Formatting Output & Speaking for Claude

Extract structured information from the following job posting.
Return a valid JSON object — no markdown, no explanation, just JSON.

Schema:
{
  "job_title": string,
  "company": string,
  "location": string | null,
  "remote": "yes" | "no" | "hybrid",
  "salary_range": {
    "min": number | null,
    "max": number | null,
    "currency": string
  },
  "required_skills": string[],
  "experience_years_min": number | null,
  "seniority": "junior" | "mid" | "senior" | "lead" | "unknown"
}

<job_posting>
{job_posting_text}
</job_posting>

# Requesting rich Markdown (rendered UI, documentation)
Write a technical README for this Python library.
Use: H2 headers for sections, code blocks for examples,
bold for key terms, bullet lists for features.
Format: standard GitHub README style.

# Requesting no Markdown (plain text contexts, voice, SMS)
Write a short welcome message for our SMS onboarding flow.
Plain text only — no asterisks, no hashes, no bullet characters.
Just clean sentences.

# Controlling Markdown granularly
Write a product comparison.
Use a Markdown table for the feature comparison.
Use plain prose for the recommendation paragraph.
No headers — this will be embedded in a larger document.

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": "Extract the key financial metrics from this earnings call transcript.\n\n<transcript>{transcript}</transcript>"
        },
        {
            "role": "assistant",
            "content": "{"  # <-- Prefill forces JSON output from character 1
        }
    ]
)

# Claude continues from the "{" — response is guaranteed to start as JSON
raw = "{" + message.content[0].text
import json
data = json.loads(raw)