Bing Copilot API: Programmatic Access Guide (2026)

Microsoft does not offer an official public API for Bing Copilot answers as of May 2026. To access Copilot's AI-generated responses programmatically, you need a third-party scraper or wrapper API — the two production-ready options are SerpAPI's Bing AI endpoint and Apify's Bing AI Search (Copilot) Answers actor, both of which return the generative summary, citations, and source URLs as structured JSON.

Quick Answer

Bing Copilot API programmatic access requires a third-party tool because Microsoft has not exposed Copilot Search behind a documented developer endpoint. You send a query through a scraper that loads Bing, triggers the Copilot answer, and parses the AI response plus inline citations into JSON. The two working paths in 2026 are SerpAPI (subscription, $75+/month) and the Bing AI Search (Copilot) Answers actor on Apify (pay-per-use, typically $0.50–$2 per 1,000 queries). Both return the same core fields: the answer text, citation URLs, and the underlying SERP for fallback.

Is there an official Bing Copilot API?

No. Microsoft's official Bing Search APIs were retired in August 2025, and Copilot Search — announced at Build in April 2025 — was never released with a public developer endpoint. Microsoft directs enterprise customers to Azure OpenAI for generative AI, but that doesn't include Bing's web grounding or the exact Copilot answer surface you see at bing.com.

What this means practically:

• You cannot get a free Microsoft API key for Copilot answers.
• The Azure AI Agents "Grounding with Bing Search" tool exists, but it costs $35 per 1,000 transactions and only works inside Azure AI Foundry agents — not as a standalone REST endpoint you can call from a Python script.
• Every other path involves automated browsing or a third-party wrapper.

How do I scrape Bing Copilot answers myself?

You can technically script it, but Copilot answers are rendered by a streaming JavaScript front-end with anti-bot detection. A from-scratch scraper needs to handle:

1. JS rendering — Copilot answers don't appear in the static HTML; they stream in via websocket after the page settles.
2. Session cookies — Bing requires a valid _EDGE_S and MUID cookie pair, often with a region/market hint.
3. Anti-bot fingerprinting — Microsoft uses TLS fingerprint checks and IP reputation scoring. Datacenter IPs get blocked within ~50 requests.
4. Citation extraction — Citations are numbered [1], [2] inline; you have to map them to the source URLs in a sidebar element.

A working DIY stack costs roughly: residential proxies ($300/month for 50 GB), a headless browser cluster (Playwright on 4 CPUs ≈ $80/month), and 2–3 days of engineering. That breaks even against a managed API at around 80,000 queries per month.

How to use the Apify Bing Copilot actor

The fastest path is the Bing AI Search (Copilot) Answers actor. It handles rendering, proxies, and parsing — you POST a query and get JSON back.

Step 1: Get an Apify token

Sign up at apify.com, open Settings → Integrations, copy your API token.

Step 2: Call the actor

curl -X POST "https://api.apify.com/v2/acts/YOUR_USERNAME~bing-ai-copilot/run-sync-get-dataset-items?token=APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "queries": ["best vector database for RAG in 2026"],
    "market": "en-US",
    "maxResults": 1
  }'

Step 3: Parse the response

[
  {
    "query": "best vector database for RAG in 2026",
    "copilotAnswer": "For RAG workloads in 2026, Pinecone, Weaviate, and Qdrant lead...",
    "citations": [
      {"index": 1, "url": "https://example.com/pinecone-review", "title": "Pinecone Review"},
      {"index": 2, "url": "https://example.com/weaviate-bench", "title": "Weaviate Benchmark"}
    ],
    "relatedQuestions": ["Is Pinecone better than Weaviate?"],
    "sources": 7
  }
]

Step 4: Wire it into your app

In Python:

import requests

def get_copilot_answer(query):
    url = "https://api.apify.com/v2/acts/YOUR_USERNAME~bing-ai-copilot/run-sync-get-dataset-items"
    r = requests.post(
        url,
        params={"token": "APIFY_TOKEN"},
        json={"queries": [query], "market": "en-US"}
    )
    return r.json()[0]

answer = get_copilot_answer("LangGraph vs CrewAI for agents")
print(answer["copilotAnswer"])
for c in answer["citations"]:
    print(f"[{c['index']}] {c['url']}")

Runtime is typically 6–12 seconds per query. For batch jobs, pass an array of up to 100 queries in a single run to amortize startup cost.

