Comment capturer les requêtes
dérivées de ChatGPT

# fanout.py · read ChatGPT's own search calls off the wire
import asyncio, json
from playwright.async_api import async_playwright

PROMPT = "meilleures plateformes de sécurité IA pour PME"
found  = set()

def harvest(evt):
    msg = evt.get("message") or {}
    # the fan-out queries ride inside the search tool call
    if msg.get("author", {}).get("name") != "web":
        return
    for part in msg.get("content", {}).get("parts", []):
        if isinstance(part, str) and part.startswith('search("'):
            found.add(part[8:-2])

async def main():
    async with async_playwright() as pw:
        # a real logged-in profile. clean headless gets challenged.
        ctx  = await pw.chromium.launch_persistent_context(
            "~/.chatgpt-profile", headless=False)
        page = await ctx.new_page()

        async def on_response(res):
            if "/backend-api/conversation" not in res.url:
                return
            body = await res.text()          # SSE, not JSON
            for line in body.splitlines():
                if line.startswith("data: "):
                    try: harvest(json.loads(line[6:]))
                    except ValueError: pass

        page.on("response", on_response)
        await page.goto("https://chatgpt.com/?search=1")
        await page.fill("#prompt-textarea", PROMPT)
        await page.keyboard.press("Enter")
        await page.wait_for_timeout(45_000)  # no "done" signal
        print(json.dumps(sorted(found), indent=2))

asyncio.run(main())