Bing Proxies

Essentially, yes, if you need raw results. Microsoft retired the Bing Search APIs on August 11, 2025 — existing instances were decommissioned, not grandfathered — and the official successor, Grounding with Bing Search, returns AI-mediated answers inside the Azure AI Agents framework rather than the structured SERP JSON pipelines were built on. That leaves two realistic paths: third-party SERP APIs (which do the Bing SERP scraping for you and charge per request) or your own collection through proxies, where the cost is a flat package price regardless of query count. The math favors doing it yourself earlier than people expect — Bing's tolerance for straightforward HTTP clients means the engineering overhead that justifies a managed API for other engines mostly isn't there.

Two parameters and one matching IP. mkt sets the market (de-DE, en-GB, fr-FR…), setlang pins the interface language, and both should be sent explicitly on every request — without them Bing guesses from the IP and cookies, which makes results drift between sessions and ruins comparability. The exit address should then agree with the declared market: a de-DE query from a Dutch IP mostly works, but ads, shopping results, and some local blocks key on the request's physical origin, so full-fidelity collection uses a German address for the German market. One more Bing-specific tip: pagination is an offset, not a page number — first=1, first=11, first=21 — and requesting deep offsets in rapid sequence is one of the few patterns that reliably draws its attention.

Most likely you took “tolerant” as “unlimited” — that's the standard failure mode on this engine. Bing's velocity scoring is lenient but real: a single IP pushing hundreds of queries an hour, no delays between requests, or systematic deep-pagination pulls will earn blocks that look sudden but were accumulating. The recovery is straightforward: drop the per-IP rate to a few queries a minute with jittered spacing, spread volume across more of the package, and let flagged addresses cool off — unlike stricter engines, Bing's flags tend to expire rather than persist, so a rate-limited IP usually returns to service within hours. If blocks continue at modest rates, check the boring things: a stale or missing User-Agent, requests without an Accept-Language header, and cookie handling disabled — Bing's checks are basic, but they do exist.

Control and cost, mainly. AI-mediated search services return synthesized answers — useful for chat, but wrong for pipelines that need to rank, filter, deduplicate, or archive the underlying results; once a model has summarized the SERP, the original ordering and coverage are gone. Scraping gives you the raw ranked list to process on your terms, and Bing is the pragmatic engine for it: no CAPTCHA farm, plain-HTML responses, and datacenter-friendly economics that survive the request volumes RAG backends generate. The honest counterpoint: if you only need a handful of grounding queries per user session and never touch the raw ranking, a managed AI search API is less code to own. The crossover comes with volume — flat-price proxy packages don't bill per request, and retrieval-heavy workloads hit that crossover fast.