If all the thought you’ve put into a proposal is three bullet points, I’d rather read those than the document your AI made out of them.
More and more people use AI to create beautiful-looking documents that say very little. Someone types three bullet points and asks for a proposal, a price quote or an internal report. The AI fills in the missing details with average-looking slop. The result looks coherent. It has the shape of a finished document. But the polish hides how little information went into it.
Then someone receives it. They don’t want to read 10 pages of slop, so they put it into their own AI and ask for a summary. They get back something that broadly resembles the original three bullet points. Sometimes it doesn’t even resemble them.
That’s a stupid ritual that we must stop doing. The sender expands a small amount of thought into a long document, and the recipient pays the attention cost of trying to recover that thought, or delegates the recovery to another AI.
When I get a document that obviously smells like AI slop, I ask for the prompt. Send me the fucking prompt. Save us both the time of pretending we needed a polished document when the input already contained everything you had to say.
Our AI starts with more material than it publishes
You might think that’s hypocritical. We run an AI company that creates content.
But our process goes in the opposite direction. Our system starts with an interview with an expert over WhatsApp, through a voice note. Sometimes that interview entails more than one question. The text from these voice notes is often twice or three times the length of the finished article.
The expert has already supplied more material than the reader will see. The AI summarizes it and can bring in information from previous conversations. It has actual source material to work with, rather than a few bullets whose gaps it has to fill.
Summarizing that material is very different from expanding three bullet points into a document. In one case, the work is to reduce the amount of text while retaining the expert’s knowledge. In the other, the amount of text grows without a corresponding increase in what the person supplied.
That’s the difference I care about: signal-to-noise ratio. A polished document with very little signal still gives the reader very little. More words don’t fix the missing thought.
Having substantial expert material to summarize is one reason we can create AI content that doesn’t read like AI slop. It isn’t the only reason. We have other tricks up our sleeve, and I wouldn’t claim that summarization alone explains the result.