What's the difference between SerpAPI and the Apify Copilot actor?

Both return AI answers and citations, but the economics and integration model differ.

If you're running 50,000+ queries every month, SerpAPI's flat rate wins. For under 20,000 queries or unpredictable usage (research jobs, content pipelines, monitoring), the pay-per-use model is 5–10x cheaper.

How do I extract citations and source URLs?

Citations are the most valuable part of Copilot output — they're what makes Copilot useful for grounded answers versus pure LLM hallucinations. In the actor response, each citation includes:

• index — the inline reference number (e.g., [1])
• url — the source URL
• title — page title as Bing shows it
• domain — extracted root domain for filtering

Practical patterns:

Bias detection — count citations per domain across 100 queries. If microsoft.com shows up in 70% of tech queries, you have measurable platform bias.

Citation diffing — run the same query weekly and diff the citation set. Sources entering/leaving the answer reveal how Copilot's ranking shifts.

SEO tracking — for a target query, check whether your domain appears in citations. Copilot citations now drive an estimated 8–14% of zero-click traffic for AI-heavy queries (per Similarweb's Q1 2026 report).

Can I use Bing Copilot API for SEO or AEO tracking?

Yes — this is the highest-ROI use case right now. AEO (Answer Engine Optimization) measurement requires knowing which sources AI search engines cite for your target keywords. Manual checking doesn't scale past ~50 queries.

A working tracking pipeline:

1. Maintain a list of 200–500 target queries in a database.
2. Run them through the Copilot actor weekly (cost: ~$1–$4 per run for 500 queries).
3. Store (query, date, citations[]) rows.
4. Dashboard the citation share for your domain vs. competitors.

This is what the new generation of AEO tools (Profound, Peec AI, Otterly) charges $200–$2,000/month for. Building it on a pay-per-use actor costs cents per scan.

What rate limits should I expect?

The Apify actor itself has no enforced query limit — you pay for what you use. Practical limits come from:

• Concurrency — by default, Apify free accounts run 1 actor concurrently; paid plans go up to 32+.
• Bing-side throttling — sustained traffic above ~10 queries/second from the actor's proxy pool can trigger CAPTCHA fallbacks, which the actor handles but slows runs.
• Cost ceilings — set a max compute units cap on your Apify account to avoid runaway bills.

For most teams, 1,000–10,000 queries per day is comfortably in the smooth-running zone.

FAQ

Q: Does Microsoft offer a free Bing Copilot API? No. Microsoft retired the Bing Search APIs in August 2025 and has not released a public Copilot endpoint. The closest official option is Azure AI Agents' Grounding with Bing Search, which costs $35 per 1,000 calls and only works inside Azure AI Foundry.

Q: Is scraping Bing Copilot legal? Scraping publicly visible search results is generally permitted under U.S. case law (hiQ v. LinkedIn, 2022 ruling) when you don't bypass authentication. Bing's terms prohibit automated access, so most teams use third-party services that handle the risk and proxy infrastructure on your behalf.

Q: How accurate are scraped Copilot answers vs. the live website? Modern actors like the Apify Copilot scraper return the same answer text and citations a real user would see, with one caveat: Copilot personalizes results based on signed-in history. Anonymous scraper runs reflect the unpersonalized baseline, which is actually what you want for SEO/AEO benchmarking.

Q: Can I get streaming/real-time Copilot responses via API? Not currently. Both SerpAPI and Apify return the fully-rendered answer after Copilot finishes streaming on the back-end, with 4–12 second latency. True token streaming would require a direct WebSocket connection to Bing, which no third-party provider exposes today.

Q: How much does it cost to track 1,000 queries weekly on Bing Copilot? With the Bing AI Search (Copilot) Answers actor at roughly $1–$2 per 1,000 queries, weekly tracking of 1,000 queries runs $4–$8/month. The same volume on SerpAPI requires the $75/month plan. For monitoring under 20,000 queries/month, pay-per-use is consistently the cheaper option.