← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Conceptupdated Wed Nov 04 2026 08:00:00 GMT+0800 (Philippine Standard Time)

AI Slop

ai-slopworkslopai-writingcognitive-offloadingread-budgetqualityculturemerriam-websterdeezerbetterupglean

AI Slop

Merriam-Webster's 2025 word of the year: "digital content of low quality that is produced usually in quantity by means of artificial intelligence." Before that, slop was what you fed to pigs. Nufar Gaspar's working definition is sharper and is the one the vault uses: work where the person who should have done the thinking skipped it, partly or altogether, and the cost moved to whoever received the work. Humans produced slop long before AI; AI made it cheap and in volume (How to Deslopify Your AI (AI Daily Brief)).

The definition matters because it supplies the standard. Non-slop work is the most effective output for the least effort on the receiving side, so compacting an artifact to the format, length and quality the audience needs is the AI user's job, not the reader's. Nathaniel Whittemore's gloss on why people react so viscerally: the sender is "leaving them holding the work bag." It is the artifact-side twin of Cognitive Offloading (the producer skipped the thinking) and the receiver-side problem behind Authenticity Crisis (AI Communication) (everything sounds the same) and Verification Tax (someone still has to check it).

The numbers (as cited in the episode; verified 2026-11-04)

Figure Source as cited Note
>50% of music uploaded to Deezer daily is fully AI-generated; AI tracks are ~1–3% of listens Deezer the scale outside the office
40% of US desk workers received "workslop" in the past month; 1h56m cost per instance on the receiving side; ~$186 per employee per month; ~15% of content received Stanford Social Media Lab + BetterUp Labs, HBR, Sept 2025, n=1,150 the receiver's bill
69% of AI users admit to "botshitting" (shipping AI-generated work they haven't verified); 41% sometimes deliver work they couldn't explain; 74% where only productivity is measured vs 64% where quality is measured too Glean Work AI Index 2026, n=6,000 (see Botsitting) the producer's confession; Gaspar's "without reading" is a paraphrase
41% of longer LinkedIn posts detected as AI-drafted Pangram, via AI Often Makes Writing Worse (Economist) the channel share

Where it comes from

In the machine. A language model predicts the most likely answer, so a generic question returns the average of what the internet thinks. Post-training narrows it further through Typicality Bias: raters reward what feels familiar, the range contracts, and the default comes out long and listy because length looks like effort. Models from different vendors converge on the same answers (the NeurIPS 2025 Best Paper Artificial Hivemind measured it across 70+ models), so switching tools buys no originality; image models each carry a house style. Gaspar's incognito test (three frontier models, memory off, one generic LinkedIn prompt) produced near-identical posts in structure and content.

In us. The brain is a cognitive miser; pilots call the result automation complacency (when the machine is usually right you stop checking); knowledge-worker research finds that more trust in the AI means less critical thinking applied to its output. Gaspar's three routes to slop, almost never from bad intent:

  1. It's easy. It looks fine, you are under pressure, you forward it.
  2. Not knowing the craft. A naive ask gets a naive result; the senior person gets better work from the same tool because they know what to ask. This is the trainable route and maps onto Capability Gap.
  3. Outsourcing the thinking. Asking the tool what it thinks before deciding what you think. The most troubling route, and the one Cognitive Offloading and Productive Resistance already track from the usage and design sides.

How to detect it

Gaspar's five questions, for your own work or someone else's:

  1. Could the sender defend every sentence without rereading it?
  2. Did the sender read all of it before it went out?
  3. Is there anything only this person could have contributed: a point of view, a decision, real data? If anyone could sign it, it is not high-quality work.
  4. Is there a more effective way to convey this? Does it save the reader time or cost them time?
  5. Would you put your name on it in capital letters?

Technology does not substitute for the questions. Detectors measure structure (bullet density, uniform sentence length) and light up most of LinkedIn, but they do not detect quality or originality; humanizers are "slop trying not to look like slop", still flagged about three-quarters of the time, and the reader's complaint was never the style but the missing point of view.

What it is not

Slop is proportional to the stakes. An AI-written internal email that is concise and does its job is not slop, because its purpose was never to provoke thought (Whittemore); nobody wants artisanal meeting notes (Gaspar). The burden rises with what the artifact is for: anything that carries your brand, is customer-facing, or is supposed to be your original work needs the full Deslopifying Playbook. Related: Wade Foster's "runaway train" of low-effort volume in internal channels and the declare-your-role rule in AI Communication Etiquette; the Economist's finding that AI lifts the median writer and flattens the best (AI Often Makes Writing Worse (Economist)).

The culture fix

Name it when it arrives, especially as a leader ("this isn't a good use of my time; send me your take in five sentences"), hold the gate, never let it reach a customer, and do not hide AI but label it: Figma's published rules for AI writing require AI-written content to say so at the top so readers know how to read it. Gaspar's bet is that once the agentic hype passes, work with a human fingerprint becomes the premium product and disciplined quality becomes job security.

Sources