The best line in your next ad has probably already been written. Not by a copywriter. By a customer with a grievance and a keyboard, somewhere in a 3-star review, wedged between a complaint about the packaging and a photo of their hallway. Review mining is how you get it out: a deliberate pass through reviews, threads and comments that turns real customer language into ad copy before anyone opens a script doc.
Conversion copywriters have mined reviews for landing pages for over a decade. Far fewer teams do it for UGC ads, which is odd, because a UGC ad has one job in its first 3 seconds: sound like a person, not a brand. Nothing sounds more like a person than a sentence a person actually typed.
The worked example: a cordless stick vacuum, sold DTC at around $280, up against household names with 10x the budget. Every mined phrase that follows is the kind of thing an hour of digging surfaces in that category, and by the end they become 4 usable hooks and a 30 second script skeleton.
Why review mining beats writing ad copy cold
The canonical proof is old and still unbeaten. Joanna Wiebe of Copyhackers mined Amazon book reviews for a rehab clinic and swapped the control headline "Your addiction ends here" for a line lifted from a reviewer: "If you think you need rehab, you do." Clicks on the main call to action rose over 400%. She did not write a better headline. She found one.
Copyhackers built its method around landing pages, where you have paragraphs to work with. UGC ads are a harsher environment. You get 3 seconds before the thumb moves, so the gap between "language a marketer wrote" and "language a customer typed" gets brutally exposed at the hook. A marketer writes "tired of vacuums that lose suction?" A customer writes "the battery dies before I finish the stairs." One of those stops a scroll. It is never the first one.
Our stance is not the consensus one: verbatim beats paraphrase, every time. The instinct to tidy the grammar, shorten the sentence or make the phrase "on brand" is the instinct that kills it. The slightly wonky rhythm of a real sentence is what makes it read as speech. Capture it exactly as typed, typos included, and only clean it up at the final scripting step if a creator genuinely cannot say it.
One legal note before the digging starts. The FTC's 2024 rule on consumer reviews and testimonials bans fake or misrepresented testimonials, with civil penalties currently at $51,744 per violation. Mining reviews for language patterns is research and entirely fine. Scripting a creator to voice a common complaint in their own words is fine too. Presenting an invented person's invented experience as a genuine customer testimonial is not. Keep that line bright and this whole method is clean.
Where to mine: 4 surfaces, ranked
Not all review surfaces give you the same raw material. For the vacuum brand, this is the order we would work them.
1. Amazon reviews (yours and your competitors')
Even if you sell purely DTC, your two biggest competitors are on Amazon, and their review sections are a free focus group. Go straight to the 3-star reviews. 5-star reviews gush, 1-star reviews rage, but 3-star reviews negotiate: "suction is genuinely great, but the battery dies before I finish the stairs." That sentence contains a proof line and an objection, which is a hook and a script beat in one. Since Amazon banned incentivised reviews, the middle of the star range is also where the least performative writing lives. For the vacuum, mine 60 reviews across the two nearest competitor models before touching your own.
2. Reddit threads
People go to Reddit specifically to escape marketing language. An Android Authority poll found almost 70% of respondents append "reddit" to their searches to get unfiltered answers, and category subreddits reward long, specific answers. A thread titled "which cordless stick is actually worth it?" is a pre-run debate: every objection your ad needs to answer has already been raised, argued and upvoted. Reddit is also where identity language lives. Nobody writes "as someone with two shedding dogs and a toddler" in an Amazon review. On Reddit they open with it.
3. TikTok comments
Pull up competitor ads and organic reviews in your category and read the comments, especially the sceptical ones. TikTok comments arrive pre-compressed to hook length. "ok but does it work on carpet or just hard floors" needs almost no editing to become an opening line. Comments under ads are doubly useful because they are reactions to advertising itself: they tell you which claims the audience already refuses to believe.
4. Your own post-purchase reviews and support tickets
The only surface where every voice is a confirmed buyer of your product. Post-purchase reviews give you the surprise bucket ("did not expect it to pull this much out of a carpet we vacuum weekly"). Support tickets give you friction nobody posts publicly, and one question mines better than any other if you run a post-purchase survey: "what nearly stopped you buying?" The answers are your objection list, ranked by the people who overcame them.
Capture phrases verbatim: the phrase capture sheet
The discipline that separates review mining from casually reading reviews is the capture sheet. One spreadsheet, four buckets, and a tally column. Every time a phrase repeats in substance, you do not capture it again. You add a tally mark to the existing row. That tally column is the whole point: it converts a pile of anecdotes into a ranked creative brief.
The four buckets:
- Complaint. What the category or product gets wrong, in the customer's words.
- Desired outcome. What they were actually trying to buy, which is rarely the product itself.
- Surprise. The thing they did not expect. This is where believable proof lives.
- Identity. How they describe themselves when explaining why the product matters to them.
Here is the vacuum brand's sheet after 60 competitor reviews, two Reddit threads and one afternoon of TikTok comments:
| Verbatim phrase (as typed) | Bucket | Tally |
|---|---|---|
| "the battery dies before I finish the stairs" | Complaint | 17 |
| "I just want to do the whole flat on one charge" | Desired outcome | 14 |
| "two shedding dogs and a toddler" | Identity | 13 |
| "something I can grab for crumbs without dragging out the big vacuum" | Desired outcome | 12 |
| "heavier than it looks, my wrist ached by the end" | Complaint | 11 |
| "emptying the bin is fiddly, dust goes everywhere" | Complaint | 9 |
| "did not expect it to pull this much out of a carpet we vacuum weekly" | Surprise | 8 |
| "my husband and I argue about whose turn it is now" | Surprise | 6 |
If a phrase shows up 17 times in 60 reviews, it is not an anecdote. It is the market dictating your hook.
