Call · 30 min
MarketingMattia Esposito11 November 202614 min read

Customer language. Where to find it before you write, and how to tell the phrases people repeat from the ones that decide a purchase.

Copy that sells uses the buyer’s words, and those words are already written down somewhere: reviews, forums, complaints, Google searches. Here is the Excel template to collect them, with a worked example.

In brief

Your customers’ words are wherever customers talk without a form in front of them: reviews, forums, comments, complaints and support emails, plus the interviews you run yourself. Google autocomplete adds how they phrase the question, and the Meta Ad Library shows how competitors already talk to them.

The firmest figure comes from MIT. In a 1991 working paper Abbie Griffin and John Hauser found that 20 to 30 one-to-one interviews capture 90-95% of customer needs, and that how often a need is mentioned does not measure how much it matters.

Collecting by hand is fine; scrapers are not. The terms of Google, Amazon and Tripadvisor forbid automated extraction, and a reviewer’s name and photo are personal data: the sheet holds the phrase and its source, never the author.

The Excel template can be downloaded below, with ten columns, a summary that calculates itself and a worked example, also as a PDF.

Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.

The voice of customer template, to download

An Excel sheet with ten columns, one row for each customer phrase, and a summary that counts phrases by source and by level, plus the important rows no competitor uses. Every column comes from a public source, shown in the table below. The second sheet holds a worked example for a made-up olive mill, with made-up customers and competitors.

Worked example of the voice of customer sheet: nine phrases from made-up customers of an olive mill, with source, rewritten need, level, count, importance and competitors’ words
FileContentsLink
Excelto fill in, in Italian

The blank 40-row sheet with drop-down menus, the calculated summary and the example sheet.

capire-i-clienti-prima-di-scrivere-foglio.xlsx

PDFworked example, in Italian

Nine phrases from made-up customers, the count by source and the reading.

capire-i-clienti-prima-di-scrivere-esempio.pdf

ColumnWhat you enterWhere it comes from
Exact phrase

The customer’s words in quotation marks, without correcting grammar or word order.

ASQ, voice of the customer table

Source, with link or date

The type first (interview, review, forum, complaint), then the link or the date, so the summary can count it.

ASQ, list of sources

Situation

Where and when the person says it: before buying, on a first order, after a problem.

ASQ; Harvard Business Review, the circumstance

Need, rewritten

One positive sentence, on a single theme, with no product names.

ASQ, QFD criteria

Job

The progress the customer is after, with its functional, social or emotional side.

Harvard Business Review, 2016

Level

Primary, secondary or tertiary, from the broad need down to the detail.

Griffin and Hauser, the hierarchy of needs

Times mentioned

The number of phrases collected about the same need.

Griffin and Hauser: count it, but don’t weight by it

Importance

From 1 to 5, asked of customers rather than decided by you.

ASQ; Griffin and Hauser

Competitors’ words

How competitors say the same thing in their ads, or “no competitor says it”.

Ad libraries, DSA Art. 39

Personal data?

Yes or no. If yes, remove the name, photo and profile before saving.

GDPR, Arts. 4 and 14

In mid-October, for a positioning project in fitness, I read 1,153 comments under Italian videos in that niche. The word discipline came up 3 times. Around the same time the rapper Wiz posted five words on X, “Discipline is the new luxury”, and against the median of his last hundred posts it got 44 times the likes and 202 times the shares.

Three comments out of 1,153 suggested nobody cared about the theme. The reaction to that post said otherwise, and Google searches pointed the same way. The rule of the sheet comes from there: how often a word appears goes in one column, how much it matters to the buyer goes in another, and the second you ask people directly. The full post of 21 October.

In August, while preparing Itria’s first Meta campaign, I did the same job from the other side: about a hundred ads for AI courses, read before writing my own. The competitors’ words were almost all identical, and for that very reason they showed what nobody was saying. They fill the ninth column of the sheet.

Two readings made only from public words

1,153 comments, discipline 3 times. A word rarely seen in comments can still decide a choice: the same idea, in five words, earned 202 times its author’s usual shares. On Google Trends, in Italy, searches for run club were 22 times what they had been five years earlier.

