How to turn customer reviews into a creative strategy for your ads
This is the fifth article in a series. Part 1: How to analyze 1,000+ competitor ads with AI. Part 2: How to stress-test AI competitor ad research before briefing your next ads. Part 3: How to break down competitor video ads beat by beat. Part 4: How to turn a competitor’s ad structure into new ad concepts (without copying the ad).
I ended the last article with a warning: the situation in your script is only an assumption until your own customers’ words back it up. A competitor’s ad shows you what that competitor decided to put on screen. It doesn’t tell you why.
In this one I’ll show you how to get those words. We’ll take the public reviews of one app, work out who is writing them and what they were struggling with, and turn that into a strategy brief you could hand to a scriptwriter.
The example uses 3,207 public App Store reviews of an AI language tutor app, in the same category as the ads from the earlier articles. We don’t work with this app, so treat this as an exercise on public data.
Why do customer research before you write the ad?
Because you can’t blindly copy a competitor. Their ad was built for their customers and their product. If you lift the situation without checking it, you can end up paying to produce an ad about a problem your customers don’t have.
Take the ads we broke down in Part 3. They tell the same story three times: a skilled worker is held back at work by their English, practises with an AI tutor, and ends up with a better role or more income.
Now suppose you run creative for a different AI language tutor and you like those ads. Before you write a workplace story of your own, go and read what your customers say. When we looked closely at a sample of 450 reviews for this app, only two mentioned a job or an interview.
So a workplace ad for this app could be well made and still say very little to the people who use it.
None of this means the competitor got it wrong. They can see results we can’t, and their customers may simply be different people. It just means their ad isn’t evidence about your customers.
Which customer reviews should you collect, and how many?
Collect every recent review you can get, good and bad, from each country you advertise in. If your product is too new to have many, use the reviews of your closest competitor and treat whatever you find as a starting assumption.
What we collected
| What | Detail |
|---|---|
| Source | Apple’s public App Store review feed. It returns up to 500 recent and 500 most helpful reviews per country store. |
| Country stores | 12: United States, United Kingdom, Canada, Australia, India, Mexico, Spain, Colombia, Argentina, Chile, Peru and Brazil. |
| Reviews | 3,207, dated April 2023 to November 2026. Most are from 2025 and 2026. |
| Languages | English, Spanish and Portuguese. |
| Ratings | 1,960 with four or five stars, 1,078 with one or two stars, and 169 with three. |
App Store Review Analysis
You don’t need to write any code for this step. Give the skill an app ID and the country stores you care about. It downloads the recent App Store reviews, keeps a record of what it could and couldn’t fetch, and then walks your AI assistant through the analysis in this article.
-
Downloads: recent App Store reviews, up to 500 per country store.
-
Also accepts: App Store and Google Play exports you already have.
-
Cost: free. It runs with your own assistant.
Most reviews say nothing specific
Be ready to set most of them aside. A two-word five-star review, or a complaint about a charge, tells you nothing about who the customer is.
-
Of the 1,960 positive reviews, only 399 run longer than 120 characters.
-
We gave the AI a random sample of 450 of the longer reviews: 300 positive and 150 negative.
-
In that sample, about 66 describe something about the reviewer’s own life. That’s roughly one in seven.
We started with 3,207 reviews and ended up with a few dozen we could really use. That’s why it’s worth collecting many more than you think you’ll need.
What prompt turns customer reviews into a creative strategy?
Hand the AI the reviews along with their star rating, country and date. Ask it four questions about the customers before you ask for any strategy, and have it count the evidence behind each answer.
What to ask once you have the reviews
If you used the App Store Review Analysis skill, it has already downloaded the reviews and prepared the file, and it will find angles and briefs on its own. The brief below adds the strategy questions this article is about: who the enemy is, what the customer believes, and what change they want.
Replace the bracketed parts with your own reviews and competitor research.
I want to build a creative strategy for our ads from what customers say, not from what competitors show. Attached are [number] app store reviews of [our app, or the closest competitor], with star rating, country store and month.
Read all of them before you group anything. Then answer four questions.
1. Who is writing? Group the situations people describe when they explain why they use the app. Give a count for each group and two or three short examples. Count only reviews that actually describe the situation. Say how many reviews say nothing specific.
2. What were they struggling with before? Keep their own phrases, in the original language with a translation. Do not tidy them into marketing language.
3. What changed for them? Separate what the product did (a feature) from what they could then do (the change in their life).
