--- title: "Google search APIs in 2026: the last official option dies January 1" category: "field-notes" url: https://keirolabs.cloud/blog/google-search-apis-2026 --- ## The 60-second version The last official way to buy Google results is dying. Google closed the Custom Search JSON API to new customers, and existing customers have until **January 1, 2027** to move off it ([Google's own overview page](https://developers.google.com/custom-search/v1/overview) says exactly that). After that date it returns errors, not results. What's left is a market with two halves, and you should pick deliberately: - **Scrapers**: SerpApi, Serper, DataForSEO, Zenserp, Bright Data. They fetch Google the way a browser would, parse the HTML, and sell you JSON. Cheap, fast, and legally uncomfortable. - **Own-index APIs**: Brave and Keirolabs. They crawl the web themselves and sell you their own index. Slightly different results, and nobody can sue them off the planet. One disclosure up front, stated once: Keirolabs is ours, and it's the own-index option below at $0.25 per 1,000 searches, ranked #1 on AIMultiple's 2026 agentic benchmark and first on FinanceBench at 78%. That doesn't make scraping Google wrong in every case, and I'll be specific about when it's the right call. Everything numeric below was checked against provider pricing pages and primary sources in September 2026. Prices move; check before you sign anything. ## The official options, honestly assessed **Custom Search JSON API (legacy).** This was the deal for twenty years: one HTTPS call, Google results as JSON, 100 queries/day free, then $5 per 1,000 queries capped at 10,000/day. It returns at most 100 results per query. It was never a full-web API; it searched a Programmable Search Engine. But at some point Google quietly tolerated "search the entire web" mode and everyone built on it. That era ends January 1, 2027. New signups are already locked out, and Google has also been stripping the free full-web option out of Programmable Search Engine itself. **Vertex AI Search.** The product Google points migrating users to. Read the docs before you believe the marketing: Vertex AI Search is an enterprise semantic-search product over *your own corpus*. Google's [configurable pricing docs](https://cloud.google.com/generative-ai-app-builder/docs/enable-configurable-pricing) set minimum subscription thresholds around 1,000 queries per minute and 50 GiB of storage, with per-query fees that land roughly between $1.50 and $4 per 1,000 queries depending on edition. It does not return plain public-web Google results as JSON. If your app needs "query in, ranked web results out," Vertex is not a drop-in; it's a different product with a different shape and a monthly floor. So the official path is closed, and the official replacement is an enterprise build-your-own-index product. That's not an accident. It's Google exiting a business. ## The scrapers that inherited the market When Bing retired its Search APIs in August 2025 ([Microsoft's lifecycle notice](https://learn.microsoft.com/en-us/lifecycle/announcements/bing-search-api-retirement)) and Google followed by closing Custom Search, the entire "official search API" category evaporated in twelve months. What filled the gap was infrastructure that fetches Google pages at scale (rotating proxies, real browser fingerprints, CAPTCHA solving) and sells the parsed output. | Provider | What it actually is | Index | Why people use it | |---|---|---|---| | SerpApi | The premium scraper. Broadest vertical coverage (Maps, News, Shopping, Jobs, AI Overviews) | Google + others | Deep features, SDKs, 24/7 support | | Serper | The speed-and-price scraper. Credits, no subscription | Google | 1–2s responses, cheap at volume | | DataForSEO | The SEO-industry data platform. Queue-based or live | Google + 20+ engines | Cheapest per-SERP at scale | | Zenserp | Mid-size scraper with a small free plan | Google + a few | Simple, monthly plans | | Bright Data | The proxy giant's SERP product. Pay per successful request | Google + Bing + more | Scale, compliance story, no concurrency limits | All five share one structural fact: their entire product is Google's page, fetched without Google's blessing. Keep that in mind for the next two sections. ## What it costs Here's the pricing table I wish existed when I started benchmarking this market. Effective cost is what you actually pay per 1,000 searches at each provider's most common working volume. | Provider | Entry cost | Effective price per 1,000 | Free tier | Source | |---|---|---|---|---| | Keirolabs `/search/lite` (own index) | Free tier, one-time packs from $10, plans