Technical Guides
How to scrape Microsoft Copilot: WebSocket events + auth sessions
Scrape Microsoft Copilot in 2026: intercept the WebSocket protocol, manage Microsoft account session cookies, and extract source URLs from event metadata.
Get the real Copilot UI responses: Bing-grounded sources, inline shopping cards, and Bing maps with place data. All the data Azure OpenAI never returns. Markdown out, any country, any scale.
4.7 on G2
No credit card required.
curl -X POST https://api.cloro.dev/v1/monitor/copilot \
-H "Authorization: Bearer sk_live_your_api_key_here" \
-H "Content-Type: application/json" \
-d '{
"prompt": "What do you know about Tesla\'s latest updates?",
"country": "US",
"include": {
"markdown": true
}
}' {
"success": true,
"result": {
"text": "...",
"sources": [],
"html": "...",
"markdown": "..."
}
} cloro extracts the Bing index that Copilot cites. The same key gets you ChatGPT, Perplexity, Gemini, AI Overview, and AI Mode.
Copilot is the only major AI search surface that grounds against Bing, not Google. Your competitive landscape on Copilot is different, and the Microsoft Graph / Azure OpenAI APIs return none of the surface that drives it.
Microsoft hardens Copilot's anti-automation to punish headless-browser patterns within hours. DIY pipelines hit success-rate cliffs that look fine in a small test and collapse at production volume. cloro absorbs the access fight so your Copilot monitoring keeps running across Bing's update cycle.
Copilot re-ranks its Bing-cited sources per session and per country. The same prompt can return a different top-source on every run, and Bing's regional indexes diverge meaningfully. A single API call tells you nothing; you need repeated sampling to see the real distribution. cloro routes per `country` so you can capture the regional divergence directly.
Azure OpenAI returns ungrounded model output. The Bing Search API returns SERP rows. The Microsoft Graph API exposes M365 tenant data. None of them returns the Bing-cited Copilot answer a real user reads on Copilot.com. Surfer's analysis measured ~20% overlap between API responses and the rendered UI across LLMs.
Copilot embeds inline shopping cards (price, ratings, seller) and Bing maps place entries (with reviews, photos, coordinates) directly inside the answer, alongside the Bing-grounded source citations. None of this lives in any Microsoft API. cloro returns `shoppingCards`, `map`, and `sources` as structured JSON in one response.
Parse markdown, Bing-grounded sources, shopping cards, and Bing maps from one endpoint.
import requests
response = requests.post(
"https://api.cloro.dev/v1/monitor/copilot",
headers={
"Authorization": "Bearer sk_live_your_api_key_here",
"Content-Type": "application/json"
},
json={
"prompt": "What do you know about Tesla's latest updates?",
"country": "US",
"include": {
"markdown": true
}
}
)
print(response.json()){
"success": true,
"result": {
"text": "Based on my analysis of Tesla's...",
"html": "https://storage.cloro.dev/results/copilot-result-page.html",
"markdown": "### Tesla Recent Analysis\n\nBased on my analysis of Tesla's recent developments...",
"sources": [
{
"position": 1,
"url": "https://tesla.com/blog/fsd-beta-v12",
"label": "Tesla FSD Beta v12 Update",
"description": "Latest improvements in Tesla's..."
}
],
"shopping_cards": [
{
"position": 1,
"product": {
"name": "Model Y",
"brand": "Tesla",
"price": "$43,990",
"currency": "USD",
"rating": 4.5,
"reviewCount": 2847,
"imageUrl": "https://example.com/tesla-model-y.jpg",
"productUrl": "https://tesla.com/modely",
"description": "All-electric compact SUV..."
}
}
]
}
} Pick a plan that fits your volume. Price per credit drops as you scale.
Credit cost per request varies by provider. The rates below apply to async/batch requests; sync requests add a +2 credit surcharge.
Google News uses the same pricing as Google Search.
Bing Search returns raw SERP data (the URLs, not the Copilot answer). Azure OpenAI returns model output without Bing grounding. Neither shows you what a user reading copilot.microsoft.com actually sees. cloro extracts the rendered Copilot UI, including the Bing-cited sources panel that drives brand visibility.
Bing's index refreshes on a different cadence than Google's: roughly every 24–48 hours for high-traffic queries, longer for the long tail. Copilot's citations track that. Daily cloro sampling on the same prompt typically shows 1–3 source changes per week; more for trending topics. Every cloro request hits Copilot live so you catch each shift.
Bing's index weighs different signals than Google: different domain authority, different freshness curves, different sub-domain handling. Sites that rank well on Google can be invisible on Copilot and vice versa. cloro lets you measure the Copilot-specific picture without conflating it with Google data.
More than most teams expect. Bing maintains separate regional indexes (Bing US, Bing UK, Bing DE, Bing JP), each with its own ranking signals and content freshness. From our sampling, Copilot's per-country citation overlap on the same prompt averages around 30%. If you're a global brand, you need country-level Copilot tracking to know what each market actually sees.
cloro targets the public copilot.microsoft.com consumer surface, the one that matters for organic brand visibility. M365 Copilot lives inside enterprise tenants with different behavior and access controls; it's out of scope.
Yes. The `result.map` array contains Bing maps place entries (each with reviews, photos, coordinates, and contact info) when the prompt has local intent. Combined with `shoppingCards` for shopping intent, you get the full set of Copilot rich-result objects, none of which live in Azure OpenAI or the Bing Search API.
Bing's per-country indexes × the prompts you care about = typically 6–15 permutations per tracked query. Sync polling against Microsoft's anti-automation at that volume needs a job queue, retries, and a residential-proxy budget. Realistic in-house cost: $4–7k/month before engineering time. cloro's Hobby plan ($100/month) absorbs the whole permutation matrix.
Scrape Microsoft Copilot in 2026: intercept the WebSocket protocol, manage Microsoft account session cookies, and extract source URLs from event metadata.
We tested 12 LLM visibility tracking tools on real brand-monitoring workflows across ChatGPT, Perplexity, Gemini, and Google AI Overview. What works, what doesn't.
From Perplexity to ChatGPT Search, AI search engines are replacing traditional keywords with conversational answers. Here is everything you need to know about the shift to answer-first discovery.