The comparison is not an analogy someone invented after the fact. It is where the word came from.
On 7 May 2024, the writer deepfates observed that “slop” was becoming a term of art, arriving in the language the way “spam” had. Simon Willison picked it up the next day, titled his post “Slop is the new name for unwanted AI-generated content”, and drew the line that gave the word its definition: plenty of promotional email is legitimate and plenty of machine output is fine, while material generated mindlessly and thrust upon someone who did not ask for it is exactly what the new word describes.
So the two words are siblings. The differences are what make them worth telling apart.
The one-line difference
Spam is unwanted bulk content, historically email, that wants something from you. It is addressed to a person, it usually makes an offer, and it arrives in a channel you own.
AI slop is unwanted bulk content that wants something from an algorithm. It is addressed to a ranking system, it often offers nothing, and it arrives in a channel somebody else owns. The payoff comes later, through reach, advertising revenue, or a funnel at the end of the funnel.
The comparison
| Spam | AI slop | |
|---|---|---|
| Intended recipient | You | The recommendation system |
| Typical channel | Email, comments, messages | Feeds, search results, catalogues, storefronts |
| What it wants | A click, a payment, a credential | Reach, and whatever reach converts into |
| Production cost | Low | Near zero |
| Age | Since the 1990s | Named in 2024 |
| Legal status | Defined by statute in many jurisdictions, with enforcement bodies | No dedicated legal category |
| Main defense today | Filters, blocklists, authentication standards | Ranking changes, platform policy, reader judgment |
| Characteristic harm | Wasted time, fraud, malware | Wasted time, degraded search and feeds, eroded trust |
Why the intended recipient changes everything
Spam needs to reach you, which makes it addressable. That is why spam got a technical arms race: sender authentication, reputation systems, filters trained on decades of examples. The fight happens at the point of delivery, and it can be automated because the target is a message going to a mailbox.
Slop does not need to reach you. It needs to be ranked. A generated video that nobody watches still performs its function if it occupies a slot, and there are always more attempts behind it. Filtering it means changing what a ranking system rewards, which is a decision a platform has to make deliberately and pay for in engagement metrics.
That is the practical consequence of the difference, and it explains why slop feels less solved than spam. The defense for spam sits between the sender and you. The defense for slop would have to sit inside the machinery that decides what you see.
The economics are the same
Strip away the channels and both run on one calculation: when distribution is nearly free, volume is a rational strategy even if most attempts fail.
With spam, the near-zero cost was the cost of sending an email. With slop, it is the cost of producing the thing in the first place, which has collapsed. A creator who posts hundreds of generated videos a week is running a brute-force search for the one that travels. Wikipedia's documentation of the term records exactly this pattern, noting that slop proliferates where platforms pay for engagement, and that the incentive draws producers who are not the audience for what they make.
The money is different in one respect. Spam usually wants a transaction. Slop often wants attention that gets sold onward, which means the platform and the slop producer are not necessarily adversaries. Sometimes they are business partners with different tastes.
Where the two now overlap
Three areas have stopped being separable.
Search and retail listings. Producing pages purely to rank predates generative AI by years. What changed is the cost, which is why academic writing on slop describes “fodder for websites whose only purpose appears to be optimizing for search engines” as a canonical example.
Bot armies and comment sections. Generated comments, replies and reviews that used to require human farms now come from models. The behavior is spam. The material is slop.
Funnels and phishing. Fabricated people, synthetic product photos and generated landing pages now serve fraud directly. A Columbia University report on the topic makes the point that slop often functions as cover rather than as the payload, normalizing synthetic material so that targeted manipulation has somewhere to hide.
Why the distinction still earns its keep
Because it tells you which lever to reach for. If something is spam, you can filter it, report it, and expect a legal framework to have an opinion. If something is slop, filtering does not solve it, because it is not trying to get into your inbox. It is trying to be the thing your feed shows you instead.
And it tells you who to argue with. Spam is a crime or a nuisance committed by a sender. Slop is an equilibrium that a platform's incentives produce, which is a less satisfying target and a more accurate one. Complaining about slop to the person who made it rarely changes anything. Asking what a platform pays for is the version of the question that has somewhere to go. For how that plays out platform by platform, see AI slop on social media.