from $30/mo | **$0.25** headline, $0.24 down to $0.08 on monthly plans | 1,250 credits/mo = 12,500 lite searches | [keirolabs.cloud/pricing](https://keirolabs.cloud/pricing) | | Serper | $50 credit pack | **$1.00** entry, drops to $0.30 at volume | 2,500 credits, one-time | [serper.dev](https://serper.dev/) | | DataForSEO (Standard queue) | $50 min deposit, no subscription | **$0.60** (queued) / $2.00 live | None, but pay-per-SERP | [dataforseo.com](https://dataforseo.com/apis/serp-api) | | Bright Data SERP | Pay-as-you-go, $0 minimum | **~$0.75–$1.50** (Google at the high end) | Effectively none | [brightdata.com](https://brightdata.com/pricing/serp) | | Parallel | Usage-based, no subscription | **$1.00–$5.00** (by endpoint) | None | [parallel.ai](https://parallel.ai/) | | Google Custom Search (legacy) | Free tier then pay-as-you-go | **$5.00**, dies Jan 1, 2027 | 100 queries/day | [developers.google.com](https://developers.google.com/custom-search/v1/overview) | | Brave Search API (own index) | Metered, no subscription | **$5.00** | $5 in credits every month | [brave.com/search/api](https://brave.com/search/api/) | | Zenserp | $29/mo Small plan | **$5.80** (5,000 searches) | 50 searches/month | [zenserp.com](https://zenserp.com/) | | Exa | Metered credits, pay-as-you-go | **$7.00** | $10 in free credits | [exa.ai](https://exa.ai/pricing) | | Tavily | Usage-based credits | **$8.00** | 1,000 credits/mo | [tavily.com](https://www.tavily.com/#pricing) | | SerpApi | $25/mo Starter (1,000); $75/mo Developer (5,000) | **$15** at the $75 Developer plan | 100 searches/month | [serpapi.com/pricing](https://serpapi.com/pricing) | Same picture, drawn: USD per 1,000 searches · Sep 2026 Keirolabs /search/lite $0.25 · $0.08 on plans Serper (volume) $0.30 DataForSEO Standard $0.60 Bright Data SERP $0.75–$1.50 DataForSEO Live $2.00 Parallel $1.00–$5.00 Brave Search API $5.00 Google CSE (legacy) $5.00 Zenserp (Small) $5.80 Exa $7.00 Tavily $8.00 SerpApi Developer $15.00 Two honest footnotes on that chart. First, DataForSEO's Standard queue is the cheapest SERP on the market, but results arrive queued, not live: fine for rank tracking, wrong for agents. Second, SerpApi's $15/1k buys you the deepest feature set in the market: every vertical, every engine, a support team that answers. If your product lives on SERP features nobody else parses, that price can be right. (One parenthetical on the cheapest bar: it plots Keirolabs' `/search/lite`; the heavier `/search/fast` endpoint is $2.40 per 1,000 on Essential and $0.80 per 1,000 on Startup.) ## The risk you're pricing in Now the part the scraper pricing pages don't put above the fold. On December 19, 2025, Google sued SerpApi; its own blog post announcing the suit [is titled "We filed a suit today against the scraping company SerpApi"](https://blog.google/innovation-and-ai/technology/safety-security/serpapi-lawsuit/). The allegations: circumventing Google's SearchGuard bot-blocking with fake browsers and bot networks, at a scale Google describes as hundreds of millions of automated queries a day, with DMCA claims layered on top ([The Verge's coverage](https://www.theverge.com/news/848365/google-scraper-lawsuit-serpapi), [PPC Land](https://ppc.land/texas-api-firm-strikes-back-after-googles-dmca-web-scraping-lawsuit/)). > When the entire market's product is one company's page, the market's supply chain risk is one company's mood. However the case ends up, the operational lesson is already priced in by anyone who lived through 2025: Google shipped SearchGuard in January 2025 and broke nearly every scraper's night. Microsoft killed Bing's APIs with six months' notice. Google closed Custom Search with twelve. If your product's search layer is a lease on someone else's anti-bot arms race, your uptime is their litigation calendar. This is the risk ranking I'd give, for a production system: 1. **Own-index APIs** (Brave, Keirolabs). No Google dependency at all. Supply risk is your vendor's crawling budget, which they control and you can monitor. 2. **Scrapers with queue/aggregate models** (DataForSEO Standard). Cheap, and the queue absorbs breakage, but you're still renting Google. 3. **Live SERP scrapers**. Cheapest per result, and the most exposed to the next SearchGuard, the next lawsuit, the next 25,000% usage spike that invites one. ## When you actually need Google's ranking There are also real cases where scraping Google, or paying someone who does, is the correct engineering decision: - **Rank tracking and SEO tooling.** If your customers pay you to know where *Google* puts them, an independent index answers a different question. You need the Google answer, full stop. - **Local and Maps data.