Bing Search API Alternative: What to Use After the Retirement
Michael Carter · · 17 min read · Last updated

Key takeaways
- Microsoft retired the Bing Search APIs on August 11, 2025 and decommissioned every existing instance; the v7 reference now sits in Microsoft Learn's previous-versions archive.
- Grounding with Bing Search is an agent tool, not a results API: developers never see the raw results, and it costs $14 per 1,000 transactions (Microsoft's pricing page, checked November 17, 2026).
- Microsoft Web IQ returns titles, URLs and snippets for $7 per 1,000 web calls, but its terms allow the content only alongside LLM output and forbid storing or caching it.
- Most Bing v7 web fields map one to one to a Google SERP API (name to title, url to link, snippet to snippet, displayUrl to displayedLink); totalEstimatedMatches and dateLastCrawled have no equivalent.
- In our November 17, 2026 test, a Python adapter of about 40 lines returned Bing-shaped Google results from BestSERP in 3.7 seconds for 1 credit, so the code that reads the results did not change.
The right Bing Search API alternative depends on what your code did with the results. Microsoft retired the Bing Search APIs on August 11, 2025, and points customers to Grounding with Bing Search and now Web IQ, both built to feed AI models. For a plain list of ranked results, use a Google SERP API or Brave's own index.
This guide is for developers who had a Bing v7 call in production: a search box, an SEO tool, a data pipeline or an AI agent. You get the dated facts from Microsoft's own pages, what each replacement returns and costs, a field and parameter mapping, and tested Python and JavaScript code that keeps Bing's response shape. For a wider comparison of providers, see how to choose a SERP API.
What happened to the Bing Search APIs?
Microsoft shut them down. The Microsoft Lifecycle notice, dated May 15, 2025, says: "Bing Search APIs will be retired on August 11, 2025. Any existing instances of Bing Search APIs will be decommissioned completely, and the product will no longer be available to be used or new customer signup."
The same notice told customers to migrate to "Grounding with Bing Search as part of Azure AI Agents." Eighteen months later, Microsoft's direction has moved again:
| Date | What happened | Source |
|---|---|---|
| May 15, 2025 | Retirement notice published | Microsoft Lifecycle |
| August 11, 2025 | Bing Search APIs retired; existing instances decommissioned | Microsoft Lifecycle |
| As of November 17, 2026 | "Grounding with Bing is upgrading to Web IQ"; Web IQ's web, news, image and video APIs are generally available | Microsoft's Grounding with Bing and Web IQ pricing pages |
The v7 reference survives in Microsoft Learn's "previous versions" section, and the mapping below comes from it. The API behind it, api.bing.microsoft.com/v7.0/search, is gone.
What Grounding with Bing Search actually does
Grounding with Bing Search is a tool for AI agents in Microsoft Foundry Agent Service, not a search results API. According to Microsoft's Grounding with Bing Search documentation (page dated October 12, 2026), the agent's model decides when to search, Bing returns results to that model, and the model writes the final answer with citations.
The key sentence for v7 users: "Developers and end users don't have access to raw content returned from Grounding with Bing Search." You get the model's answer, its URL citations and a link to the Bing query, and Microsoft requires you to display both "in the exact form provided by Microsoft."
| Bing Web Search v7 (retired) | Grounding with Bing Search | |
|---|---|---|
| What you call | A REST endpoint (GET /v7.0/search) |
A tool attached to an agent in Foundry Agent Service |
| What comes back | JSON lists of web pages, news, images | The model's answer with URL citations |
| Raw results for your code | Yes | No |
| Price (checked 2026-11-17) | Retired | $14 per 1,000 transactions, counted per tool call |
| Data handling | Not applicable | Queries leave the Azure compliance boundary; Microsoft's Data Protection Addendum does not apply |
The price and limits come from Microsoft's Grounding with Bing page: $14 per 1,000 transactions, up to 150 transactions per second and 1 million a day. The same page says "Outputs are not directly accessible for use in other applications or programs."
