I had dinner with the owner of an asset management company recently. The conversation inevitably turned to AI and how he was using it at his firm. It had taken him only a few iterations to figure out the dark secret of internet data.

This fund is at the opposite end of the scale from NBIM, which I wrote about last week. It owns shares in 25 companies and researches three more a week. This is deep research, Warren Buffett style, where outperformance depends on envisaging something that others have yet to see.

The fund’s analysts started by asking Claude questions about the policies of companies that had passed their basic screening. It soon became obvious that the primary source of information for the replies was the companies themselves. This was particularly the case in emerging markets, where financial news is limited and the companies’ influence pervasive.

This is the age-old problem of garbage in, garbage out. It is far more common than you might think. By one account, 90% of everything you see on the internet is advertising.

Clipping is the business of cutting long-form content into soundbites. These are blast-posted over social media by multiple, purpose-built accounts. There are many agencies offering this service, promising far lower cost per thousand views than billboards, print and digital ads.

Popular internet figures such as Mr. Beast and Joe Rogan use clipping to boost their ratings. Trump’s comments about Haitians eating cats in Springfield were clipped extensively during the 2024 election campaign. This was a deliberate strategy to deny airtime to Kamala Harris, as she attempted to establish herself in the race.

There are two ways in which clipping works. The first is if the posts resonate with real, human accounts. Justin Bieber uses it to promote his work. His songs are popular and clipping is a means of accelerating reach that they might have got in any case.

The second way is when a niche influencer manufactures interest and catches the eye of the mainstream media. Reputable news organisations report on the emerging phenomenon, creating interest that was not there before. Several of the darker voices from the manosphere are believed to have started this way. They were boosted by the publications writing articles about how abhorrent they were.

The pernicious part of clipping is the accompanying narrative campaign. While boosting music posts may seem harmless, the power comes from the comments. It’s well known that people like to fit in and the first comment they see is the most likely to shape their opinion. Narrative campaigns may be used to spread positive or negative opinion about artists, politicians or products, depending on who is paying the bill.

Dan Brahmy of bot-detection firm Cyabra compares clipping to a soccer tournament in which teams bribe the referees. The teams that don’t know to do this keep being eliminated in the quarter-finals. If your marketing and advertising fail to land, it may be because you are paying the wrong people to promote them.

Joe Lim of Floodify, a now-closed clipping agency, estimates that up to 90% of internet traffic is soft advertising. Viral trends are manufactured, popular celebrities are kept in the spotlight, and newcomers must pay the piper or face irrelevance.

I often hear people complaining that they have fallen foul of social media algorithms. They attribute malicious intent to these programs, which they assume exist to promote slop at the expense of quality content. But algorithms just promote what appears to be popular to keep people scrolling.

Clipping illustrates that it is humans who have captured the algorithms. Social media companies are mere conduits of manufactured interest. Marketing agencies dictate what we see and hear based on which clients pay the most.

The analysts at the asset management company no longer ask Claude about a company’s policies. If they want to investigate ESG breaches, for example, they ask the LLM to surface specific examples. This forces the search towards alternative sources, ensuring that the analysts see the counter-narrative.

This is the same logic that I recommended adopting when using LLMs for corporate strategy. Rather than having the machine resurface fashionable management speak, use it to critique a strategy you have already created. This way you don’t outsource your intelligence to algorithms. You are using AI to widen your horizons while sharpening your reasoning.

Social media has been captured by commercial interests that learned how to play the system. There is no reason to believe that AI will be any different. Models that are trained on internet content and which search the internet when generating replies are subject to the quality of what they read. Rather than worrying about machine superintelligence replacing us, we should be concerned about how much the machines replicate our fallacies.

“Your Feed is Fake” - Vulture, May 13, 2026