About a hundred ads, a single lever. Nearly every ad about AI for businesses played on the fear of falling behind and sold a course as the cure. The post of 27 August tells where I started from to write the opposite.

Where to find voice of customer data: the sources, one by one

The best sources are the ones where customers talk without a form in front of them: reviews, forums, comments, complaints and support emails, plus interviews when you run them. ASQ, the American Society for Quality, adds surveys, focus groups, website behaviour, sales and technical reports, and warranty data to its list.

SourceWhat it gives youWhat it doesn’t give youThe rule
One-to-one interviews

The why, the situation, whole sentences.

Figures that represent the whole market.

An hour each, transcribed

Reviews, yours and competitors’

Praise and complaints after the purchase, in the buyer’s own vocabulary.

People who chose someone else and didn’t write about it.

By hand, no scrapers

Forums, groups and social comments

Doubts and questions before buying, voiced among peers.

A measure of how widespread a doubt is.

Phrase and link, not the author

Complaints, emails and chats

Real problems, in the words of the people who had them.

Customers who left without a word.

They stay internal data

Google autocomplete

How people phrase the question.

How many people ask it.

Private window

Google Trends

Comparisons between words, and seasonality.

Search volumes.

Rare terms show as 0

Ad Library

How competitors already talk to the customer.

The customer’s words, or how the ads perform.

Read it, don’t copy it

Reviews matter beyond copy too, and in Italy asking for them now follows new rules, gathered in the page on reviews and artificial intelligence. For forums and Reddit there is the guide to Reddit and forums in AI answers.

How many customer interviews do you need?

Between 20 and 30 one-to-one interviews, according to the most cited study on the subject. At MIT, Abbie Griffin and John Hauser interviewed 30 prospective customers of portable food containers, such as cool boxes, picnic baskets and backpacks, for about an hour each, and had every transcript read by seven analysts. The result was 220 distinct needs.

With 20 interviews they already had over 90% of the needs found with 30, and their final estimate is that 20 to 30 interviews capture 90-95% of needs. A one-to-one interview yielded about as much as a focus group, which costs more. And more analysts reading the same transcripts found more needs, so it pays to have at least two people read the sheet.

The limit has to be stated. The figure comes from two American product categories, food containers and office equipment, in 1991. For a small Italian business it is an order of magnitude to start from. In the sheet, the interviews field works out how many are still missing to reach 20.

Why the most repeated phrases are not the most important

Because in Griffin and Hauser’s data, important needs were not mentioned more often than the rest. The authors put it like this: “frequency of mention does not appear to be a good surrogate for importance”. A need everyone mentions can be a detail, and one mentioned by a few can decide the purchase.

So the sheet keeps two separate columns. Times mentioned is counted while you collect. Importance is asked of customers, as ASQ also advises: at the end of an interview, or in a short survey of people who have already bought, you read out the list of needs and ask for a score from 1 to 5.

In the downloadable example the most repeated phrase is about choosing between a light, fruity oil and an intense one: six times, importance 4. The taste of home for people living far away comes up five times, scores 5, and no competitor uses it. That makes it the first hypothesis to write up, and it stays a hypothesis until a piece of copy puts it to the test.

From sources to copy: customer phrases go into the sheet, become rewritten needs, customers score their importance, and the important rows no competitor uses become the first draft of copy to test The sourcesreviews, forums, interviews The sheetone phrase per row The needrewritten as a positive Importanceasked of customers The texta hypothesis to test by hand, no scrapers10 columnsASQ and HBR Griffin and Hauserto be measured
The path of the words: from the source to the sheet, from the phrase to the need, from the need scored by customers to the first draft of a piece of copy.

From customer statement to need: how to rewrite a phrase

Copy the phrase as it is, note the situation in which it was said, and only then rewrite it as a need. The method comes from the voice of the customer table that ASQ describes within Quality Function Deployment: customers talk about features and rarely about why they want them, and a single statement can yield 5 to 10 needs.

The need is written as a positive, on one theme, without product names. “The tin was dented and I was afraid it would leak” becomes “the oil arrives intact even from far away”. The original phrase stays in the first column: the final copy needs the customer’s words, and the rewritten need is there to count and compare them.