So Grounding with Bing Search fits one case: an agent built in Microsoft Foundry that answers from the live web. It cannot fill a search box, track rankings or build a dataset, because your code never sees the results.
What about Microsoft Web IQ?
Web IQ is Microsoft's newer search API for AI applications; the Grounding with Bing page now says "For new projects, explore Web IQ." Unlike Grounding, it returns "a structured JSON response with titles, URLs, snippets, timestamps, and provenance," according to the Web IQ overview (dated November 13, 2026). Web IQ pricing lists its web, news, images and video endpoints as generally available at $7 per 1,000 calls each; it needs a paid Azure subscription.
That makes it the closest official successor to Bing v7 for AI apps, with one large condition: the terms. The Web IQ terms of use let you display the content only "on the same user interface as content from an LLM", cap what you show from any single source at 200 characters, and forbid you to "copy, store, cache, archive, or create a database of Content" beyond narrow exceptions such as a user's chat history, which the Web IQ FAQ (dated November 16, 2026) spells out. A classic results page, a rank tracker or a stored dataset does not fit those terms. This is general information, not legal advice; read the terms for your own use.
Bing Search API alternative options compared
The right Bing Search API alternative is the one whose output matches what your code used: an answer for a model, or a list of ranked results. Here is what each option returns, with prices from the providers' own pages:
| Option | Results come from | What you get back | Price (checked 2026-11-17) |
|---|---|---|---|
| Grounding with Bing Search | Bing | A model's answer with citations, inside Foundry Agent Service | $14 per 1,000 transactions |
| Microsoft Web IQ | Microsoft's web index | Results and passages for an LLM; shown only with LLM output, not stored | $7 per 1,000 web calls |
| Google SERP API (BestSERP and others) | Google's results page | Ranked results as JSON: organic, news, images, videos, places | BestSERP: $0.50 per 1,000 after 3,000 free searches |
| Brave Search API | Brave's own index | Results as JSON plus LLM context; storage rights need a custom plan | $5 per 1,000 requests, $5 free credits a month |
| SERP API with a Bing engine | Bing's results page | Bing's page as JSON | See the provider |
| AI-native search APIs | Varies | Content or answers prepared for a model | Varies |
- Google SERP APIs run a Google search for you and return the results page as data (see what a SERP API is). BestSERP is one of several. It returns organic results, Top stories, video carousels and the news, images, videos, shopping, places and maps tabs, but not ads, People Also Ask, related searches or AI Overviews; details are on the Google Search API page.
- Brave Search API runs on its own index. Brave's API page says "Every result comes from Brave's own independent index, crawled and ranked by Brave," lists the Search plan at $5 per 1,000 requests with 50 queries per second, and answers "Yes, with a custom plan" to whether you can store results.
- SERP APIs with a Bing engine suit teams that need Bing's results specifically. SerpApi, for example, says its Bing Search API "allows you to scrape results from the Bing search page."
- AI-native search APIs such as Tavily and Exa present themselves as search for AI agents on their own sites. If you are building an agent, start with our guide to web search for AI agents and the Tavily alternatives comparison.
How Bing fields map to a Google SERP API
Most of what your code read from webPages.value[] has a direct equivalent in organic[]. The diagram shows the mapping for web, news and image results; the tables below repeat it with notes.

