स्तम्भ Stambh · Udyamnīti

AI Slop = AI Flop?

Half the new articles on the web are now written by machines. The search engines and the answer engines still quote humans. The tool is not the problem. Having nothing to say is.

Nikhil Sharma · उद्यमनीति Udyamnīti · 17 October 2026 · 5 min read

Fabulous.Media sells AI-led content and visibility work. The business whose figures appear under “The proof” is one I run.

When the dictionaries name a thing the market has usually already priced it.

Last December Merriam-Webster chose its word of the year. It chose slop. Its definition: digital content of low quality that is produced usually in quantity by means of artificial intelligence. Australia’s Macquarie Dictionary went further and chose the full phrase. AI slop.

So the question in the title deserves a straight answer. Is AI slop an AI flop?

For the brand that publishes it, yes. For the tool that wrote it, no. The difference is the whole of this column.

The flood

Graphite studied 65,000 English articles from the open web published between 2020 and May 2025. By late 2024 machine-written articles briefly outnumbered human ones. Since then the split has stayed close to half and half.

Making content now costs almost nothing. That is the attraction. It is also the trap.

The flop

The same researchers then looked at what gets found. Across more than 31,000 Google searches 86 per cent of the articles that ranked were written by people. Only 7 per cent of the first results were machine-written. In ChatGPT and Perplexity answers 82 per cent of the cited articles were human.

Half of what is published. A seventh of what ranks.

Read those two numbers together and the verdict is plain.

Google has stopped being patient about it. Its spam policy has a name for the practice: scaled content abuse. The spam update that ran from 18 to 21 August this year hit sites that had pumped out programmatic pages with machine text on top. One lost rankings for more than 200,000 searches. An affiliate site whose pages were flagged as 97 per cent machine-written lost 14,000. The analyst who studied the cases warned that the losses now reach into AI Overviews and AI Mode too. Slop does not just fail to win the answer. It gets you removed from it.

YouTube moved first. Since July 2025 its partner programme refuses to pay for what it calls inauthentic content: mass-produced and repetitive videos.

And buyers notice. In August DoubleVerify reported that 42 per cent of consumers across Europe the Middle East and Africa would feel worse about a brand that ran low-quality AI ads. In the UK it was 49 per cent. Only 24 per cent across the region would feel better.

Fewer people find it. The platforms stop paying for it. The people who do see it think less of you. That is a flop.

The tool is not the culprit

Here is the part the anti-AI crowd leaves out.

Ahrefs ran its detector over half a million pages that Google’s AI Overviews cite. Pure human writing was 8.6 per cent of them. Pure machine writing was 3.6 per cent. The other 87.8 per cent were a mix. Most of what the machine quotes was written by a person working with a machine. The correlation between how much AI a page used and where it was cited was 0.017. Effectively zero.

So Google does not care who held the pen. It cares whether the page says something no other page says.

That is the real definition of slop. Not AI-written. Information-free. A page that restates the ten pages above it adds nothing an answer engine needs. It will be summarised and never quoted. A page with a number nobody else has or a result only you can show or a price you are willing to print will be quoted. The machine has to name its source.

Slop does pay. Someone else.

I will not pretend slop never makes money. Kapwing set up a fresh YouTube account last year. About one in five of the first 500 Shorts it was shown were AI-generated. The most-viewed slop channel in the world in its study was Indian. It had more than two billion views and an estimated $4 million a year in ad revenue.

That is a business. It is not a brand. The money comes from the platform’s feed while the feed allows it, and the July 2025 rule shows how quickly that can change. A company that sells to customers has nothing to gain from it. Its customers are not looking for content. They are looking for a reason to trust it.

The proof

I run a business that sells to buyers abroad. It has no marketing team. I rebuilt it to be found by AI first and I used AI to do the rebuilding.

Since 4 July it has received 67 enquiries from AI answers across 12 countries. They are worth about ₹1.5 crore. Between ₹40 and 50 lakh of that has already converted.

None of that came from volume. It came from facts a machine could read and quote. AI did the heavy lifting. The business supplied the substance. The longer argument is in [Revenue Defence in the AI Era](/soch/vimarsh/revenue-defence-in-the-ai-era).

My interest is plain. Fabulous.Media sells this work. Check every figure here against its source below.

The test

Before anything goes out ask three questions.

Does it contain one fact a competitor could not have written? A price. A result. A date. A name your customer would recognise.

Would you put your own name on it?

If an answer engine quoted only one line from it, is there a line worth quoting?

Three yes answers and it does not matter who drafted it. Three no answers and it does not matter how many you publish.

So, AI slop equals AI flop? For the brand, yes, every time. AI used to say something true and specific is not slop. It is the cheapest good writing in history.

Gurugram, 17 October 2026.

Sources, checked 17 October 2026. Merriam-Webster’s word of the year, TechCrunch, 15 December 2025. Macquarie Dictionary’s word of the year, The Nightly. Graphite’s web sample, Axios, 14 October 2025. Graphite, How does AI-generated content perform in search and answer engines?, 14 October 2025. Glenn Gabe, August 2026 Google spam update case studies. YouTube’s inauthentic content rule from 15 July 2025, Gulf News. DoubleVerify’s EMEA Global Insights 2026, Advanced Television, 6 August 2026. Ahrefs, AI Overviews cite AI-generated content more than human writing, 14 July 2025. Kapwing’s study, The Decoder, 29 December 2025. Graphite and Ahrefs both rely on AI detectors, which make mistakes. The figures for the business I run are its own enquiry records and are not public.