How many reviews is enough? For a single product, 40 to 60 usually gets you to saturation: the point where new reviews stop producing new rows and just push the tallies up. When three consecutive reviews add nothing but tally marks, stop mining and start scripting. More is not better past that point. It is procrastination with a spreadsheet.
Turning an objection into a hook
The highest-tally complaint is almost always your hook, but you cannot just read it out. The move is to take the objection and do one of two things with it. If your product genuinely solves it, put the objection in a sceptic's mouth and answer it on camera. If your product shares the weakness, own it and reframe what actually matters. Never ignore it. The audience mining taught you about has read the same reviews you have.
This is where the buckets map onto our 7 hook angles framework almost one to one. Complaints feed the problem-solution and myth-busting angles. Surprise phrases feed curiosity and demonstration hooks. Identity phrases feed the callout angle, where the first line names exactly who the ad is for. Desired outcomes feed transformation hooks. If you mine properly, you never sit down to "brainstorm hooks" again. You sit down to assign phrases to angles, which is a much shorter meeting. For what separates a hook that holds from one that gets skipped, the patterns in our breakdown of UGC hooks that convert apply directly to mined lines too.
One more thing mining tells you that brainstorming never will: which awareness stage your audience is actually at. A category drowning in "which one is actually worth it" threads is solution aware, not problem aware, and your hooks should meet them there. The full mapping is in our guide to stages of awareness in UGC ads.
The payoff: 4 hooks and a 30 second skeleton
Here is what the vacuum brand's sheet becomes. Four hooks, each traceable to a row in the table:
- Hook 1 (objection flip, from the 17-tally complaint): "I almost didn't buy this vacuum because one review said the battery dies before you finish the stairs. So I timed it."
- Hook 2 (identity callout, from the 13-tally identity phrase): "If you've got two shedding dogs and carpet, ignore the spec sheet. Watch this instead."
- Hook 3 (surprise proof, from the 8-tally surprise): "We vacuum every week. This is what it still pulled out of the carpet."
- Hook 4 (desired outcome, from the 14-tally outcome): "Whole flat. One charge. No wrist ache. That's the entire pitch."
And the 30 second skeleton those hooks drop into:
- 0 to 3s, hook: one of the four lines above, delivered to camera, product visible.
- 3 to 8s, problem in their words: voice the top complaint verbatim ("every cordless I've owned died halfway up the stairs").
- 8 to 18s, demo as proof: show the claim, don't state it. Stairs, full flat, dog hair out of a clean-looking carpet.
- 18 to 24s, objection answered: hit the second-biggest complaint head on (weight, or the bin) before the viewer raises it.
- 24 to 30s, CTA with the outcome phrase: "whole flat on one charge" does more work here than any discount code.
Notice every beat is populated from the sheet. The script writes itself because the market wrote it first. The full beat-by-beat version of this structure, with timings and variant guidance, is in our UGC ad script structure template.
Key takeaway
Mine before you write. 40 to 60 reviews, captured verbatim into complaint, desired outcome, surprise and identity buckets with tally counts. The highest-tally complaint, flipped, is usually your best hook, and the rest of the sheet fills the script.
Where review mining sits in a real production process
At Spark, this is not a nice-to-have research step. It is the first stage of how every client engagement works, before concepts, before scripts, before a creator or avatar is briefed. The mined sheet becomes the angle map, the angle map becomes scripts, and the scripts carry customer phrasing all the way to the edit.
It shows up in the output quality in a way that surprised even us. Across our last 40 DTC creative batches, 94% of first cuts came back usable, and a large part of that is scripting from mined language: creators are performing lines that already sound like speech, so the delivery lands on the first take instead of the third. And when we have run a marketer-written hook against a mined hook on the same concept, the mined line is consistently the one that holds attention past the 3 second mark. The pattern is reliable enough that we now treat an unmined script as a draft, not a script. That research-first pipeline is a big part of why we hold an 8.7x blended ROAS across Spark client accounts.
If you want to go deeper on the scripting side, the frameworks in our resources library cover the rest of the chain from mined phrase to live ad.
FAQ: review mining for ad copy
What is review mining for ad copy?
Review mining is the practice of reading customer reviews, forum threads and comments in bulk, capturing the exact phrases customers use, and building ad copy from that language instead of writing it cold. For UGC ads, the highest-tally complaint or desired outcome usually becomes the hook, because it is already phrased the way the audience talks.
Can I quote a customer review word for word in an ad?
You can use the language pattern freely: script your creator to voice the same complaint or outcome in their own delivery. What you cannot do is present a fabricated or misrepresented testimonial as a genuine customer experience. The FTC's 2024 rule on consumer reviews and testimonials bans fake testimonials outright, with civil penalties currently $51,744 per violation. Mining phrasing for a script is research; inventing a reviewer is a violation.
How many reviews do you need to mine before scripting?
Around 40 to 60 reviews per product or competitor is usually enough. Capture verbatim phrases into four buckets (complaint, desired outcome, surprise, identity) and tally repeats. When new reviews stop producing new phrases and the tallies just climb, you have hit saturation and can stop.
Which reviews are most useful for UGC hooks?
3-star reviews are the most valuable because they hold a complaint and a compliment in the same breath, which is exactly the tension a hook needs. 1-star reviews give you the objections your script must answer, and 5-star reviews give you proof lines and surprise moments for the middle of the ad.