Shooting in the Dark

Stirling is a small town in central Scotland with a castle where Mary Queen of Scots was crowned, a statue of William Wallace on a hill, and very little else a weekend traveler is expected to do after sundown. I was there on a Saturday because I wanted a quiet Saturday. You can drink in any of a dozen pubs on the old hill. You can watch the proverbial paint dry. You can walk up Broad Street and see what is on at the Tolbooth.

The Tolbooth has been at the top of Broad Street since 1473. It was a courthouse, a jail, the seat of the Scottish parliament during a civil war in 1571. In 1820 two leaders of the Radical War, Baird and Hardie, were executed on the street outside. The Victorians eventually repurposed the courtroom as a concert hall. After 2000 a Scottish architect named Richard Murphy converted the whole building into an arts venue. The main auditorium holds 166 seated and 225 standing. When I walked past the poster outside, I did not know any of that.

What I knew, because the poster said it, was that a musician called Brìghde Chaimbeul was playing that night. I did not know how to pronounce her name. I did not know what she played. I typed her name into my phone.

What came back was not a Spotify top-ten, not a Pitchfork review, not a TikTok clip. It was a sentence. She plays the Scottish smallpipes — a quieter, mellower cousin of the Highland bagpipes — and she is from the Isle of Skye. That was it. No music, no reviews, no context. Just enough.

I went to the Tolbooth.

The sentence that closed the sale

That sentence — “Scottish smallpipes, quieter cousin of the Highland bagpipes, Isle of Skye” — is not a marketing strategy. It is a sentence. Somebody wrote it once, somewhere, at some point in the last decade. Search engines have been rediscovering it ever since. A weekend traveler in Stirling made a choice on a Saturday because that sentence existed and because a physical piece of cardboard outside a venue cued him to go looking for it. Three things closed the sale: lack of choice, preference for a particular instrument, a daredevil mood about the unknown.

From Brìghde Chaimbeul’s perspective, I am dumb luck. From mine, she is. From the perspective of anyone trying to engineer this kind of encounter at scale — a label, a manager, a family member who carries the promotional hat so the musician does not have to — the event is something more useful than luck. It is evidence. The two things that closed the sale were the specific physical placement of the poster and the specific informational density of what I found one tap away. Not her follower count. Not her streaming numbers. Not whether she had posted that week on TikTok.

Neither of these things is what most independent musicians are actually doing with their time online.

Most are copy-pasting. Same release announcement, same photo, same three hashtags, three platforms at a time, sometimes more. The assumption is that this is how you engineer the encounter in Stirling. If your content is everywhere, surely one of the everywheres will catch. If the weekend traveler sees you three times, surely one of them will stick.

This is wrong in a specific, measurable way, and it is wrong for reasons almost nobody in music promotion describes correctly.

The folklore that isn’t true

There is a piece of folklore about social media that independent musicians repeat to each other with the confidence of people who have read it in three different places. The folklore says that TikTok checks Instagram a few hours after you post. It says Instagram checks TikTok. It says Bluesky’s algorithm penalizes posts it detects on X. It says that somewhere, a machine is watching all of it, keeping score, and silently punishing anyone caught distributing their own content twice.

There is no primary source for any of this. Not in any patent filing. Not in any engineer’s conference talk. Not in the open-sourced Twitter algorithm, which Elon Musk published on GitHub in March 2023 specifically so people could inspect exactly this kind of mechanism. What is in the Twitter codebase is a within-platform duplicate check at tweet creation time and a copypasta downranking policy; neither contacts another platform’s servers. No other major platform has been shown to. The surveillance network that was supposed to be watching your cross-posts does not exist.

This is worse news for copy-paste, not better. The three mechanisms that actually do penalize identical cross-posts each operate without any platform having to look at any other platform. They are faster than the imaginary one, cheaper, and harder to avoid. A musician only needs to lose on one of them for the post to fail. Most lose on all three.

The three horsemen of oblivion

The first horseman is the file you just uploaded. When you film inside the TikTok app and save the result to your camera roll — the way most independent musicians do, because that is where the filters and the auto-captions and the effects are — the MP4 in your gallery has @yourusername burned into the bottom-left corner. TikTok’s font. TikTok’s opacity. Upload that file to Reels, and Adam Mosseri himself has gone on record about what happens. In October 2024, on his own Instagram Reel — which is faintly absurd when you think about it, the head of Reels using Reels to explain how Reels penalizes Reels — he said: your own logo is fine; logos from other apps are not. The original April 2024 Instagram Creators post was less diplomatic, calling visibly recycled content “less discoverable” on the Reels tab. TikTok’s Creator Academy states the mirror rule: repurposed content with someone else’s watermark is typically ineligible for the For You feed. YouTube’s Shorts Fund, back when it existed, was explicit that TikTok-watermarked uploads did not qualify, and the rule survived the Fund’s replacement.