4. What makes people hesitate or regret it? Read the one and two star reviews for objections a new customer would share, not for bugs.
Then propose three strategic directions. For each one give: who it is for, the enemy (the specific failure or obstacle they recognize), the belief that makes the product make sense, the change they want, and five phrases from the reviews we could build on. Give the number of reviews that support each direction, and say plainly where support is thin.
Compare the three directions on: strength of evidence, how specific the tension is, whether the product actually does the thing, and whether competitors already say it. For reference, the competitor ads we studied tell this story:
.Recommend one direction and say what you would need to see to change your mind.
Do not infer anything about people who did not write a review. Do not infer ad performance. Treat any group with fewer than 10 reviews as anecdote and label it that way.
A few of these instructions do more work than they look like they do.
“Count only reviews that actually describe the situation” is there because, without it, the model will happily sort every five-star review into a persona. You get neat segments with big numbers next to them, when almost none of those reviewers said anything about themselves.
“Keep their own phrases” is there because a model’s summary of a fear reads like a positioning statement, while the customer’s own sentence reads like something a person would say. For a script, you want the customer’s sentence.
The rule about fewer than 10 reviews guards against the one vivid review. A long, moving story about learning a language for a partner’s family can feel like a whole strategy when it’s really one person’s experience.
And asking the AI to separate the feature from the change keeps your payoff realistic. “It corrects me as I speak” is a feature. “I can now talk to people who only speak English” is a change in someone’s life. You’ll find plenty of the first kind in reviews and very little of the second.
What the reviews said about the people using the app
The two biggest groups are people who are afraid to speak in front of others, and people who studied for years and still can’t hold a conversation. Work, travel and family goals hardly come up.
Counts in this section come from the AI’s reading of the 450-review sample. We checked the two largest groups by hand, and the results are further down. Review content is paraphrased and translated.
Who is writing
| Situation they describe | Reviews | What they say |
|---|---|---|
| Afraid or ashamed to speak in front of people | 25 | They avoid native speakers because they’re afraid of looking foolish. They feel judged when they practise with people. One was too embarrassed to speak in class. |
| Have a base and want to unblock it | 15 | They already know some of the language and want to get it moving, or bring it back after years without practice. |
| Studied before and still could not converse | 14 | Three years on a streak app, or five years of school courses, and none of it prepared them for a conversation. |
| Nobody to practise with | 10 | They don’t know anyone who speaks the language, or tutors are expensive and awkward to arrange. |
| Family or heritage language | 4 | Anecdote. |
| Exam or certificate | 3 | Anecdote. |
| Travel | 3 | Anecdote. |
| Work or a job interview | 2 | Anecdote. |
These groups overlap. Several of the most detailed reviews describe both the fear and the years of study.
Most of the people in the fear group write in Spanish or Portuguese and are learning English. That becomes relevant when we decide who the strategy is for.
What they struggled with before
-
Fear of being judged. A reviewer in Mexico is scared to talk to native speakers in case she makes a fool of herself. A reviewer in Brazil is ashamed to speak incorrectly to a real person. Another says that practising with people makes her feel judged.
-
Study that never turned into speaking. One reviewer studied German for five years, fifty years ago, and says none of it prepared him for a conversation. Another says speaking was the missing ingredient in her French. A third had already given up.
-
No partner, or the cost of one. One reviewer never found anyone to speak Italian with after university, and the ability faded. Another says that depending on other people to learn a language is painful.
What changed for them
| What the product did | What they could then do |
|---|---|
| Corrected them as they spoke (about 26 reviews). | Spoke more than ever before. One says it’s the most Japanese they have spoken in their life, and they used to live in Japan. |
| Felt like a person or a private tutor (about 20). | Lost the fear of pronouncing words or starting a conversation. |
| Talked about any topic they chose (about 16). | Moved up a level, by their own account. |
| Made them speak, and was there at any hour (8 each). | Went from reading a language to speaking it. |
The right-hand column doesn’t have much behind it. Only one review in the sample says the person went on to talk to real people. Almost every other change happens inside the app, and one person who used it for five months says they saw no real improvement.