| Bing v7 field | BestSERP field | Note |
|---|---|---|
webPages.value[].name |
organic[].title |
|
webPages.value[].url |
organic[].link |
A Google redirect until links are resolved; domain holds the site |
webPages.value[].snippet |
organic[].snippet |
Often starts with a date, such as "Nov 5, 2026 —" |
webPages.value[].displayUrl |
organic[].displayedLink |
|
Order in value[] |
organic[].position |
Starts at 1 |
dateLastCrawled, datePublished |
None | Parse the date at the start of the snippet |
deepLinks |
None | Sitelinks are not returned |
webPages.totalEstimatedMatches |
None | Page until fewer results come back |
queryContext.originalQuery |
searchParameters.q |
|
news.value[] |
news[] from /news |
name, url, description, datePublished, provider[0].name become title, link, snippet, publishedAt, source |
images.value[] |
images[] from /images |
contentUrl, width, height, hostPageUrl become imageUrl, imageWidth, imageHeight, link |
relatedSearches, spellSuggestions, entities, computation |
None | Not returned by BestSERP |
The request parameters map almost as directly. The Bing names come from the archived Bing Web Search v7 query parameters reference; the BestSERP names from the search endpoint docs:
| Bing v7 | BestSERP | Note |
|---|---|---|
Ocp-Apim-Subscription-Key header |
X-API-KEY header |
Keys in URLs are refused |
q |
q |
1 to 256 characters |
count (default 10, max 50) |
num (10, 20, 30, 40 or 50) |
Each 10 results cost 1 credit |
offset |
page (1 to 5) |
page = offset / num + 1; pages 1 to page are fetched and charged |
mkt=en-US |
gl=us, hl=en |
Country and language are two parameters |
cc, setLang |
gl, hl |
|
freshness=Day, Week, Month |
tbs=qdr:d, qdr:w, qdr:m |
Also qdr:h and qdr:y; date ranges have no equivalent |
responseFilter=News, Images, Videos |
/news, /images, /videos |
One endpoint per vertical |
safeSearch |
None | Remove it: unknown parameters fail with unknown_parameter |
Two limits change the math. BestSERP returns at most 50 web results per query (page × num/10 ≤ 5). And Google sometimes shows 8 or 9 results on a page, so never assume count results come back.
Code: replace a Bing call with Google results
The fastest migration keeps your downstream code untouched: a small adapter takes Bing's arguments and returns Bing's shape. You need a BestSERP key in the BESTSERP_API_KEY environment variable; the quickstart shows how to get one with 3,000 free searches.
Before: your Bing v7 call (retired, not run by us)
This is the kind of call your code made, written from Microsoft's v7 reference. The service is retired, so we could not run it:
"""Bing Web Search v7: retired on August 11, 2025. Shown for comparison; not run by us."""
import os
import requests
resp = requests.get(
"https://api.bing.microsoft.com/v7.0/search",
headers={"Ocp-Apim-Subscription-Key": os.environ["BING_SEARCH_KEY"]},
params={"q": "bing search api alternative", "count": 10, "offset": 0,
"mkt": "en-US", "freshness": "Month", "responseFilter": "Webpages"},
timeout=30,
)
resp.raise_for_status()
for page in resp.json()["webPages"]["value"]:
print(page["name"], page["url"], page["snippet"])
After: Python
The same arguments, now answered by BestSERP. It retries 502 and 503 (both refunded), waits on 429, polls a 202 for long searches and waits up to 2 minutes for final links. Tested on November 17, 2026 with Python 3.12.10 and requests 2.34.2, and re-run on November 18 with the longer link wait:
import os
import time
import requests
API = "https://api.bestserp.com"
HEADERS = {"X-API-KEY": os.environ["BESTSERP_API_KEY"]}
FRESHNESS = {"day": "qdr:d", "week": "qdr:w", "month": "qdr:m"} # Bing freshness -> Google tbs
def call(method, path, body=None, tries=3):
"""One BestSERP request: polls 202, waits on 429/503, retries 502 (failed searches are refunded)."""
for attempt in range(1, tries + 1):
r = requests.request(method, API + path, headers=HEADERS, json=body, timeout=130)
while r.status_code == 202: # long search: poll until it is done (polling is free)
time.sleep(2)
r = requests.get(f"{API}/requests/{r.json()['requestId']}", headers=HEADERS, timeout=130)
if r.status_code not in (429, 502, 503) or attempt == tries:
r.raise_for_status()
return r.json()
time.sleep(int(r.headers.get("Retry-After", 3)))
def bing_search(q, count=10, offset=0, mkt="en-US", freshness=None):
"""Takes Bing v7 parameters, returns {"webPages": {"value": [...]}} like Bing did."""