** Google's local pack is the product. No independent index replicates it. - **AI Overviews and People-Also-Ask.** These features only exist on Google's page. If you're building monitoring around them, you need a scraper that parses them; SerpApi and Bright Data do, own-index APIs don't. - **Ad verification.** Checking that an ad serves correctly in Google, in a given geo, is definitionally scraping-adjacent work. If your use case is on that list, pick a scraper with a track record and budget for breakage. What you should not do is put a live Google scraper inside an AI agent's hot path and call it reliable infrastructure. Agents hammer search thousands of times a day for cheap queries; that workload belongs on an index you can meter. ## Own your index instead: Brave and Keirolabs The alternative to renting Google is buying from someone who crawled the web themselves. **Brave Search API** runs on Brave's own index. Since February 2026 it's fully metered (Brave removed the old free plan) at **$5 per 1,000 requests** with $5 in free credits landing every month ([Brave's announcement](https://brave.com/blog/most-powerful-search-api-for-ai/), [their pricing page](https://api-dashboard.search.brave.com/documentation/pricing)). The endpoints bundle web, images, news, and an "LLM context" mode. You get an index with no Google-shaped legal cloud, and results that skew independent-web rather than SEO-farm-web. The trade: a materially smaller index than Google, and no SERP features. **Keirolabs** is ours, so here are the exact numbers. [`/search/lite`](https://keirolabs.cloud/pricing) is $0.25 per 1,000 searches headline, on our own index. The free tier gives you 1,250 credits per month with no card, which is 12,500 lite searches at the 30 req/min limit. Monthly plans bill lite at 0.1 credits per request: Essential at $30/mo buys 125,000 lite searches ($0.24/1k), Pro at $50/mo buys 375,000 ($0.13/1k), and Startup at $100/mo buys 1,250,000 ($0.08/1k). One-time credit packs start at $10 if you never want a subscription: Starter $10 for 1,500 credits, Growth $30 for 5,000, Scale $100 for 20,000. Packs bill lite at 0.5 credits per request, which works out to $2.50–$3.33 per 1,000. Quality is not the trade-off it used to be, either. On AIMultiple's 2026 agentic search benchmark (scored /20), Keirolabs scored **15.2, #1 in the field**, ahead of Brave at 14.89, Firecrawl at 14.58, Exa at 14.39, and Tavily at 13.67. On FinanceBench financial QA, Keiro scores **78%, also #1**, ahead of Valyu (73), Parallel (67), Exa (63), and Google (55). That last row is worth sitting with in a post about Google search APIs: on financial questions, our own index beats Google's own results. And head-to-head against Parallel turbo, `/search/lite` won 91 of 100 comparisons with a composite score of 341.6 to 170.4, sweeping six of nine agent archetypes ([full breakdown](https://keirolabs.cloud/bench/keiro-lite-vs-parallel-turbo)). `/search/lite` is the cheap tier: 10 results per query, built for agents that fan out. The heavier endpoints are priced per credit, published on the pricing page: `/search/fast` at 1 credit per request, `/search/content` and `/extract` at 3, `/answer` at 5, `/agentic` at 20, and `/batch` at 1 credit per query. The decision rule I use: **does the user of your product need Google's ranking, or just good results?** Rank tracking, Maps, ad verification: Google. Agents doing RAG grounding, research pipelines, monitoring, enrichment: they need fresh, relevant results at a price that survives 10,000 queries a day. That's an own-index question, and at $0.08–$0.25 per 1k on plans versus $15, the arithmetic is doing most of the arguing. ## The boring takeaway The official Google search API era ends January 1, 2027. Google will take your $5/1k legacy traffic and offer you an enterprise semantic-search product with 1,000-QPM minimums in exchange; that's not a migration, it's a chasm. The market that replaced it has two shelves: SERP scrapers at $0.30–$15 per 1,000, and own-index APIs at $0.08–$5, where the top of the quality benchmarks now sits too. Pick by use case, not by price alone. If you need Google's answer (rank tracking, Maps, SERP features), buy from a scraper that's survived a lawsuit and diversify before Google's next move. If you need good results for agents at volume, an independent index is cheaper per query, legally boring, and the only option where your vendor's roadmap doesn't include a courtroom. And whatever you choose, put an abstraction layer in front of it. This market has changed three times in two years. Assume it will change again.