcing without delivering it, and the same applies to crispness or sweeteners.

Customer language is the wrong tool when the question is a number or a choice between finished options. AskTheCustomer recommends a survey for numbers across many people, interviews for depth on why, and an A/B test to pick between finished options. It adds that customer evidence supports your judgement and does not replace it.

Can AI help find customer language?

AI can speed up reading and sorting large piles of transcripts, but it can invent or smooth over wording, so every phrase it returns needs a source you can open. AskTheCustomer advises asking for the exact record behind each phrase, checking a sample against the sources and treating the themes as a starting point.

Kanso's Reddit Agent finds relevant threads and drafts replies for you to review and post yourself. It is a way to find the conversations, not a phrase-mining tool, so reading the threads and copying the words into a phrase bank is still your job. Kanso does not post for you.

What is documented and what is our analysis?

The documented list holds what the five pages say, attributed to them, and the analysis list holds what Kanso adds. Neither list is a test result, because Kanso ran no tests for this page.

• Documented: AskTheCustomer reports a May 2018 TrustRadius poll of 769 software buyers in which about 3 in 4 called vendor marketing language jargon or fluff.
• Documented: AskTheCustomer, Doan and PhraseMine describe collecting, tagging and grouping customer phrases. AskTheCustomer adds six kinds of language and the rule to count customers, not mentions.
• Documented: Doan and PhraseMine disagree on whether interviews or anonymous threads give truer language, and PhraseMine claims 15 to 25 usable phrases from an hour of reading.
• Documented: none of the five pages offers controlled evidence that customer-language copy lifts conversion.
• Analysis: the worked example is hypothetical, built from statements printed in Doan's guide, and the shortened phrases are Kanso's restatements.
• Analysis: the thin, medium and strong cut-offs, the seven-column phrase bank and the draft lines are Kanso's and untested.
• Analysis: treating a pseudonymous public post as evidence of language and not as a testimonial.

Facts that change should be rechecked before you rely on them, including the vendor claims, which were read on 26 October 2026.

Voice of customer copywriting: frequently asked questions

Sources

• Voice of customer copywriting in 5 steps, AskTheCustomer
• How To Effectively Capture Your Customer's Voice In Your Copywriting, Daniel Doan
• Voice of Customer Research for Copywriters, PhraseMine
• How to Use Voice of Customer in Your Brand, Copyhackers
• Voice of Customer research is the secret weapon to better copywriting, Copyflight