lang, _, country = mkt.partition("-") # "en-US" -> hl=en, gl=us
num = min(50, -(-count // 10) * 10) # Google pages hold 10 results: round up to 10, 20 ... 50
body = {"q": q, "gl": (country or "us").lower(), "hl": lang.lower(), "num": num, "page": offset // num + 1}
if freshness:
body["tbs"] = FRESHNESS[freshness.lower()]
data = call("POST", "/search", body)
deadline = time.monotonic() + 120 # final links arrive by polling: usually within 30 s, up to 2 minutes
while data.get("linksPending") and time.monotonic() < deadline:
time.sleep(4)
data = call("GET", f"/requests/{data['requestId']}")
pages = [{"name": r["title"], "url": r["link"], "snippet": r.get("snippet", ""),
"displayUrl": r.get("displayedLink", ""), "domain": r.get("domain", ""), "position": r["position"]}
for r in data.get("organic", [])]
return {"webPages": {"value": pages[:count]}, "queryContext": {"originalQuery": q},
"bestserp": {k: data.get(k) for k in ("requestId", "credits", "cached", "unresolvedLinks")}}
if __name__ == "__main__":
result = bing_search("bing search api alternative", count=10, mkt="en-US", freshness="Month")
for page in result["webPages"]["value"]: # the loop your Bing code already had
print(page["position"], page["domain"] or "-", "|", page["name"][:70])
print(result["bestserp"])
Save it as bing_to_bestserp.py and run it. Our output on November 17, 2026 (Google results from the past month, US, English):
1 faun.dev | Bing Search API vs wigolo
2 apify.com | Bing Web Search Results API in Python [DEPRECATED]
3 faun.dev | Bing Search API vs Scale SERP
4 searxncrawl.io | Bing and Google Search API Alternative, Self-Hosted
5 sourceforge.net | Bing Custom Search vs. SearchApi Comparison
6 nestdaddy.com | Bing Search API is retired: how to get raw web and news ...
7 sourceforge.net | Bing News Search API vs. Serpdog Comparison
8 faun.dev | Top 5 Proprietary licensed Bing Maps for Enterprise ...
9 sourceforge.net | Bing Custom Search vs. SerpKite Comparison
10 bestwebsearchapi.com | Migrating Off the Bing Search API - Hyperresearch Reference
{'requestId': 'b17153c3-d65a-4c42-85ba-aa79b8fffa81', 'credits': 1, 'cached': False, 'unresolvedLinks': 10}
The search took 3.7 seconds and cost 1 credit. Note unresolvedLinks: 10: every link was still a Google redirect after polling. That was a link-resolution bug, fixed on November 18, 2026: in 61 searches through the public API that day, polling turned all 401 pending redirects into final URLs, median 34 seconds and at most 54 seconds after the request, so the adapter now waits up to 2 minutes (see link resolution). Re-run on November 18, 2026 with the longer wait: 10 of 10 links were final URLs after 33 s. The adapter also passes domain through, for code that matches results by site.
After: JavaScript
The same adapter for Node.js 18 or later, using the built-in fetch. Tested on November 17, 2026 with Node.js 22.17.1 and re-run on November 18 with the longer link wait, this time with mkt: "en-GB" (sent as gl=gb, hl=en):
const API = "https://api.bestserp.com";
const HEADERS = { "X-API-KEY": process.env.BESTSERP_API_KEY, "Content-Type": "application/json" };
const FRESHNESS = { day: "qdr:d", week: "qdr:w", month: "qdr:m" }; // Bing freshness -> Google tbs
const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
// One BestSERP request: polls 202, waits on 429/503, retries 502 (failed searches are refunded).
