Paste a long article's URL into ChatGPT and out comes a summary. Perplexity pulls together several pieces on the same topic and answers your question directly. As these tools have settled into everyday use, a single piece of advice has spread among people who make content: write so that AI can read it easily. Break your structure into clear steps, put the key point at the top of each paragraph, mark the flow with subheadings — this advice keeps showing up across platforms.
The advice has its logic. Google's AI Overview feature, launched in May 2024, places an AI-generated summary at the top of the search results page. If an AI is reading the content and generating the answer, it stands to reason that writing the AI can parse easily has a better shot at being chosen as a source — and that failing to become a source means fewer clicks. It's not an unreasonable line of thinking. But follow the premise behind this advice one step further, and it starts to crack.
Writing AI Reads Easily Is Also Writing Readers Skip Easily
What are the traits of writing that AI summarizes well? The argument unfolds step by step, the key claim sits clearly at the front or back of each paragraph, and the examples support the argument directly. Even before AI summarization entered the picture, a reader could skim just the subheadings and opening sentences of writing like this and grasp almost everything.
That's where the problem starts. If an article's value survives the trip through AI summarization intact, that means the original, before summarization, never gave the reader more than the summary does. If the value fits inside the summary, the original exists for no reason other than its length. Flip that around, and you get this: writing that AI summarizes easily was writing that could have been a summary from the start.
This is where two goals that "AI-friendly writing" advice lumps together need to be pulled apart. One goal is getting included as a source for AI summaries. Being chosen as a source requires that AI can parse the content, which favors writing with a clear structure. The other goal is getting a reader who has already seen the AI summary to click through to the original. The two goals look like they overlap, but they actually call for different things. If inclusion as a source is the goal, delivering the claim clearly is the right move. If getting readers to seek out the original after the summary is the goal, the writing needs to hold something the summary doesn't capture.
A Hacker News discussion on AI reading content on readers' behalf pulled in over 400 points — a sign that this has already become a live question for content producers. But if the question stops at "how do we optimize for AI," the strategy being built is one that confuses the two goals.
The Kind of Information a Summary Can't Carry
There's territory AI summaries handle poorly. They can carry over a claim itself, but they rarely carry over the conditions under which that claim holds true or breaks down.
Say someone about to open a café wants to know what rent-to-revenue ratio is safe. An AI summary quickly hands over the claim: "rent should stay within 10 to 15 percent of monthly revenue." The reader gets that number and never clicks through to the original. It's enough. But if the AI summary instead reads, "this benchmark shifts depending on the type of commercial district and the fixed-cost structure, and in some downtown districts a business can meet this benchmark and still not turn a profit," the reader goes looking for the original because they want to know which type their own district falls into.
What's the difference between the two? The first summary carried a claim. The second carried the conditions under which the claim breaks down. The reader clicks through to the original because the summary opened up the boundary of those conditions without filling it in.
This is a difference in kind, not degree. "What is it" gets summarized. "Under what conditions does this change" resists summarization. Writing that deals in the latter can turn an AI summary into a channel that sends readers back to the original.
Having this second kind of information has nothing to do with length. A short piece can hold the boundary of a condition, and a long piece that just lines up claims will still get summarized. AI-friendly structure does make writing easier to summarize — that much is true. But what pulls readers in after the summary isn't the structure; it's the kind of information the structure holds. If you're going to change how you write, here's where to start: state the key claim, then write one more paragraph on the situation where that claim doesn't hold. An AI summary will try to capture that paragraph and fail to capture it completely — and it's that incompleteness that sends the reader to the original.