What makes people hesitate or regret it
| Objection | Reviews (of 150 negative) | What they say |
|---|---|---|
| Trial and billing | About 56 | They were charged during or straight after the trial, or a full year was taken in one payment, or a promised reminder never came. |
| “The ad said free” | About 27 | They answered a long questionnaire and then hit a paywall. One says they would probably have signed up if the price had been shown first. |
| Too hard for beginners | 21, plus about 10 fans who agree | Complete beginners felt stupid and couldn’t put three words together. Fans say the app suits people who already have a base. |
| Pronunciation and language mix-ups | About 16 | Mostly about languages other than English. |
| Product does not match the ad | About 12 | Avatars, hands-free use and listed languages that weren’t as shown. |
These objections are useful for the creative too. The second and third rows tell you what your ads shouldn’t promise and who they shouldn’t try to attract.
How do the reviews compare with the competitor’s ads?
They describe a different person with a different problem. In the competitor’s ads, other people shut the main character out. In the reviews, people are mostly holding themselves back.
| In the competitor’s ads | In this app’s reviews | |
|---|---|---|
| Who | A skilled worker with years in the job and little English. | Someone who has studied the language and understands it, but doesn’t speak it. |
| What holds them back | A boss or a customer. They are sent back to the kitchen. | Their own fear of looking foolish in front of another person. |
| What they tried before | Not shown. | Apps, courses and years of classes that never got them talking. |
| The objection | “What if I get it wrong?” and “Is it free?” | “The ad said free.” “I have tried apps for years.” |
| The payoff | Tips, a new role, recognition at work. | Speaking at all, and being able to hold a conversation. |
| Evidence in the reviews | 2 reviews. | At least 35 reviews that we confirmed by hand. |
A few cautions before you read too much into this table:
-
The ads may still be right for the company running them. We can’t see their results or their customers.
-
Reviewers aren’t prospects. People who write reviews have already installed the app, so they aren’t the same people your ads need to reach.
-
One beat shows up in both places. The fear of getting it wrong appears in the ads and in the reviews, so you can carry that hesitation across with some confidence.
What the table does show is that a workplace story would be a borrowed situation for this app, when its own customers have described a different one.
Three strategic directions from the reviews
Each direction says who it’s for, the enemy that person would recognize, the belief that makes the product make sense, and the change they’re after. These three came back from the brief above.
| A. Mistakes cost nothing here | B. You studied. You still can’t hold a conversation. | C. A speaking partner whenever you are ready | |
|---|---|---|---|
| Who it is for | People who avoid speaking because a person is listening. | People with a real base from apps, school or courses, who read and understand but don’t speak. | People with nobody to practise with, or who find tutors costly. |
| The enemy | The moment of looking foolish in front of a native speaker, a class or a teacher. | Study that never got them talking: the streak app, the course, the years of classes. | Finding, scheduling and paying a person. |
| The belief | You only lose the fear by speaking a lot, and you speak more when nobody is judging. | Speaking is its own skill, and you only build it by speaking. | What was missing was hours of speaking time, not more lessons. |
| The change | Speaking without freezing, and eventually with people. | Finally holding a conversation. | Daily speaking practice. |
| Support | 25 reviews, plus about 10 on confidence. | 26 reviews across two overlapping groups. | About 19, of which 10 describe a situation. |
| Where it is thin | One review reports speaking to real people afterwards. Three say the app itself made them anxious. | Results are self-reported. One user without a real base failed badly. | It leads with price and access, which is where the billing complaints sit. |
How do you choose between strategic directions?
Compare them on the same four questions, make a call, and write down what would change your mind. The evidence usually won’t make the choice for you.
| Question | A. No judgment | B. Studied, can’t speak | C. Partner on demand |
|---|---|---|---|
| How strong is the evidence? | 25 reviews, and they are vivid. | 26 reviews. | 10 that describe a situation. |
| How specific is the tension? | The most specific of the three: one moment anyone would recognize. | Specific: a named failure, with years attached. | Less specific. It’s a convenience. |
| Does the product do it? | Yes, for people with a base. Some beginners report the opposite. | This is the strongest. Even unhappy reviewers agree the app gives you conversation practice. | Yes for access, but price claims run into the billing reviews. |
| Do competitors already say it? | Not in the ads we studied. | Not in the ads we studied. | Not in the ads we studied. |
The direction we would test first
We’d lead with B and let it carry the emotion of A. The two have almost the same number of reviews behind them, by the AI’s count and by ours, so this is a judgment call. Here’s what tipped it:
-
It picks out the right audience. B is aimed at people who already have a base, and the app’s fans say that’s who it suits. A on its own would also pull in complete beginners, who wrote most of the “too hard” reviews.
-
It rests on the safest product claim. Even unhappy reviewers agree that the app gives you conversation practice.
