{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihnbpejhug3osbo4emise7gg5mogu2tfllbluw6mjvtvfssjmjmw4",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mhgrwitt2ba2"
  },
  "path": "/t/structure-vs-tone-in-real-world-llm-workflows/1377238#post_1",
  "publishedAt": "2026-03-19T19:42:43.000Z",
  "site": "https://community.openai.com",
  "textContent": "I’ve been testing structured LLM workflows in real-world use (clinical + operational), and something interesting came up.\n\nA lot of discussion around “humanizing” outputs focuses on tone or wording.\n\nWhat’s been more impactful in practice is structure and cognitive load.\n\nExample:\n\nInput:\nClinic: Local Animal Hospital\n\nEntry 1\nDate: March 20, 2026\nTime: 10:00 AM – 6:30 PM\n\nEntry 2\nDate: March 21, 2026\nTime: 9:00 AM – 3:00 PM\n\nOutput:\n→ structured timesheet entries (multi-day)\n→ a single combined invoice (auto-calculated totals)\n→ a ready-to-send email referencing a PDF attachment\n\nAll consistent, no reformatting needed.\n\nWhat made the difference wasn’t stylistic prompting — it was:\n\n• enforcing consistent output structure\n• separating input variability from output format\n• designing for immediate usability (not completeness)\n\nCurious if others working with LLMs in real workflows have found structure to matter more than phrasing.",
  "title": "Structure vs Tone in Real-World LLM Workflows"
}