Paul Graham claims that AI has an unexpected benefit: clickbait titles that tease but do not clarify are becoming ineffective. Instead of clicking through to see "the new theory explaining x," he directly asks ChatGPT what that theory is.

This is another side of the same observation. He has just mentioned that long articles that take too long to get to the point will be replaced by shorter ones; now even the distribution tactic of "hiding the point first and then trading for clicks" is also failing. The title creates a gap, and the body is responsible for filling that gap with a fee—AI models have removed the need for sites to fill that gap.

He has long viewed density as a defense against summary walls: writing an article and then deleting parts can hurt the content, and summaries inevitably lose some points. Clickbait takes the opposite approach, deliberately leaving gaps. With a model that allows readers to ask immediate follow-up questions, gaps no longer need to be repaid with clicks. Previously, he could identify ghostwritten content from words like "delve" and hardcore news emails, comparing the experience of recognizing AI to being deceived.

Advertising and media metrics still rely on clicks and dwell time. Surveys from companies like BetterUp show that a significant amount of low-quality generated content is received in American workplaces, with about half of respondents lowering their evaluations of the sender's creativity, ability, and credibility. Related research from Ohio State University also indicates that recipients often interpret generated private messages as lazy and insincere. The metrics of traffic and trust are beginning to diverge.

Listicles, SEO long-form articles, and "you won't believe the reason" belong to the same inventory. The training corpus for models is already filled with this type of text, so it can produce clickbait while also exposing it to readers. Graham also finds ChatGPT's tone reminiscent of corporate communications and HR: it loves using upbeat bullet points, unlike real conversations around it.

In market mechanisms, buyers are readers who do not want to pay for a riddle with a click, while sellers are sites that rely on the gap in titles to recover advertising costs. Funding is shifting from click funnels to question interfaces. Beneficiaries are knowledge sources where answers can be extracted in a single sentence, and authors who have already written to the point of no longer hiding it; those under pressure are programmatic advertising, affiliate marketing, and content factories that bury conclusions on the third screen. The event-driven shift comes from the "right to explain" moving from links to dialogue boxes.

Additionally, he did not distinguish the proportions of academic preprints, product reviews, and gossip entertainment that are skipped, nor did he quantify how often the "theory reduction" provided by ChatGPT is incorrect.

Source: Public Information