A working process for mining customer reviews for ad copy: where to pull reviews, how to extract exact phrases, and the filter that turns a review line into a hook.
You are three creative refreshes into the quarter and the hook doc is running on fumes. Every new line is a remix of the last one: "Struggling with X?", "What if X didn't have to be this hard?". Meanwhile, last night, someone who bought your competitor's product wrote four sentences on Trustpilot that would out-hook everything in that doc. They described the exact moment the problem became unbearable, in words no copywriter in a brainstorm would invent.
Mining customer reviews for ad copy is the cheapest research in this job. Here is the workflow I use, start to finish: where to pull reviews, how to get usable phrases out of them, and the filter that decides which line becomes a hook.
The takeaways
Four public places cover most products: Amazon for physical goods, Trustpilot for services and DTC brands, G2 for B2B software, and the App Store and Google Play for apps. Start with your own reviews if you have volume. If you don't, start with your biggest competitor's: their reviewers are the same buyers you are targeting, describing the same pains, in public.
The platforms read differently, so use them differently. Trustpilot skews toward the service wrapper: shipping, support, refunds, "did this company treat me fairly". G2 is comparative by design ("What do you dislike?" is a built-in prompt), which makes it the best source of objection language. App-store reviews are short and emotional, written seconds after a feeling, which makes them hook material more than body-copy material.
Volume matters less than you'd think. Thirty reviews read carefully beat three hundred skimmed.
Copy lines word for word into one document, with the star rating and source next to each. The moment you paraphrase, you sand off the phrasing that made the line work. "I stopped checking it obsessively after the first week" becomes "reduces anxiety", and the ad dies in that rewrite.
Then sort the pile by the job each line does. I tag four: the before-state (life with the problem), the outcome (what changed, in their words), the objection (why they almost didn't buy), and the surprise (what they didn't expect). From each review, two to four short phrase markers are usually all that's worth keeping. The rest is filler even when the review is glowing.
One habit from Copyhackers' old Amazon-mining playbook still holds: read the 3-star reviews first. Five-star reviews gush and one-star reviews rant, but the middle band weighs pros against cons in the same breath, and that tension is what believable copy sounds like.
When you rank the collected lines, rank by intensity, not frequency. The pain mentioned most often is usually the obvious one your competitors already lead with. The pain described in the most charged language is where the unworked angles live.
A review line earns a hook slot when it passes three checks: it's in first person, it contains one concrete specific (a number, a timeframe, a scene you can picture), and it carries a before-state. "Great app, works well" fails all three. "I didn't realize how much I dreaded Sunday nights until they stopped" passes all three, and it's already a hook. You mostly just trim it.
The rewrite is lighter than people expect:
Keep the grammar a little crooked. Reviewers write "it just... works??" and "lowkey fixed my mornings", and that texture is what stops the scroll, because review language reads as a person talking. Resist cleaning it into marketing English.
One honesty rule: anything you present as a quoted review (attributed, in quotation marks) has to be real and yours. Competitor reviews are for language and angles only. Quoting one as your own social proof crosses into fabrication, and that catches up with brands.
Review-mined hooks fail too. They just fail less embarrassingly, because at least one human cared enough to write the underlying sentence. Put five to ten of them into a normal creative test and hold them to the same bar as everything else, including the bar for sample size: a review-born hook can ride a lucky three days like any other ad. I wrote separately about how I tell a winner from a lucky streak.
When I was running around €150k a month in paid social, the review-mined batches were the ones I could defend in hindsight: when one bombed, I at least knew the angle came from a real buyer and could trace why it didn't transfer. That traceability is why this workflow was one of the first things I automated when I built Adscalr. Its audience research pulls from five voice-of-customer sources (Reddit, Amazon reviews, the App Store, Google Play, and custom forums), seeded with your URLs and your competitors', and it keeps two to four exact phrase markers per quote, plus the emotion and language register, so the wording survives all the way into the creative work. If you want to see where that sits in a full research stack, the audience intelligence page walks through it.
This is the thinking behind Adscalr.
See the product →