None of these platforms had to contact any other platform to enforce any of this. The evidence was in the pixels of the file you uploaded. The gift you handed the algorithm on the way in.

The second horseman is within-platform matching. In 2019, Meta open-sourced two fingerprinting algorithms through a project called ThreatExchange: PDQ for images, TMK+PDQF for video. Subsequent benchmarks — one of them containing a direct statement from Meta itself — have clarified that these are reference implementations rather than production matchers. The production matchers are proprietary. They run, stitched into a system called Rights Manager, against everything uploaded to Facebook and Instagram. The consequences are documented in specific language on specific dates, and they have hardened over two years. Instagram, April 2024: when two or more identical pieces of content appear on Instagram, only the original is recommended, and the copy is replaced in the feed. An account that reposts others’ content 10 or more times in a rolling 30 days is removed from recommendation surfaces entirely. Facebook, July 2025: repeat offenders lose monetization access and receive reduced distribution on everything they share, not just the copies. Meta, March 2026: such accounts can be marked “non-recommendable” and demonetized.

SoundCloud does the audio version of this. Artists signed to distributors have already discovered that their second upload of their own track, from a second account, is silently blocked by Content ID. Medium does not bother with the subtlety. Upload the same article to two Medium accounts and Medium suspends one of them outright. Its no-duplicate-content rule runs three short paragraphs and reads like a warning label on a blender.

All of this fires on identical duplicates, uploaded to the same platform, without any outside visibility required. It fires on your own copies of your own posts.

The third horseman is the one most creators underestimate, and it is the cheapest of the three to operate because it requires no infrastructure at all. The audience.

A YouTube Shorts viewer who spent an hour earlier in the evening on TikTok can tell a reuploaded TikTok from a native Short even when the watermark is scrubbed. The pacing. The on-screen caption in TikTok’s particular text-overlay style. The way the audio sits slightly hot under the first beat, because TikTok mixes audio a specific way. The comment that follows is usually some variant of “go back to TikTok,” which costs nothing to type and roughly a thousand times that in lost watch time. Watch time is the signal YouTube actually reads.

Bluesky is the extreme version. The user base is small and literate, and the culture is allergic to anything that reads as X cadence — declarative hot takes, dunk-reply architecture, main-character-of-the-day framing. Cross-post your X thread verbatim to Bluesky and the replies come within the hour. They are not interested in your music.

The audience does not need an algorithm to enforce any of this. The audience is the algorithm, and the algorithm reads the numbers the audience produces.

The arXiv paper nobody in music has read

Small edits matter at a scale you cannot dismiss as noise. The cleanest evidence is a 2025 arXiv paper that collected 59.6 million Reddit posts linking to YouTube videos and cross-referenced each Reddit title against the YouTube title it was meant to represent. Twenty-one percent of the Reddit titles were not copies. Somebody — the person sharing the video on Reddit — had rewritten the YouTube headline. Sometimes by one word. Sometimes by several. Rarely by all of them.

The authors then trained a BERT classifier on the edits. After controlling for the underlying video’s popularity, the subreddit, and the time of day, the classifier reached 74% accuracy predicting which of two titles would be more popular. It did this without knowing anything about the video itself. Just by looking at the words.

What this tells you is not that adaptation produces a magic uplift. It tells you that the identical re-post is leaving a measurable fraction of its engagement on the table, and that the fraction is recoverable with effort measured in seconds.

Hootsuite ran the cruder version of this test four years earlier, in 2021. Five near-identical videos, two zero-follower accounts, TikTok and Reels. The Reels videos peaked at 2 views apiece. The TikTok videos averaged 450. The sample is small enough that you cannot take the specific numbers seriously, but you can take the direction seriously, and the direction matches the benchmark data. Socialinsider’s 2026 aggregate of 70 million posts shows TikTok engagement at 3.70% and Facebook at 0.15%. A 25-fold spread between what the same video format can do on two properties owned by the same company. No ranking system in history has ever managed to make identical content perform identically across a gradient that steep.

Viral faster. Converts worse.

