Copilot sources

`sources` returns the cited pages in order, so citation share is measured the same way as on other engines. `citationPills` records which sources sit behind each inline chip. The searches Copilot ran are coming.

Returned in result

Request

curl -X POST "https://api.answerline.dev/v1/monitor/copilot" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Cheapest way to send money abroad","country":"US"}'

What Copilot’s sources are

A Copilot source is a web page Copilot cites for its answer. Copilot marks cited passages with chips and lists the referenced pages, and when several citations follow one another it groups them into one chip. This API returns the page list as sources and the chip structure as citationPills. Because Copilot searches before it answers, the queries it ran (searchQueries) are coming; they are not returned yet.

Today sources shows which pages Copilot cites in your category and how often your domain is one of them. Once searchQueries ships, one record will hold the prompt, the searches and the cited pages, so you can check whether a domain that is never cited doesn’t rank for the queries Copilot writes. The market changes the answer, so compare markets deliberately; see state-level AI answers.

Field reference

result.sources[]

Copilot sources carry no description. Key on url.

result.citationPills[]

One row per (chip, source) pair, ordered to match the chips in the answer:

A chip that groups three citations therefore produces three rows. The whole array is absent from result when the answer shows no chips at all.

result.searchQueries[]

Coming; not returned yet.

Annotated excerpt

A shortened response. The content is made up; the field names and types are those of the POST /v1/monitor/copilot result schema.

{
  "success": true,
  "result": {
    "text": "For a small team, a password manager with shared vaults and...",
    "sources": [
      {
        "position": 1,
        "url": "https://www.example-security.com/reviews/team-password-managers",
        "label": "Best team password managers"
      },
      {
        "position": 2,
        "url": "https://help.example-vault.com/sharing",
        "label": "Sharing items with your team"
      }
    ],
    "citationPills": [
      { "citationPillId": 1, "position": 1, "url": "https://www.example-security.com/reviews/team-password-managers", "domain": "example-security.com", "label": "Best team password managers" },
      { "citationPillId": 1, "position": 2, "url": "https://help.example-vault.com/sharing", "domain": "help.example-vault.com", "label": "Sharing items with your team" }
    ]
    // searchQueries is coming; not returned yet.
  }
}

Cost and request options

No include flag is needed for sources. A Copilot request costs $0.00274; include.markdown, include.rawResponse and include.rawWire (always null on Copilot) add nothing, and include.html is refused with a 400.

The request takes prompt and country; a request that sends state is refused with a 400. Keep the market fixed within a time series.

Example: citation share by domain

The script runs each prompt several times and computes answer share per domain: the share of answers that cite the domain at least once.

import os
import time
from collections import Counter
from datetime import date
from urllib.parse import urlparse

import requests

API = "https://api.answerline.dev"
HEADERS = {"Authorization": f"Bearer {os.environ['API_KEY']}"}
PROMPTS = ["best password manager for a small team", "1Password vs Bitwarden for business"]
SAMPLES = 5
DAY = date.today().isoformat()
MY_DOMAIN = "example-vault.com"


def run_tasks(tasks):
    resp = requests.post(f"{API}/v1/async/task/batch", headers=HEADERS, json=tasks, timeout=60)
    resp.raise_for_status()
    pending = {item["task"]["id"] for item in resp.json()["results"] if item["success"]}
    results = []
    while pending:
        time.sleep(10)
        for task_id in list(pending):
            body = requests.get(f"{API}/v1/async/task/{task_id}", headers=HEADERS, timeout=30).json()
            status = body["task"]["status"]
            if status in ("COMPLETED", "FAILED"):
                pending.discard(task_id)
                if status == "COMPLETED":
                    results.append(body["response"]["result"])
    return results


def registrable(url):
    parts = (urlparse(url).hostname or "").split(".")
    return ".".join(parts[-2:])


tasks = [
    {
        "taskType": "COPILOT",
        "payload": {"prompt": prompt, "country": "US"},
        "idempotencyKey": f"pw-sources-{p}-s{s}-{DAY}",
    }
    for p, prompt in enumerate(PROMPTS)
    for s in range(SAMPLES)
]
answers = run_tasks(tasks)

answer_hits = Counter()
for result in answers:
    answer_hits.update({registrable(src["url"]) for src in result.get("sources", []) if src.get("url")})

n = len(answers) or 1
print(f"{len(answers)} answers")
for domain, hits in answer_hits.most_common(15):
    print(f"{domain:32} answers {hits / n:5.0%}")
print(f"{MY_DOMAIN} cited in {answer_hits[MY_DOMAIN] / n:.0%} of answers")

registrable here drops everything but the final two host labels. That merges help. and www. into one brand, which is what you want, and mis-groups .co.uk domains, which is not; reach for a public-suffix library if your query set is international.

Pitfalls

sources can be empty. Default it to an empty list. citationPills is absent when the answer shows no chips. searchQueries is not returned yet.

sources[].position is a number. Key on url when merging with other engines’ sources.

Copilot’s answers vary between runs of the same prompt, so measure citation share across many runs (sampling vs census).

Related

• Copilot engine guide: sync and async calls, webhooks, targeting and billing.
• Copilot shopping and map results for the commercial and local modules.
• Monitoring Copilot with an API.
• Compare with ChatGPT sources, ChatGPT search queries and Gemini sources.
• AI visibility tracking and keyword research.

More Copilot data

All Copilot fields · API reference