Campaign audit · 1,953 clips reviewed
The views were real. That was never the problem.
We pulled every submission from the clipping campaign on Can't Look Away and checked each clip against the record itself. Only 3% shows any sign of botting. Daily Spotify streams are running 2.2x higher now than while the campaign was live, which tells you where the problem actually sits.
Part one is what happened. Part two is the brief we would run for $2,000. We read 1,494 of their comments: four asked about the song, 147 asked about the footage. That ratio is the whole problem, and it decides every lane in the plan.
What actually happened
Every one of the 1,953 submissions, checked against the record, the brief and the streaming data.
01 · The record
It is a house record, and the brief was written for a different genre.
We measured the audio rather than reading the tags. The kick lands every 480 milliseconds with zero deviation across the kick, snare and hat bands, which is a machine locked four on the floor at 125.0 BPM.
BPM, four on the floor
Confirmed independently in three frequency bands. Full song runs 2:40.
Where the drop lands
The TikTok sound covers song 0:41 to 1:41, so the drop sits 4.1 seconds into the sound.
Minimum clip length
Three seconds before the drop and twelve after. Start at song 0:42, which reads 0:01 on the TikTok trimmer.
The master is flat at -11.5 dB with no dynamic build, so there is no audible drop moment for a clipper to cut against. Without a stated start time in the brief, they were guessing.
02 · The brief
The brief said everything allowed. That is the whole campaign in two words.
Read against what we publish for a clipping campaign, the gaps are not stylistic. There was no length rule, no start time, no format rule and no lane. Every failure further down this page traces back to this page of the brief.
| Rule | What their brief said | What we publish |
|---|---|---|
| Minimum length | Not mentioned | 15s floor, three seconds before the drop and twelve after |
| Start time | Not mentioned | One published second, given in song time and in TikTok trimmer time |
| Format | Slideshows allowed. Only soccer slideshows and 18+ anime slides banned | 9:16 video, hook in the first second, cut on the beat |
| Edit lane | "Everything allowed except the niches below" | One named lane, with a scene bank and pre made edits supplied |
| Platforms | TikTok only | Instagram primary, TikTok secondary |
| Rate | $130 per 1M views, which is a $0.13 CPM | A working CPM plus a weekly prize for the best clip |
| Per account cap | 50 submissions, unlimited per day | No campaign-only accounts, max 20% of budget to one creator |
| Creative direction | Comment the song name and like your own comment | Hook headline bank, on screen text bank, caption bank, each claimed so nobody duplicates |
| Review gate | Engagement threshold only | Format, length, sound, lane and bio link all checked before payout |
| Watermarks, safe zones | Not mentioned | No watermarks, text in safe zones, artist name spelled correctly |
Two numbers in the brief work against each other. Max earnings per post is $130 and max earnings per creator is $160. A clipper who lands one strong clip is capped almost immediately, so the best performers had no reason to submit a second one. The cap rewards volume from weak accounts and penalises the exact people you want.
The brief set a views goal of 12,307,000, which is simply the budget divided by the rate. The campaign delivered 17.0M, so on its own terms it succeeded. That is the clearest argument that views were never the right target: the goal was hit by 38% and the record still got nothing out of it.
03 · Compliance
Of 1,953 submissions, 18 met the full spec.
Each gate below is a rule we hold our own clippers to. This is how many clips survived each one, cumulatively.
Photo slideshows
Typically one or two still images with the song underneath. Not an edit in any sense.
Under 15 seconds
Of the clips that were genuinely videos. A further 30.4% ran past 35 seconds.
Correct sound
The one gate that held. Sound discipline was close to perfect, so attribution was never at risk.
04 · Real versus botted
Botting accounts for 2.8%. The other 97% is the real finding.
We built a control group from the same accounts' own non-campaign posts, which removes account quality as a variable, then tested every clip above 10,000 views against it.
| Bucket | Clips | Share of live clips | Views | Share of views |
|---|---|---|---|---|
| Already deleted or unreachable | 242 | 12.4% of submitted | n/a | n/a |
| Botted flagged | 48 | 2.8% | 1,446,440 | 6.7% |
| Real, but under 1,000 views | 915 | 53.5% | 400,165 | 1.9% |
| Real and measurable | 783 | 45.8% | 19,726,099 | 91.5% |
The 48 botted clips are 39 the platform flagged itself plus 9 our own signature test caught. A further 124 were rejected for falling under the engagement threshold. This campaign was not defrauded at scale.