async function call(method, path, body, tries = 3) {
for (let attempt = 1; ; attempt++) {
let res = await fetch(API + path, { method, headers: HEADERS, body: body && JSON.stringify(body) });
while (res.status === 202) { // long search: poll until it is done (polling is free)
const { requestId } = await res.json();
await sleep(2000);
res = await fetch(`${API}/requests/${requestId}`, { headers: HEADERS });
}
if (![429, 502, 503].includes(res.status) || attempt === tries) {
if (!res.ok) throw new Error(`BestSERP ${res.status}: ${await res.text()}`);
return res.json();
}
await sleep(1000 * Number(res.headers.get("Retry-After") ?? 3));
}
}
// Takes Bing v7 parameters, returns { webPages: { value: [...] } } like Bing did.
export async function bingSearch(q, { count = 10, offset = 0, mkt = "en-US", freshness } = {}) {
const [lang, country = "us"] = mkt.toLowerCase().split("-"); // "en-GB" -> hl=en, gl=gb
const num = Math.min(50, Math.ceil(count / 10) * 10); // Google pages hold 10 results
const body = { q, gl: country, hl: lang, num, page: Math.floor(offset / num) + 1 };
if (freshness) body.tbs = FRESHNESS[freshness.toLowerCase()];
let data = await call("POST", "/search", body);
const deadline = Date.now() + 120_000; // final links arrive by polling: usually within 30 s, up to 2 minutes
while (data.linksPending && Date.now() < deadline) {
await sleep(4000);
data = await call("GET", `/requests/${data.requestId}`);
}
const value = (data.organic ?? []).slice(0, count).map((r) => ({
name: r.title, url: r.link, snippet: r.snippet ?? "", displayUrl: r.displayedLink ?? "",
domain: r.domain ?? "", position: r.position,
}));
const { requestId, credits, cached, unresolvedLinks } = data;
return { webPages: { value }, queryContext: { originalQuery: q }, bestserp: { requestId, credits, cached, unresolvedLinks } };
}
const result = await bingSearch("grounding with bing search", { count: 10, mkt: "en-GB" });
for (const page of result.webPages.value) console.log(page.position, page.domain || "-", "|", page.name.slice(0, 70));
console.log(result.bestserp);
Save it as bingToBestserp.mjs and run node bingToBestserp.mjs. Our output on November 17, 2026:
1 learn.microsoft.com | Unable to use grounding with bing search with openai (ai ...
2 learn.microsoft.com | BingGroundingSearchConfigurat...
3 learn.microsoft.com | Inquiry on Using "Grounding with Bing Search" Without ...
4 - | Introducing "Grounding with Bing Search" in Azure AI ...
5 bitbasti.com | Azure and the Bing API Shutdown - No Notice, No API - bitbasti
6 learn.microsoft.com | BingGroundingSearchToolPara...
7 github.com | Grounding with Bing search - Run failed: {' ...
8 - | Using APIM and Azure AI Foundry + Grounding Bing Search
9 ppc.land | Grounding with Bing Search enhances Azure AI Agent ...
{
requestId: 'bf16feaa-00f6-4c96-8c10-c122cf24f95a',
credits: 1,
cached: false,
unresolvedLinks: 9
}
Google returned 9 results for count: 10, and two came back without a domain, so treat every field except position and title as optional. Re-run on November 18, 2026 with the longer wait: 9 of 9 links were final URLs after 32 s.
