Raw LLM Output Pasted Into Communication Is a UX Failure

There is a specific failure mode that has become common enough to warrant a label: taking whatever an LLM produces and forwarding it directly, without editing, without verification, without any signal that a human was in the loop.

The site dontquotetheai.com makes the case against this pattern. The core argument is not that AI output is wrong — it is that pasting it verbatim treats the recipient as someone who cannot tell the difference, and usually they can.

For engineers, the practical stakes are concrete. A pull request comment generated wholesale by an LLM and posted without review adds noise instead of signal. A Slack message that reads like a ChatGPT completion — hedged, verbose, structured with bullet points that nobody asked for — tells the reader the sender did not think about them. A spec doc drafted by an AI and sent without a pass to strip the filler wastes the reader's time and makes the author look absent.

The distinction the framing draws is between using AI as a drafting tool versus outsourcing the communication entirely. The first is a workflow accelerant. The second is a trust deficit. Senior engineers and technical founders notice the difference faster than most because they read more AI-generated text than most.

The implication for teams building with AI is direct: if your internal culture normalizes unedited AI output as acceptable communication, you are training your team to stop reading carefully. That compounds. Code review quality drops. Spec clarity drops. Debugging conversations get slower.

Editing AI output before sending is not extra work. It is the work. The generation step is just a faster first draft. The team behind the site frames this as basic professional hygiene, not a technology critique. That framing is correct.

Source