Engagement matched the accounts' normal output
Campaign clips versus the same accounts' organic posts. Statistically indistinguishable, which is what proves the traffic was real.
But the clips underperformed their own creators
Median views per post, campaign clips versus what those same accounts get organically.
The content did 1.9x worse than what these creators post on their own. The audience was not the constraint. The creative was.
05 · Edit lanes
Half the campaign went into the two lanes that do not work on a house record.
We classified every clip by what it actually shows. Scene based edits, meaning film, television, Korean drama and meme footage under a hook headline, carried the campaign. Anime and sports consumed it.
| Lane | Clips | Share of clips | Share of views | Views per clip | Index |
|---|---|---|---|---|---|
| Scene edits works | 363 | 21.2% | 55.6% | 33,019 | 2.62x |
| Anime and sports does not | 890 | 52.0% | 24.5% | 5,936 | 0.47x |
| Everything else | 458 | 26.8% | 19.9% | 9,388 | 0.74x |
A scene edit was worth 5.6x an anime or sports clip. Their single biggest clip, at 3.99 million views, was a drama scene edit. Across our own finished house campaigns, film and television edits carry roughly 75% of all views, so this campaign independently reproduced a pattern we already price against.
The same clips, allocated correctly
Holding clip count and spend fixed, and moving the anime and sports volume into the scene lane.
06 · What works on house
We have 42.5M views of house clipping data. It says the same thing.
Three finished house campaigns of ours, Fisher, aka Vinnie and Chris Lake & Tiga, with 696 machine-labelled posts. Film and television edits take 71.8% of views, and two lanes that perform well on other genres fall apart on house.
| Edit lane, our house campaigns | Posts | Views | Share | Views per post | Index |
|---|---|---|---|---|---|
| Film / TV edit core | 449 | 16,805,671 | 71.8% | 37,429 | 1.11x |
| Memes / humour core | 156 | 4,929,470 | 21.1% | 31,599 | 0.94x |
| Football / soccer | 26 | 771,970 | 3.3% | 29,691 | 0.88x |
| Cars / automotive | 9 | 335,390 | 1.4% | 37,266 | 1.11x |
| Gaming | 12 | 204,950 | 0.9% | 17,079 | 0.51x |
| Superhero / comic dies | 30 | 112,794 | 0.5% | 3,760 | 0.11x |
| Emotional / storytelling dies | 8 | 18,487 | 0.1% | 2,311 | 0.07x |
The genre effect is not small. A superhero edit averages 51,380 views on our non-house campaigns and 3,760 on house. An emotional storytelling edit averages 113,648 off house and 2,311 on it. Cars run the other way and do 5.7x better on house than anywhere else. Lane selection is genre specific, and guessing costs an order of magnitude.
Views are not the same question as conversion
Saves are the step between watching a clip and going to find the song. Here is what each lane in their own campaign actually produced, ranked by saves rather than views.
| Lane, their campaign | Clips | Share of clips | Saves | Share of saves | Saves per clip | Photo |
|---|---|---|---|---|---|---|
| K-drama scene edits | 119 | 7.0% | 75,869 | 51.8% | 638 | 0% |
| Anime | 377 | 22.0% | 31,253 | 21.3% | 83 | 93% |
| Film / TV | 53 | 3.1% | 6,700 | 4.6% | 126 | 23% |
| Memes | 82 | 4.8% | 5,714 | 3.9% | 70 | 71% |
| Sports worst | 513 | 30.0% | 5,531 | 3.8% | 11 | 85% |
| Superhero | 109 | 6.4% | 1,174 | 0.8% | 11 | 18% |
K-drama versus sports
On saves per clip. Seven percent of the campaign produced more than half of all the intent it generated, and every one of those clips was a real video.
Real video versus slideshow
159 saves per video against 41 per slideshow. The 62% of the campaign that was slideshows produced 29.7% of the saves.
07 · The bias check
We checked our own playbook against records we did not brief.
Our house numbers have a flaw: they measure what we commissioned. If we mostly brief film and television edits, film and television edits will take most of our views by construction. So we went and found house records that broke on TikTok without us, and looked at what actually carried them.