News: /v7.0/news/search becomes /news
Each Bing vertical becomes its own endpoint. This function reuses call and FRESHNESS from the Python adapter and returns Bing's NewsArticle shape (news endpoint docs):
from bing_to_bestserp import FRESHNESS, call
def bing_news(q, mkt="en-US", freshness=None):
lang, _, country = mkt.partition("-")
body = {"q": q, "gl": (country or "us").lower(), "hl": lang.lower()}
if freshness:
body["tbs"] = FRESHNESS[freshness.lower()]
data = call("POST", "/news", body)
return {"value": [{"name": a["title"], "url": a["link"], "description": a.get("snippet", ""),
"datePublished": a.get("publishedAt", a.get("date", "")),
"provider": [{"name": a.get("source", "")}]}
for a in data.get("news", [])]}
if __name__ == "__main__":
for article in bing_news("Microsoft Ignite", freshness="Week")["value"][:5]:
print(article["datePublished"], "|", article["provider"][0]["name"], "|", article["name"][:60])
The first five of the 10 articles it returned on November 17, 2026:
2026-11-17T03:20:23+00:00 | Microsoft Source | Microsoft Ignite 2026 | Live News
2026-11-15T14:03:01+00:00 | Yahoo Finance | Microsoft Ignite Starts Nov. 17. Will Microsoft Update Its P
2026-11-12T14:01:42+00:00 | The New Stack | See what AI looks like in production at Microsoft Ignite
2026-11-10T23:16:48+00:00 | TechTarget | Microsoft Ignite 2026 conference coverage
2026-11-14T13:00:09+00:00 | PCMag | Microsoft Ignite Is Next Week. Here's What It Needs to Prove
Images work the same way with /images.
Choosing by use case: apps, AI agents, SEO tools
Pick a Bing Search API alternative by what your product shows the user and whether it keeps the data:
| Use case | Best fit | Why |
|---|---|---|
| Search box or results list in an app | Google SERP API or Brave Search API | You need the list itself; Grounding returns no list, and Web IQ content may only appear with LLM output |
| AI agent built in Microsoft Foundry | Grounding with Bing Search or Web IQ | Native tools, if you accept the display rules and that data leaves the Azure compliance boundary |
| AI agent on another stack | Google SERP API, Brave, Web IQ or an AI-native API | Compare on answer quality and terms; see the AI agents use case |
| Rank tracking and SEO tools | Google SERP API | You track Google's own results page and keep the history |
| Bing rankings specifically | SERP API with a Bing engine | Returns Bing's results page; Microsoft's FAQ says Web IQ results can differ from Bing.com |
If you are also moving off Google's Custom Search JSON API, which existing customers must leave by January 1, 2027, the same adapter idea works there; see the Custom Search JSON API migration guide.
Common migration problems
| Problem | Cause | Fix |
|---|---|---|
400 unknown_parameter |
Bing parameters (mkt, count, safeSearch) sent as is |
Map them with the table above and drop the rest |
Fewer results than count |
Google showed 8 or 9 results on the page | Loop over what came back; never index a fixed length |
Links start with google.com/goto |
Read before link resolution finished (linksPending: true) |
Poll until linksPending is false, for up to 2 minutes, as the adapters do; unresolvedLinks counts any left |
400 for deep pages |
More than 50 results requested (page × num/10 > 5) | Stop at 50 results per query |
429 rate_limited |
More than 60 searches a minute (the default) | Wait for Retry-After; the adapters already do |
502 search_failed |
The search was blocked or timed out | Retry once or twice; the credit is refunded |
How we checked
On November 17, 2026 we read Microsoft's Lifecycle notice, the Grounding with Bing Search docs and pricing page, the Web IQ overview, pricing, FAQ and terms, and the archived Bing v7 Web, News and Image Search references. Competitor facts come only from the providers' own pages, read the same day.
We did not run Bing v7 (retired), Grounding with Bing Search or Web IQ; their descriptions quote Microsoft's documentation. We ran the BestSERP code above against the live API: one Python search (US), one JavaScript search (UK) and two news searches (the first, for a narrower query, returned one article and is not shown). On November 18, 2026 we re-ran both web adapters once with the 2-minute link wait.
Next steps
Whichever Bing Search API alternative you choose, start with an adapter, keep your downstream code, and test the parameters you actually use before you switch traffic. Useful next reads:
- Choosing a provider: the best SERP API options compared.
- Building an agent: web search for AI agents.
- Writing code today: the quickstart and the pricing page, where BestSERP costs $0.50 per 1,000 searches after 3,000 free ones.