-
A fits inside it. Fear of being judged is the reason these people never spoke, so the fear can be a scene in a B story.
What would change our mind
-
A recount. If a second reading found the fear group to be clearly larger, we’d lead with A.
-
Beginners doing well. If product data showed complete beginners succeeding, A would be safe to lead with.
-
The language mix. The support for B comes mainly from people learning English. If most paying customers are learning other languages, we’d test again.
-
Results outside the app. If a larger set of reviews showed people getting results at work, the workplace story would deserve another look.
Check the AI’s counts before you rely on them
We checked the AI’s work in two ways. First, we read the reviews behind the two biggest groups ourselves.
-
The citations held up. Every review the AI cited said what the AI claimed it said.
-
The counts were on the high side. The AI counted 25 reviews in the fear group and we could confirm 17. It counted 29 across the two study groups and we could confirm 18.
-
The order didn’t change. Both groups are still much bigger than work, travel or family.
So I’d treat an AI’s count as the most it could be, until someone has read the reviews behind it.
Second, we ran a plain keyword search in three languages across all 3,207 reviews. It’s a cruder method than reading, and it pointed the same way.
-
Fear, shame or confidence: 49 of the 399 detailed positive reviews, about 12%.
-
A job, interview or meeting: 11 of the 399, about 3%.
-
Trial, charges or refunds: 437 of the 1,078 negative reviews, about 41%.
It’s the same habit as in Part 2: spot-check the AI before you rely on it.
What goes in a creative strategy brief?
One page that tells a scriptwriter who the ad is for, what that person is up against, and what the ad can and can’t claim. Here’s the brief for direction B.
| Field | For this app |
|---|---|
| Who it is for | Adults who have studied a language for years and understand it, but go quiet when it’s their turn to speak. Most of them are learning English. |
| The enemy | Study that never got them talking, and the moment of looking foolish in front of someone. |
| The belief | Speaking is its own skill. You build it by speaking a lot, somewhere mistakes don’t cost you anything. |
| The change | Being able to hold a conversation. We aren’t promising fluency, a promotion or a new life. |
| Their words | Brazilian reviewers call it “destravar”, to unblock. Others talk about losing the fear, speaking without shame, and speaking being the missing ingredient. |
| What the product does | Open conversation on any topic with an AI tutor, with correction as you speak, at any hour. |
| Do not say | “Free”. Anything aimed at complete beginners. Pronunciation coaching. Any result outside the app that we can’t show. |
| Still an assumption | That people who haven’t installed feel the same way as reviewers do, and that the change carries over to conversations with real people. |
How does the strategy change the ad concept?
It changes what goes into each beat. The eight-beat structure from Part 4 stays the same, but the situation, the objection and the payoff now come from the reviews.
| Beat | Borrowed from the competitor | Built from the reviews |
|---|---|---|
| Opening barrier | A boss sends her back to the ovens. | He understands every word of the conversation. When it’s his turn, he says nothing. |
| Concrete cost | A few coins instead of tips. | Years of classes and a long app streak, and someone else still answers for him. |
| Hesitation | “What if I get it wrong?” | “I have done three years of lessons on my phone.” The price is shown before he signs up. |
| Product use | Practises one work phrase. | Talks with the tutor about something he cares about, gets corrected, and keeps going. |
| Demonstrated change | Takes the order from the same customers. | The same conversation comes round again. This time he answers, with mistakes. |
| Recognition | The front counter is hers. | Nobody comments on his mistakes. The conversation just carries on. |
One of the two example scripts in Part 4 was written before we had read these reviews. Its barrier holds up: a father who understands the teacher and then loses his words. Its setting doesn’t, because family goals were only anecdote here. Catching that kind of thing before production is why this research is worth doing.
In the next article, we’ll write the full script from this brief.
What customer reviews cannot tell you
Reviews tell you what some of your existing customers chose to write down. They can’t tell you which ad will perform.
-
They come from people who installed. The people your ads need to reach haven’t installed yet.
-
They are self-reported. A reviewer who says they moved up a level hasn’t been tested.
-
They come from one store. We only read App Store reviews, and Android customers may be different.
-
They may be about someone else’s product. If you used a competitor’s reviews, your own customers may not match.
So before production, check the direction against other sources: customer interviews, survey answers, support conversations and the comments on your own ads.
The brief is still a hypothesis, and you’ll only find out whether it holds by testing it.
Published