How the comparison set was built
We took TikTok's top 2,000 tracks, kept the 809 that entered the chart in 2025 or 2026, screened them for dance records, then confirmed the genre by measuring the audio, the same tempo and four-on-the-floor test used on this record. Seven came back as genuine house. Two of those, a Taylor Swift record and a Dua Lipa record, were dropped: they are house tempo carried by megastar fandoms, which is not a comparison an independent artist can use. That leaves five dance-act house records and their 930 top-performing videos, 1.23 billion views.
59.5% of the views on those records came from content a clipping page cannot make. Travel, beauty, food, nightlife, dance, outfit and lifestyle clips, filmed by real people with real lives. Only 21.8% came from the kind of footage a faceless edit page can produce at all.
| Lane, organic house hits | Videos | % views | Views per video | Index | Can a clipper make it? |
|---|---|---|---|---|---|
| Dance performance | 23 | 6.3% | 3,358,545 | 2.54x | No, needs a real person |
| Beauty | 39 | 7.4% | 2,328,369 | 1.76x | No, needs a real person |
| Artist / music content | 53 | 8.7% | 2,015,107 | 1.52x | yes |
| Relationship / couple | 15 | 2.0% | 1,619,267 | 1.22x | yes |
| Travel | 105 | 13.4% | 1,570,723 | 1.19x | No, needs a real person |
| Sports | 27 | 2.8% | 1,266,956 | 0.96x | yes |
| Film / TV | 16 | 1.6% | 1,264,885 | 0.96x | yes |
| Memes | 41 | 3.1% | 941,973 | 0.71x | yes |
| Gaming | 31 | 1.3% | 532,245 | 0.40x | yes, but don't |
| Anime | 20 | 0.7% | 430,254 | 0.33x | yes, but don't |
What the check actually changed
Our ranking holds
Where both datasets cover a lane they agree. Film and TV beats memes, memes beat sports, sports beat gaming, and anime is last in both. Our order was not wrong.
But the dominance was an illusion
Film and TV took 71.8% of our house views because it was 64% of our posts. Per post it indexes only 1.10x, and 0.96x organically. It is a safe, average lane we scaled, not a standout.
The best lanes are untested
Dance, beauty, relationship and artist-led content all beat film and TV on real house records, and we have never briefed any of them at volume.
For this record the practical read is: keep scene edits as the clipping floor, since they are the best thing a paid page can reliably make, drop anime and gaming outright, which both datasets agree are the worst lanes on house, test relationship and artist-led angles, and accept that the 60% of house reach that comes from real people is a creator seeding job, not a clipping job.
Two limits, stated plainly. These are the top videos on each sound, so this describes what wins rather than a full census. And a sound's lane mix is partly downstream of it already trending, so this is association, not proof of what caused the record to break. Follower counts are not returned by this endpoint, so none of it is controlled for account size.
08 · Streaming reality
Streams went up after the clipping stopped.
Daily Spotify streams from Soundcharts, plotted against clips posted per day. If the clipping were driving streams, the line would follow the bars. It does the opposite.
Streams per day, during
Average daily Spotify streams across the days the campaign was live and tracked.
Streams per day, after
Average across the 21 days since the last clip was posted. The record is growing, just not from this.
TikTok views per stream
17.0M campaign views against 574,316 lifetime streams. Even crediting the campaign with every stream the song has ever had, that is a 3.4% conversion ceiling.
The song has 32 Spotify playlist placements and not one is editorial. Thirty are listener playlists, one is a radio and one is algorithmic. The growth since the campaign ended is coming from algorithmic pickup, listener adds and the paid ads already running, which is consistent with clipping having contributed close to nothing.
One limit worth stating plainly: Soundcharts begins tracking this song on 18 August, seven days into the campaign, so the first and largest posting spike on 12 August sits before the streaming data starts. The comparison above covers the campaign's declining tail against the period after it. It cannot rule out an unmeasured lift in that first week, though the flat trajectory through 18 to 27 August argues against one.
09 · Root cause
At nine cents per thousand views, you get the roster you pay for.
The campaign paid $0.13 per 1,000 views against a $130 per clip cap, $1,575.44 in total, which works out to an all in CPM of $0.093. Ours runs $1.50 to $2.00. Seventy nine percent of submissions earned nothing at all.