MIDiA Research published a study in September 2025 called “All Eyes, No Ears,” drawn from a survey of 10,000 global music consumers. The findings are brutal on the TikTok-equals-streams assumption. Forty-eight percent of listeners who had heard music on social media in the previous month had not subsequently streamed any of it. YouTube outperformed TikTok as a discovery source by 15 percentage points. Among users whose favorite discovery platform was a digital streaming service, 74% streamed more after finding something they liked. Among TikTok-primary discoverers, only 48%. The people MIDiA describes are not failed listeners. They are a listening audience that treats social-media music as ambient wallpaper. They watched the video. They did not save the song.

Duetti’s 2025 Music Economics Report is more granular and more direct. Built on a dataset of more than 6 million tracks, it defines TikTok virality with unusual precision: a doubling of video-creation growth in a month, with at least 250,000 video creations. Fewer than 1% of tracks cross the threshold. Of those that do, roughly 15% produce a durable 30%-or-better lift in Spotify streams at four months. YouTube-first artists develop durable catalogs at a 16% higher rate than artists who begin elsewhere. The TikTok-to-Spotify funnel has narrowed year over year since 2020, even as the time-to-virality on TikTok has dropped from roughly 340 days to 48.

Viral faster. Converts worse.

Those numbers describe the shallow end of the conversion problem. The deeper end is what a Spotify stream actually pays. It pays a fraction of a cent. A 2,000-stream bump from a TikTok moment is a rounding error against a single $17 cassette sold from the merch table, which is itself a rounding error against a $40 vinyl pressing sold direct-to-fan on Bandcamp, and all of it is an infinite rounding error against a ticket to the gig. The musician optimizing for TikTok reach is usually, whether they know it or not, optimizing for the least profitable conversion path on the chart. The musician posting the identical clip to TikTok, Reels, and Shorts is optimizing for three of the least profitable paths simultaneously, and underperforming on each.

The copy-paste question is not whether it hurts reach. It hurts reach less than the folklore says. The copy-paste question is what it collapses: three possible destinations into one generic call to action, when three different-shaped posts could send three different audiences toward three different outcomes. Some of which pay you.

Upload originals, not re-downloads

The single most deterministic cross-platform penalty — the watermark rule — is almost always a re-download problem, and the workflow that produces it is the one most independent musicians default into without thinking about it.

You film inside the TikTok app because that is where the filters and auto-captions live. TikTok saves the result to your camera roll with its watermark baked in. You open Instagram, hit New Reel, select that file. Reels demotes it. You open YouTube, hit Create Short, select the same file. Shorts demotes it. Three platforms, three demotions, one unopened export of the master footage that would have solved all of them.

The fix is not to edit the re-downloaded file differently. The watermark is not a layer you can remove. It is welded into the pixels, and any edit that genuinely obscures it will also visibly degrade the image. If you filmed inside TikTok’s camera, you cannot produce a clean Reels cut from what TikTok gave you. You have to reshoot, or re-edit from the original.

Film in a neutral editor first. Your phone’s stock camera. CapCut. Premiere. Anything that does not burn a logo into the corner. Edit the master once. Export three versions, each tuned to its platform — hook timing, aspect ratio, on-screen captions, thumbnail. This is not glamorous work. It is the one workflow change that eliminates the most documented cross-platform reach penalty in the industry, and a surprising number of musicians have never heard it.

The Substack trap

The SEO side of the copy-paste problem is rarely discussed in music circles and deserves more attention than it gets.

Google’s position on duplicate content has been consistent since 2008, when a Google employee named Susan Moskwa published a Search Central post titled “Demystifying the ‘duplicate content penalty.’” Her thesis was that there is no such penalty. The engine clusters duplicate pages together and picks one URL to represent the cluster in search results. The mechanism a writer has for saying “please pick this one” is the rel=canonical tag. John Mueller, Google’s long-serving Search Advocate, has been saying for years that the tag is a hint, not a directive. Google uses it when it agrees. Overrides it when it thinks the author set it wrong. Picks algorithmically when nobody set it at all. Click behavior, via a mechanism called Navboost that became public only through the 2023 DOJ antitrust disclosures, feeds back into the choice.

The typical release-announcement workflow for an independent musician distributes identical copy across four destinations: the artist’s own site, the Bandcamp description field, a Medium post, and a Substack post. The artist expects the main site to win.

It does not always win.

