ChatGPT 导出 JSON — 将对话导出并保存为结构化数据（2026）

{
  "metadata": {
    "title": "How to Process CSV Files in Python",
    "user": {
      "name": "John Smith",
      "email": "[email protected]"
    },
    "dates": {
      "created": "2026/04/15 14:30:45",
      "updated": "2026/04/15 15:22:10",
      "exported": "2026/04/17 09:00:33"
    },
    "link": "https://chatgpt.com/c/abc123-def456"
  },
  "messages": [
    {
      "role": "Prompt",
      "say": "How do I read a CSV file and filter specific columns in Python?",
      "time": "2026/04/15 2:30:45 PM"
    },
    {
      "role": "Response",
      "say": "You can use the pandas library:\n\nimport pandas as pd\ndf = pd.read_csv('data.csv')\nresult = df[['name', 'age']]\n\nThis filters the name and age columns.",
      "time": "2026/04/15 2:31:22 PM"
    }
  ]
}

{
  "title": "Conversation Title",
  "create_time": 1713350400.0,
  "update_time": 1713354000.0,
  "mapping": {
    "aaa-msg-id-1": {
      "id": "aaa-msg-id-1",
      "message": {
        "author": { "role": "user" },
        "content": {
          "content_type": "text",
          "parts": ["Your message here"]
        },
        "create_time": 1713350400.0
      },
      "parent": null,
      "children": ["aaa-msg-id-2"]
    },
    "aaa-msg-id-2": {
      "id": "aaa-msg-id-2",
      "message": {
        "author": { "role": "assistant" },
        "content": {
          "content_type": "text",
          "parts": ["The AI response here"]
        },
        "create_time": 1713350415.0
      },
      "parent": "aaa-msg-id-1",
      "children": []
    }
  }
}

import json
from collections import Counter
 
with open("ChatGPT-my-conversation.json") as f:
    data = json.load(f)
 
# Count messages by role
roles = Counter(msg["role"] for msg in data["messages"])
print(f"Your prompts: {roles.get('Prompt', 0)}")
print(f"AI responses: {roles.get('Response', 0)}")
 
# Find longest response (skip unselected messages with empty "say")
responses = [m for m in data["messages"] if m["role"] == "Response" and m["say"]]
longest = max(responses, key=lambda m: len(m["say"]))
print(f"Longest response: {len(longest['say'])} chars")
 
# Average response length
avg = sum(len(m["say"]) for m in responses) // len(responses)
print(f"Average response: {avg} chars")

import json
import re
 
with open("ChatGPT-my-conversation.json") as f:
    data = json.load(f)
 
for msg in data["messages"]:
    if msg["role"] == "Response" and msg["say"]:
        blocks = re.findall(r"```(\w+)?\n(.*?)```", msg["say"], re.DOTALL)
        for lang, code in blocks:
            print(f"--- {lang or 'text'} ---")
            print(code)

import json
import csv
 
with open("ChatGPT-my-conversation.json") as f:
    data = json.load(f)
 
with open("conversation.csv", "w", newline="", encoding="utf-8") as f:
    writer = csv.writer(f)
    writer.writerow(["Role", "Content", "Time", "Word Count"])
    for msg in data["messages"]:
        if msg["say"]:  # skip unselected messages
            writer.writerow([
                msg["role"],
                msg["say"],
                msg.get("time", ""),
                len(msg["say"].split())
            ])

import json
 
ROLE_MAP = {"Prompt": "user", "Response": "assistant"}
 
with open("ChatGPT-my-conversation.json") as f:
    data = json.load(f)
 
# Convert to OpenAI fine-tuning format
training_data = []
messages = [m for m in data["messages"] if m["say"]]  # skip empty
for i in range(0, len(messages) - 1, 2):
    if messages[i]["role"] == "Prompt" and messages[i+1]["role"] == "Response":
        training_data.append({
            "messages": [
                {"role": "user", "content": messages[i]["say"]},
                {"role": "assistant", "content": messages[i+1]["say"]}
            ]
        })
 
with open("training.jsonl", "w") as f:
    for item in training_data:
        f.write(json.dumps(item) + "\n")
 
print(f"Created {len(training_data)} training examples")

# Count total messages
jq '.messages | length' ChatGPT-my-conversation.json
 
# Get all your prompts (non-empty only)
jq -r '.messages[] | select(.role == "Prompt" and .say != "") | .say' ChatGPT-my-conversation.json
 
# Show conversation title and message count
jq '{title: .metadata.title, messages: (.messages | length)}' ChatGPT-my-conversation.json
 
# Count messages by role
jq '[.messages[].role] | group_by(.) | map({(.[0]): length}) | add' ChatGPT-my-conversation.json
 
# Word count per AI response
jq '.messages[] | select(.role == "Response" and .say != "") | {words: (.say | split(" ") | length)}' ChatGPT-my-conversation.json

import json
import sqlite3
 
with open("ChatGPT-my-conversation.json") as f:
    data = json.load(f)
 
conn = sqlite3.connect("chatgpt.db")
conn.execute("""
    CREATE TABLE IF NOT EXISTS messages (
        id INTEGER PRIMARY KEY,
        role TEXT,
        content TEXT,
        time TEXT,
        word_count INTEGER
    )
""")
 
for msg in data["messages"]:
    if msg["say"]:  # skip unselected messages
        conn.execute(
            "INSERT INTO messages (role, content, time, word_count) VALUES (?, ?, ?, ?)",
            (msg["role"], msg["say"], msg.get("time", ""), len(msg["say"].split()))
        )
 
conn.commit()
print(f"Imported {sum(1 for m in data['messages'] if m['say'])} messages into chatgpt.db")

-- Message stats by role
SELECT role, COUNT(*) as count, AVG(word_count) as avg_words
FROM messages GROUP BY role;
 
-- Full-text search across all messages
SELECT role, content FROM messages WHERE content LIKE '%repository pattern%';

mkdir chatgpt-archive && cd chatgpt-archive
git init
 
# After each export, add and commit
git add ChatGPT-python-api-design.json
git commit -m "Export: Python API Design conversation"

# Using jq (install: brew install jq / apt install jq)
jq . ChatGPT-my-conversation.json > /dev/null && echo "Valid JSON"
 
# Using Python (no install needed)
python3 -m json.tool ChatGPT-my-conversation.json > /dev/null && echo "Valid JSON"

# Pretty-print to understand the structure
jq '.' ChatGPT-my-conversation.json | head -30
 
# Quick look at metadata
jq '.metadata' ChatGPT-my-conversation.json
 
# Count messages by role
jq '[.messages[].role] | group_by(.) | map({(.[0]): length}) | add' ChatGPT-my-conversation.json

import json
import glob
 
all_conversations = []
for file in sorted(glob.glob("ChatGPT-*.json")):
    with open(file) as f:
        conv = json.load(f)
        all_conversations.append({
            "title": conv["metadata"]["title"],
            "exported": conv["metadata"]["dates"]["exported"],
            "messages": [m for m in conv["messages"] if m["say"]]
        })
 
with open("all-conversations.json", "w") as f:
    json.dump(all_conversations, f, indent=2, ensure_ascii=False)
 
print(f"Merged {len(all_conversations)} conversations")