The job column comes from an article by Clayton Christensen and colleagues in the September 2016 Harvard Business Review. Knowing more and more about customers and hunting for correlations, they write, does not explain why people buy. What counts is “the progress that the customer is trying to make in a given circumstance”: the job a customer hires a product to do, with a functional, a social and an emotional side.

The article’s example is a builder of flats for retired people who sold few of them until he grasped that his buyers, with that purchase, were changing their lives. The job column in the sheet captures this: a customer giving the oil as a present wants to make a good impression, and a customer ordering it from another city wants the taste of home.

Google autocomplete and Google Trends: what they really tell you

Autocomplete tells you how people phrase a question, Google Trends how interest changes over time, and neither tells you how many people search. Google writes that predictions “reflect real searches that have been done on Google”, yet they are not simply the most common searches and should not be compared with Trends.

Predictions change with language, location, the interest of the moment and the searcher’s past queries. To collect them neutrally you need a private window with no account signed in. Google also filters them under its policies: dangerous, violent or disparaging predictions don’t appear, so some customer phrases will never show up there.

Google Trends works on a sample of real searches, anonymised and aggregated, and reports it on a scale from 0 to 100 by period and place: two regions with the same value can have very different volumes. Searches made by very few people appear as 0, and for privacy Google adds statistical noise that shows on rare terms. It is not a poll, and in Italy many niche searches stay at zero.

On 21 October, after setting the comments against the searches, I put it like this: “Two rappers don’t make a trend. Google searches, with all their limits, do.” Trends helps to confirm or rule out a direction, and the exact word to use still comes from customers.

Competitors’ words in the Meta Ad Library

Competitors’ ads are public, and for ads shown in the European Union the repository is an obligation under the Digital Services Act. Article 39 requires very large platforms to keep them searchable until one year after they were last presented, with the content, the advertiser, who paid, the period, the audience and the recipients reached. Spend is not included.

In Meta’s Ad Library you search for a competitor by name and read their ads. They give you the words the market already uses with the customer, which go in the ninth column, next to the need they answer. They don’t give you the customer’s words, and they don’t tell you which ad works. The guide to the Meta Ad Library explains how to read them in an orderly way, with its own sheet.

The useful comparison is between the two columns. When a need scores high on importance and the competitors’ column reads “no competitor says it”, you have found open ground. The summary in the sheet counts these rows by itself.

Can you copy reviews and comments? Terms, scrapers and personal data

Reading them and copying out a phrase by hand to understand the market is fine; harvesting them with a program is not, on sites whose terms forbid it. Google bans accessing its services by automated means in breach of machine-readable instructions. Amazon bans data mining, robots and extraction tools. Tripadvisor bans robots, scrapers and artificial intelligence systems, and even manual processes used for purposes its agreement does not allow.

Then there is the GDPR. Under Article 4, personal data is any information about an identifiable person, including through an online identifier: a reviewer’s name and photo, a nickname that can be traced back, a forum quotation with its author. If you store them, Article 14 requires you to inform the person, including the source, within a month, unless that proves impossible or would involve a disproportionate effort.

The simple route is to store the phrase and its source and leave out the author. For this reason the sheet has no author column, and the tenth column asks whether any personal data remains in the row. Interviewing customers is a different matter: you are collecting the data yourself, and the person needs a privacy notice.

One last rule concerns how you use the words you collect. The Italian Competition Authority’s guidelines of 13 October 2026 treat pre-filled templates of positive reviews as manipulative. The phrases in the sheet are for writing your own copy, and should not be suggested to customers when you ask them for a review.

TopicItaly and EUUnited KingdomUnited States
Storing a reviewer’s name and text

GDPR, Arts. 4 and 14: notice including the source within a month, unless the effort is disproportionate.

UK GDPR, built the same way.

No general federal law; individual state laws.

Scrapers on review sites

Banned by the terms of Google, Amazon and Tripadvisor; Amazon also invokes database rights.

The same terms.

The same terms.

Public ad repository

DSA, Art. 39: ads with audience and recipients reached, for one year.

The DSA does not apply; platform rules apply.

Platform rules apply.

Autocomplete and Trends

The same everywhere; in Italy low volumes mean more zeros and more noise.

