answerline

Python client for the AnswerLine API — structured JSON answers from AI assistants (ChatGPT, Perplexity, Gemini, Copilot, Grok) and Google surfaces (Search, News, AI Overview, AI Mode), on httpx.

ChatGPT, Gemini, Copilot, Google Search (with AI Overview), AI Mode and Google News are served today; Perplexity and Grok are paused for now and coming back soon. Coverage lists what is served and from which countries. A monitor call to a paused engine fails with a 404 and an async task with a 422, each saying so, and nothing is billed.

pip install answerline

from answerline import Client

with Client("sk_...") as client:
    # Synchronous monitor calls: one method per engine (chatgpt, perplexity, gemini,
    # copilot, grok, aimode, google, google_news), or client.monitor(engine, **body).
    answer = client.monitor.chatgpt(prompt="iPhone 17 or Galaxy S26?", country="US")

    # Async tasks: create, poll, or run to a terminal status.
    task = client.async_tasks.run(
        "CHATGPT",
        {"prompt": "iPhone 17 or Galaxy S26?", "country": "US"},
        idempotency_key="phone-compare-1",
    )

    client.credits()  # {"remaining", "perCycle", "cycleResetsAt"}

AsyncClient has the same methods, awaitable. async_tasks.create_batch submits up to 500 tasks in one call; async_tasks.wait polls a task to COMPLETED or FAILED; async_tasks.status and async_tasks.clear_queue manage the queue. Prefer a webhook (webhook_url= on the task) over polling for many tasks — verify deliveries with verify_webhook(body, header, secret) against the Webhook-Signature header.

Calls raise ApiError (status, stable code, parsed body, and meta with the rate-limit, concurrency, credit and request-id headers); transport failures raise TimeoutError or ConnectionError. 429s and safely replayable failures are retried with jittered backoff; timeout, sync_timeout, max_retries and http_client are constructor options.

Measuring brand visibility

One answer is one draw from an engine that varies its answers. visibility.measure asks each engine the same prompt samples times as separate async tasks and returns the results per engine; with a brand it also runs brand_check over them: how many answers mention the brand (with a Wilson 95% interval), its average rank among the brands you list, how many cite its own domains, and which domains get cited when it is absent.

from answerline import brand_check

measured = client.visibility.measure(
    "Which CRM is best for a 10-person sales team?",
    ["chatgpt", "gemini"],
    "US",
    samples=12,
    rounds=4,
    spacing_seconds=2 * 60 * 60,  # 4 rounds of 3, two hours apart
    brand={"name": "Acme CRM", "aliases": ["Acme"], "domains": ["acme.com"]},
    competitors=[{"name": "Globex"}, {"name": "Initech"}],
)
brand = measured["chatgpt"]["brand_check"]["brand"]  # mentioned, runs, interval, average_rank, ...

# Or check results you already have, one engine at a time:
brand_check(results, {"name": "Acme CRM"}, [{"name": "Globex"}])

Samples spread over hours measure how much the engine varies its own answer to that prompt from that country, not what different users see. Small samples give wide intervals: take 10 or more per engine. See Measuring visibility.