Who actually showed up
- 584 followers is the median participating account
- 52% of accounts sit under 1,000 followers, and 76% under 10,000
- 60% have a median post that never clears 1,000 views
- 57% use a different trending sound on every recent post, which is clip farm behaviour rather than a page with an audience
What that produced
- 754 views was the median submission
- 53.5% of live clips never reached 1,000 views, contributing 1.9% of the total
- 12.4% have already been deleted by the poster
- Twelve clips produced half of everything the campaign delivered
Difficulty attracting quality clippers is not a platform problem. It is the direct and predictable consequence of the rate. Quality clippers do not work for nine cents.
What we would do with $2,000
Same song, same platform problem, different brief. Everything below is chosen on saves per clip, because saves are the step before a stream, and restricted to footage a clipper can actually source.
10 · The budget
$2,000 buys a 1,000,000 view floor, and the floor is the guarantee.
We price clipping at a $2.00 CPM, so $2,000 is a guaranteed million views. Views past the floor cost nothing extra. Typically a campaign lands 3x to 30x its goal and 2x to 5x the estimated clip count, which puts the realistic range at 3M to 30M views from roughly 500 to 1,250 clips.
Guaranteed floor
Contractual. If we do not reach it we keep working. Their $1,575 bought a goal of 12.3M on paper and a median clip of 754 views.
Typical delivery
What our clipping campaigns usually return against the goal. Stated as typical, never guaranteed.
Saves, from the gate alone
Their slideshows saved at 41 per clip, real videos at 159. Rejecting slideshows on the same 1,711 clips would have produced 272,000 saves instead of 146,402, before any lane change.
We report on saves and streams, not views. Views were the metric that let their last campaign beat its goal by 38% while the record got nothing.
11 · The lyric
The song is about being unable to look away from someone.
There are no published lyrics for this record, so we transcribed the official audio. The register is adoring, sensual and nocturnal. It is not heartbreak, it is not motivation, and it is emphatically not action.
What the record says
The moon can't do what you do
Shine as bright as you do
The stars can't do what you do
Burn as hot as you do
When you move like that, I can't look away
When you move like that, I just wanna stay all night
All night, dancing in your light
What plays inside every compliant clip
A 15 second clip starting at song 0:42 carries these exact words, in this order:
"When you move like that,
I can't look away
When you move like that,
I just wanna stay all night"
So the brief is not a genre. It is a moment: someone who cannot take their eyes off someone else.
12 · The proof
Their audience asked about the footage, never the song.
Saves turned out to be the wrong target, so we went and read the comments. We pulled 1,494 organic comments from across their campaign, excluded the clippers' own planted ones, and classified what each viewer actually wanted.
Comments about the song
Out of 1,494. And all four were duds: "nice song", one about Zara Larsson, one about a different track entirely, and "Dude stop with the music bro". The campaign produced zero genuine song discovery.
Comments about the footage
"drama name please". "what's the anime". "Brewing Love". "I need to watch this". A 37 to 1 ratio against the record. The borrowed footage took the curiosity the song needed.
| Lane, their campaign | Comments | Asked about the song | Asked about the footage |
|---|---|---|---|
| K-drama worst | 220 | 0.0% | 19.1% |
| Celebrity | 96 | 1.0% | 14.6% |
| Superhero | 136 | 0.0% | 11.0% |
| Anime | 251 | 0.4% | 10.4% |
| Sports | 193 | 0.5% | 10.4% |
| Film / TV | 69 | 0.0% | 7.2% |
| Aesthetic b-roll | 64 | 0.0% | 3.1% |
This is why K-drama comes out of the plan despite topping every saves table. Its saves were people saving the drama. On their best clip the organic comments read "drama name please", "Brewing Love", "looks cringe but I want to watch it". The only mention of the song was the clipper's own planted comment. And that lane's average was an illusion anyway: 84.7% of its saves came from one clip, and the median K-drama clip got 36.
So we checked what does make people ask
The same test on 4,907 organic comments from the five house records that broke without us. Song curiosity there runs at 3.7%, twelve times their campaign, and it is wildly concentrated.
Share of organic comments asking about the song. The pattern is not about genre at all: people ask about the song when the song is the subject of the video, or the reason the video works. When the footage has its own story, they ask about the footage.
One limit: a viewer can tap the sound instead of commenting, so this undercounts song curiosity in absolute terms. It does not affect the comparison between lanes, which is what the plan is built on.
13 · The lanes
Three lanes a faceless page can actually run.
Every reference below was vetted account by account. We pulled each one's recent posts and kept only faceless pages that assemble borrowed or generic footage. Anyone who appears in their own content was thrown out, because a clipper cannot replicate a person.