Medium handles canonical tags correctly if you use its Import Story tool, which automatically points the canonical back to the original source. Medium also lets you set it manually, in the Advanced tab. Substack does neither. There is no canonical URL field in Substack’s publishing UI. There is no manual override. Substack’s own SEO guide lists the editable metadata fields — title, description, slug, preview image — and canonical URL is conspicuously absent, on a list where it would be standard in any publishing tool of comparable weight. Substack self-canonicalizes every post it hosts. Because Substack’s domain authority outranks most independent artist sites, the Substack copy of your release announcement can, and routinely does, outrank the one on your own domain. Your own fans, typing your own band name into Google, land on Substack first. You did this to yourself.

The only fix, given Substack’s tooling, is to not repost in full. Post an excerpt. Link back. Treat Substack as a teaser channel until Substack ships a canonical field. There is no technical reason they haven’t. They just haven’t.

There is a reason this is not your job

Every rule above is current as of early 2026 and will shift. Meta’s unoriginal-content policy dates only to April 2024 for Instagram, July 2025 for Facebook, and March 2026 for the non-recommendable designation. The next restatement will add something else. TikTok’s originality policy is a 2024 product whose enforcement thresholds are still being tuned. Navboost became public through legal filings, and its role in syndicated-content ranking is being mapped by the SEO community in something that looks a lot like archaeology.

Any rule of thumb you read on an advice blog dated earlier than last quarter is, at best, a starting hypothesis.

The honest posture is to test. Pick a release. Post identical copy on one channel, adapted copy on another. Measure reach. Measure the downstream numbers that actually matter — click-throughs to the album page, streams, unique visitors, merch sold, emails captured, direct messages received. The platforms do not tell you what they are rewarding. Your conversion numbers do. Do this once a quarter, including on principles you think you have nailed, and you will catch the platform shifts before the advice blogs catch up.

This is also why promotional work is usually not the artist’s job, and why the artists who do it well have almost always handed it to somebody else.

Some artists hand it to a label. Some to a manager. Some to a parent — which did not work out for Britney Spears, whose father held her career inside a conservatorship from 2008 to 2021, and which has worked beautifully for Brìghde Chaimbeul, whose family on the Isle of Skye is the kind of thick musical ecosystem where a harpist sister at Berklee, a Gaelic-writing father, and a sculptor mother all contribute some part of what an artist needs to remain an artist while somebody else handles the merchandise, the calendar, and the sentences online. Pick your poison. Or wear the hat proudly and stop pretending it is beneath you. Do not wear it badly and pretend the music is what is keeping people away.

A musician whose album has not returned its investment six months after release did not fail because the music was bad. The music might well be extraordinary. The musician failed because the discovery strategy was either non-existent or carved in stone. For most of the last ten years, the default carved-in-stone strategy has been carpet-bombing. Same post, every platform. Playlist pitching by the yard. Three variants of the same Short. Monkeys writing Shakespeare, in quantity, in the hope that one of the monkeys will get lucky.

The monkey sometimes does get lucky. The math has always been against the monkey. It is not Starcraft. The zerg-rush journey does not get you anywhere. If your album is not returning its investment in six months, the monkey was the strategy.

The poster at the top of Broad Street

I went to Brìghde Chaimbeul’s gig. The auditorium was full. She played the smallpipes through a bank of effects pedals, which is not what anyone who has ever heard “Scotland the Brave” would guess pipes could do. There is a stillness to the drones. She played tunes that were four centuries old and tunes that ran the pipes through electronic processing and tunes that blurred the distinction. I bought the album at the merch table after the show, directly from her, signed — the kind of thing you can only do at a gig like this, from someone at the beginning of their career.

The strategy that got me to that gig was not her strategy. It was the venue’s strategy — a poster at the top of Broad Street — and it worked because when I searched her name that afternoon I found one specific, dense sentence that told me exactly what I needed to know in order to take the risk. Somebody wrote that sentence. Somebody made sure it was indexed. Somebody made sure that when a stranger in Stirling typed her name into a phone on a Saturday before dinner, the stranger got the smallpipes, and not twelve identical announcements of a new album across twelve identical feeds.

That is not luck. Luck was that I was in Stirling, bored, next to a poster. Everything after the poster was somebody doing their job.

A QR code on the poster pointing to a landing page — one URL, all the streaming links, a short paragraph about the smallpipes, a couple of samples — would have made it faster. It would not have changed the outcome, because the outcome was secured by the fact that what I found when I searched was genuine and specific. The minimum viable version of Brìghde Chaimbeul’s online presence, for the purposes of closing a stranger in a pub in Stirling on a Saturday, was a sentence.

Most musicians do not have the sentence. They have the three copies.