Better Than You Think
"If you're an artist, remember that in 2026 there's not a single part of the system that isn't faked. You're probably doing better than you think."
Response
6 min read
#music industry #spotify #ai
James Blake put a note on his Instagram story in June — reposted by Pigeons & Planes, which is where most people saw it. It is a list, and each line takes away one thing an artist might have used to work out whether anything is going well.
Reviews, because the music press stopped making money. Comment sections, because they fill with fake fan accounts manufacturing a bandwagon. YouTube counts, because they get bought. Streaming numbers, because of bot farms driving discovery. And then the one that would have sounded absurd five years ago: you cannot even be sure a song was made by human beings.
If you're an artist, remember that in 2026 there's not a single part of the system that isn't faked. You're probably doing better than you think.
Instagram is blocked until you allow external media.
Privacy detailsA post like this travels because it feels true. That is exactly the reason to go through it slowly rather than nod at it, so I have taken the five claims one at a time.
Going down the list
Streaming numbers. This one is not a suspicion, it is documented. In September 2024 the US Department of Justice brought its first criminal case for AI-driven streaming fraud, against a North Carolina musician who uploaded hundreds of thousands of AI-generated tracks and streamed them with roughly ten thousand bot accounts; he pleaded guilty and agreed to forfeit more than eight million dollars. Deezer, which now publishes numbers on this, put overall streaming fraud at eight per cent of all streams in 2025. Blake is right, and the scale is not marginal.
Whether a human made it. Also documented, and moving fast. Deezer reported in January 2026 that AI-generated tracks made up around 39 per cent of daily uploads. By April it was 44 per cent, at nearly 75,000 tracks a day. By July it had passed half. On the fraud side, the company found that up to 85 per cent of streams on those AI tracks were themselves fraudulent, and demonetised them. Spotify removed more than 75 million spam tracks in the year to September 2025. The catalogue genuinely is full of things nobody made.
Bought YouTube numbers and bot farms. Artificial streaming as a paid service is real and easy to buy. The part of Blake's phrasing I would not repeat as established is labels pay for bot farms — the prosecuted cases so far involve independent operators and third-party promotion services rather than majors, and I have not seen a proven case against a major label. What is fair to say is that the platforms clearly consider label-delivered content part of the problem: since April 2024 Spotify has charged labels and distributors a per-track penalty where flagrant artificial streaming is found on music they delivered. You do not build that mechanism for a problem confined to strangers.
Reviews. The weakest line as written, and the strongest underneath. Blogs and magazines really did lose their business model, freelance rates really have collapsed, and the money that remains around music writing really does come disproportionately from labels, PR and brands — sponsored features, paid placements, publications owned by companies with acts to sell. That is a structural conflict worth naming. Journalists now get paid off by labels is a different and much heavier accusation, aimed at people mostly earning very little, and Blake offers no case. I would keep the structure and drop the word.
Comment sections. Trivially manipulable, obviously gamed, and the least important of the five, because nobody sensible was using them as evidence anyway.
They all break in the same place
What every item on that list has in common is that none of them is the thing anyone actually wants to know.
The real question — did this reach somebody, did it matter to them — is expensive to measure and always has been. So the industry substituted proxies: counts, plays, reviews, comments. Cheap to collect, easy to compare, and close enough to the truth to be useful for a while.
Then the proxies started allocating money. Playlist positions, booking fees, advances and algorithmic reach all began to key off the same handful of numbers, and at that moment faking them stopped being vanity and became a business with a positive return. Fraud is not an aberration in that arrangement. It is the predictable end state of any measure that decides who gets paid, which is why each of Blake's five collapsed in the same way and at roughly the same time.
Which is the same argument as a number without a denominator, one step further along. There the problem was reading a metric without context. Here the metric has stopped describing anything at all.
What survives
The useful move is not to stop measuring. It is to notice which signals are expensive to fake, and those tend to be the ones involving a specific person doing something that costs them.
Somebody buys the record. Somebody buys a ticket and stands in the room until the end rather than leaving after the first drink. A promoter books you a second time, having watched what happened the first time. Somebody sends a message about one track. Somebody comes back a year later.
None of that scales, none of it looks impressive written down, and all of it is absurdly uneconomic to fake — the cost of manufacturing one real person in a room is enormous compared to the cost of manufacturing a hundred thousand plays. The smallness is the feature. Those numbers are small because they are load-bearing.
This is also the practical reason the releases here sit on Bandcamp and Subvert rather than Spotify: a purchase is a decision by a person with a name, and it survives contact with a system where plays no longer mean anything.
The last line is the whole point
Blake could have stopped after the diagnosis, and the post would have travelled just as far. He ended somewhere more useful instead: you are probably doing better than you think.
That sentence does a specific job. If the visible numbers around you are inflated and yours are not, then every comparison you make is between your real figure and somebody's fiction, and you will lose that comparison permanently. Not because the work is worse — because you are the only one in the room reporting honestly.
There is a wrong conclusion available here, and it is worth naming because it is the comfortable one: if it is all fake, nothing counts, so nothing matters. That gets you out of doing the work. The better reading is narrower. The public scoreboard has stopped being information. The room has not, the record sales have not, the person who came back has not.
Count those instead, and the picture is usually less bleak than the feed suggests.
Source: James Blake, Instagram story, June 2026, as carried by Pigeons & Planes on 22 June 2026. Streaming fraud and AI upload figures from Deezer's newsroom (January, April and July 2026), Spotify's September 2025 announcement on AI and spam policies, and reporting on the US Department of Justice case against Michael Smith. Quotations are from Blake's story as posted.
What did James Blake say about the music industry in 2026?
In an Instagram story he listed signals he considers untrustworthy — reviews, comment sections, YouTube counts, streaming numbers, and even whether a song was made by a human — and concluded that no part of the system is unfaked, telling artists they are probably doing better than they think.
How much streaming fraud is there actually?
Enough to be structural rather than anecdotal. Deezer reported in January 2026 that up to 85 per cent of streams on AI-generated tracks were fraudulent and demonetised them, and put overall catalogue fraud at 8 per cent of streams in 2025. Spotify removed more than 75 million spam tracks in the year to September 2025, and since April 2024 has charged labels and distributors per track where flagrant artificial streaming is found on their content.
What can an artist measure that cannot be faked?
Things that cost a real person something: a ticket bought, a record bought, a room that stays full to the end, a promoter who rebooks, someone who comes back a second time. The numbers are small and the effort to fake them is higher than the benefit, which is exactly what makes them informative.