Indictment: A prompt so vague it could produce anything from a LinkedIn fluff piece to a Wikipedia stub — 'good', 'not too long', and 'main points people care about' are three non-instructions dressed up as a brief.
see the specimen they pasted
write me a blog post about AI agents thats good and not too long, make it engaging and hit the main points people care about
“thats good”
'Good' is not a constraint — it tells the model nothing about tone, quality bar, audience, or what failure looks like, so the model defaults to generic.
swap: 'thats good' → 'written for a technically curious but non-expert audience, no jargon without a one-line definition'
“not too long”
'Not too long' is a feeling, not a format contract — the model will happily produce 1,200 words and consider that 'not too long'.
add: explicit word count cap, e.g. '600–800 words, no more'
“make it engaging”
Tells the model how to feel about the output instead of what to do — 'engaging' is an outcome, not a technique; the model gets no instruction on how to achieve it.
swap: 'make it engaging' → 'open with a concrete real-world scenario, use one analogy per section, no bullet-point padding'
“hit the main points people care about”
Zero signal on which audience, which use case, or which 'main points' — this is the entire brief outsourced back to the model, guaranteeing a generic listicle.
add: explicit topic list, e.g. 'cover: what an AI agent is, how it differs from a chatbot, 2–3 real-world use cases, and one honest limitation'
You are a technology journalist who writes for curious, non-technical readers — think MIT Technology Review's explainer tone, not a vendor blog. Write a blog post explaining AI agents. Cover exactly these four points in this order: 1. What an AI agent is (one crisp definition, one concrete analogy) 2. How it differs from a standard chatbot or LLM prompt 3. Two to three real-world use cases (specific industries or tasks, not vague possibilities) 4. One honest limitation or risk readers should know Rules: - Word count: 600–800 words. Hard cap. - Open with a one-paragraph real-world scenario that makes the reader feel the problem before you name the technology. - No bullet-point lists inside the body — prose only. - No jargon without an immediate plain-English gloss in the same sentence. - No hype language: banned phrases include 'revolutionary', 'game-changing', 'the future of', 'unlock potential'. - End with one sentence that gives the reader a concrete next step or question to sit with — not a call to action. Audience: someone who uses ChatGPT casually and wonders what all the 'agents' buzz is about. They are skeptical of hype and will stop reading if the first paragraph feels like a press release.
https://getmeerkat.dev/roast/h77ft5zty2