The two tests every lane has to pass
Does it make people ask about the song, and can an anonymous page make it?
The first test removed K-drama, anime, film and sport. The second removed dance, which is the highest-reach lane on house at 2.54x but is real people dancing. We vetted 68 accounts behind the winning lanes by pulling each one's recent posts: 26 were faceless pages, 32 were creators filming themselves, and 10 were too ambiguous to call. Only the pages are below.
| Lane | Song intent | Footage intent | Faceless pages found | Budget |
|---|---|---|---|---|
| 1. Lyric & audio edits | 27.7% | 2.3% | 6 of 18 | 40% |
| 2. Meme & edit pages | 7.6% | 3.8% | 9 of 22 | 30% |
| 3. House curation & moodboard | 7.3% | 1.5% | 6 of 26 | 30% |
| Dance creators only | 7.4% | 3.0% | 5 of 20 | 0% |
| K-drama, anime, film, sport cut | 0 to 1.8% | 4.6 to 19.1% | n/a | 0% |
Lyric and audio edits 40% of budget
The highest song intent of any lane at 27.7%, and when we vetted the accounts behind it, six of the nine faceless pages were running exactly this format: the words on screen, a slowed or 8D version, a translation. The song is not the soundtrack, it is the entire post, which is why the curiosity lands on the record.
The brief: the hook burned across the frame over generic or stock footage, timed so "I can't look away" lands on the drop at song 0:45.1. Variants that are already working on house: slowed and reverb edits, lyric translations for Spanish and Portuguese audiences, and a plain lyric video over moving wallpaper. No franchise footage, nothing anyone could ask the name of.
Meme and edit pages 30% of budget
7.6% song intent against 3.8% footage intent, because the joke does not work without the audio. This is the largest supply of genuinely faceless operators we found: nine vetted pages, running game clips, animation, reaction footage and football memes.
The brief: the drop lands on the punchline, the cut or the reveal. Generic, unbranded or animated footage only. If a viewer could plausibly ask "what show is this", it is the wrong clip. 10 to 25 seconds.
House curation and moodboard 30% of budget
The lowest footage hijack of anything we measured at 1.5%. Nobody asks what film a sunset is from, so the only thing left to ask about is the record. These pages exist purely to post tracks and moods, and the two smallest ones here carry the best save rates on the page at 2.65% and 1.49%.
The brief: unbranded, unidentifiable, nocturnal. Neon, headlights, water, a room at night, a crowd out of focus. Stock and Pinterest footage is fine and is what these pages already use. The lyric writes the on-screen text for free: "the moon can't do what you do", "dancing in your light".
Dance, and why it is not in the plan 0% of budget
Dance is the highest-reach lane on house, at 2.54x average views per video, and it does generate song curiosity at 7.4%. We still cut it. Of the 20 dance accounts we vetted, 13 were people dancing and the five faceless ones were weak reposters. You cannot brief a clipper to dance.
What to do with it instead: this is the clearest case for pairing the clipping budget with creator seeding. The reach is real and it is the most on-theme content the record could have, given the chorus is "when you move like that". It is simply a different line item.
14 · What we hand the clippers
Nobody guesses. Everything is supplied.
The last campaign asked for a song name in the comments and left every creative decision to a stranger being paid nine cents per thousand views. This is the opposite.
One start time
Song 0:42, which reads 0:01 on the TikTok trimmer, 15 second minimum so the drop at 0:45.1 lands three seconds in and plays twelve after. Printed on the brief in both clocks.
A scene bank
Named titles per lane with approximate timestamps and season or episode numbers, so clippers are not hunting for footage or guessing at the mood.
Pre made edits
Ready-cut edits in a shared folder, each claimed by one clipper so two people never post the same thing.
Hook and caption banks
On-screen text lines and captions written for this record, claimable, so the hook is never left to whoever is posting.
Instagram as well
Across our tracked campaigns Instagram carried 97.1% of clipping views. Running TikTok only, as they did, forfeits the surface that produces the reach.
Review before payout
Format, length, sound, lane and bio link all checked before a cent moves. Their only gate was an engagement threshold applied after the fact.
And the honest caveat, repeated from the audit: about 60% of what makes a house record travel is real people filming their own lives, which no clipping page can produce. Clipping is the right lever for the other 40%. If the goal is the whole record, this sits alongside creator seeding rather than replacing it.