The same.

The same, with higher volumes.

From the sheet to the copy: how to use voice of customer

Start from the rows scoring 4 or 5 that no competitor uses: the customer’s phrase becomes the first draft of a headline, to be put to the test. Use the phrase in its own words, even when they are less polished than yours, because it is the one the customer recognises.

From there the work moves to the copy. The starting point is the conversation already going on in the customer’s head, covered in the page on the conversation in the customer’s mind. How much the customer already knows about the problem and the solutions decides where the copy begins, as the Schwartz levels of awareness explain.

Openings are chosen with the examples of hooks and headlines, and for advertising there is the guide on how to write Facebook and Instagram ad copy.

The full picture is in the guide to copywriting for small businesses, and every marketing guide is gathered in the index of marketing for small businesses. When the same phrase holds up across several pieces of copy, it becomes material for brand positioning. Before the customers’ words comes the size of the market, with the public sources in market research with AI.

The sheet needs redoing: words shift with the seasons and when products and competitors change. In September, analysing the websites of 142 Italian companies that also sell abroad, I saw language cause problems 38 times: the words of people buying from another country need collecting separately, in their own language. The post of 7 September.

Questions and answers

What is voice of customer in copywriting?

It means collecting the words customers use to describe a problem, a wish or a purchase, and putting them to work in your copy. In quality management, where ASQ places it, the voice of the customer is used to design products and services; in copywriting it is used to write headlines, ads and pages in the buyer’s own language.

How many customer interviews do you need?

Between 20 and 30 one-to-one interviews, as a starting point. In Griffin and Hauser’s study (MIT, 1991), 20 to 30 interviews captured 90-95% of customer needs, across two American product categories. For a small Italian business it is an order of magnitude rather than a guarantee.

Can you copy your competitors' reviews?

Reading them and copying out a phrase by hand to understand the market, yes. Harvesting them with a program, no, on sites whose terms forbid it, such as Google, Amazon and Tripadvisor. A reviewer’s name, photo and profile are personal data: the sheet keeps the phrase and its source, never the author.

Does Google autocomplete show the most frequent searches?

No. Google writes that predictions come from real searches but do not simply display the most common ones, depend on language, location and past searches, and should not be compared with Google Trends. They tell you how people phrase a question, not how many people ask it.

Where can you find customer language if you have no customers yet?

From your competitors’ customers: reviews, forums, comments under their posts and questions in autocomplete. Your competitors’ own words are in the Ad Library. After that you need interviews with prospective customers: Griffin and Hauser interviewed prospective customers too.

Notes on sources

  1. ASQ, American Society for Quality, Voice of the Customer, read on 11 November 2026: the sources of the voice of the customer, the table with statement, situation and rewritten need, and priorities set by customers.
  2. Abbie Griffin and John R. Hauser, The Voice of the Customer, MIT Sloan working paper, October 1991, read on 11 November 2026; source of the quotation on frequency. Published version: Marketing Science, 12(1), 1993, not read.
  3. Clayton M. Christensen, Taddy Hall, Karen Dillon, David S. Duncan, Know Your Customers’ “Jobs to Be Done”, Harvard Business Review, September 2016; source of the quotation on progress.
  4. Google, How Google autocomplete predictions work (source of the quotation) and FAQ about Google Trends data, read on 11 November 2026.
  5. Terms of use: Google, effective 30 July 2026; Amazon, last updated 28 November 2025; Tripadvisor, US version effective 15 July 2025. Read on 11 November 2026.
  6. Regulation (EU) 2016/679, GDPR, Art. 4(1) and Art. 14, in the text published by the Italian Data Protection Authority.
  7. Regulation (EU) 2022/2065, Digital Services Act, Art. 39 (ad repository).
  8. Italian Competition Authority (AGCM), Guidelines on online reviews, decision of 13 October 2026, no. 32067, para. 25.
  9. The readings in the box come from Mattia Esposito’s LinkedIn posts of 27 August, 7 September and 21 October 2026. The example sheet uses a made-up company, customers, phrases and figures.
·Who does it for you

The sheet collects your customers’ words. Choosing the one to build the copy on is the work that follows. Positioning is built starting from the market.

Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.