Google’s vanished “Web Search Service API”, autopsied page by page.
The documentation for a partner-only Google search API was public for roughly a month, was maintained right up to 19 Sep 2026, and became a hard 404 by 20 Sep. The trade press reported the disappearance. Here is what the archived document itself actually said — what Search Engine Journal documented from the deeper pages before they died too — and what it means for anyone who pays for search results programmatically.
First: what is provable from the archived copy
The Wayback Machine captured developers.google.com/web-search-service/overview on 20 Sep 2026 at 10:09 UTC — hours before the URL turned into the 404 that our public ledger began recording the same day. The archived page carries the notice “Last updated 2026-09-19 UTC”: the document was being actively maintained one day before deletion. That single stamp changes the shape of every plausible explanation of what happened next.
The page body is quoted below verbatim where quoted. The archived page states its own license — Creative Commons Attribution 4.0 — and it is quoted accordingly, with the archive link beside it.
Six verifiable facts from the archive. Everything else in this post is labelled as inference or attributed reporting.
“REST or gRPC”
Quoting the page: partners could use “REST or gRPC requests to retrieve search results in JSON format.” A dual-interface API serving JSON — the same modern serving-stack pattern as Google’s Vertex/Gemini-era services. This is not a renovated legacy endpoint.
API key + a contractual client ID
The prerequisites, verbatim: “Google Cloud Project”, “API Key: created in the Google Cloud Console”, and “Client ID: a designated client_id associated with the partner agreement.” API-key plumbing is public-facing; the partner agreement is the gate. A commercial tier was being documented, not an experiment.
“Programmatic partners”
The service is described as allowing “programmatic partners to develop websites and applications that retrieve and display Google Search results.” The document was written for external partners building products on top of Google’s search output — not as internal runbook material.
Cloud-native monitoring, already wired
Monitoring is documented as available natively through the Google Cloud Console APIs Dashboard — quota, traffic and error observability for this service was already plumbed into Cloud Console before the page vanished.
A page tree, not a page
It lived under a dedicated hub — developers.google.com/web-search-service/ — and Search Engine Journal reported that Google updated the overview plus four related pages (an introduction, REST usage guides, a reference). Deleting it meant uprooting a hub, not removing a stray file.
No pricing, no quotas, no on-ramp
The archived overview contains no pricing and no rate limits, and SEJ’s read of all the pages agreed: nothing on what the service costs, what limits apply, or how a company becomes a partner. The Wayback Machine holds three captures in total, all between 13–20 Sep 2026, of the overview URL only — the deeper pages died unarchived and unquotable.
The timeline, every date checkable
| Date (2026) | Event | Where to verify |
|---|---|---|
| January | Google announces web-search product changes: the Custom Search JSON API is closed to new customers and will be discontinued 1 Jan 2027; a separate full-web solution is promised to partners who need “our entire index”. | Google’s January blog post |
| 9 Sep | Google updates the Web Search Service API overview page and four related pages. | SEJ’s report |
| 11 Sep | Search Engine Journal publishes the first detailed public description of the API. | Same SEJ report |
| 13 Sep | First Wayback Machine capture of the overview URL. | Wayback capture index |
| 19 Sep UTC | The document’s own “Last updated” stamp — its final edit. | The archived copy itself |
| 20 Sep, 10:09 UTC | Wayback archives the page — the surviving copy. | Archive snapshot, linked above |
| 20 Sep, later | The URL returns HTTP 404. Our Policy Radar begins logging the 404, and has logged it every day since — day eight on 28 Sep. | Public Policy Radar ledger · change log |
| 25 Sep | Our running log post dates the 404 publicly, day five. | The running log |
| 28 Sep | Search Engine Roundtable publishes the removal story, sourced via Manas Panda. | Search Engine Roundtable · the LinkedIn note |
What SEJ documented from the deeper pages
Our archive holds only the overview. SEJ’s Matt G. Southern read the hub’s deeper pages while they lived — all of those URLs are 404 now and were never archived, so the inventory below is SEJ’s reporting, not our archive. We repeat it because it materially fills the picture:
- One primary method,
Search, atreference/rest/v1/TopLevel/search, described as performing a “full web search”. - Each request carries the partner client ID, the end user’s IP address (per the docs, for regional routing and misuse prevention), and the query.
- Up to 20 results per request, 10 by default, with a token for further pages.
- Filters: language, country, date range, SafeSearch, date sorting.
- Response fields: title, URL, snippet, MIME type, file format, an estimated total-results count, and a corrected query when Google suggests one.
- The client ID follows a standard format encoding the partner, product and feature set.
Why the January context matters
SEJ also noted the match: in January, Google told Custom Search users needing full-web results that a separate solution existed for “those requiring our entire index”, with an interest form and no details. The vanished API matches that promise on both points — full web, partners only — at the exact moment the legacy Custom Search JSON API is walking toward a hard 1 Jan 2027 deadline.
So the document was not trivia. It read like the migration path nobody had been shown yet — documented to the level of request fields and IP routing, maintained on 19 Sep, then pulled on 20 Sep with no redirect and no note.
Whether it returns under the same name, a new name, or not at all: we keep the URL in the ledger and we keep watching.
Fact, inference, unknown — kept strictly apart
FACT (archived text)
- Dual REST + gRPC interfaces returning JSON.
- Cloud Project + API key + a
client_idbound to a partner agreement. - Cloud Console monitoring wired in.
- “Last updated 19 Sep”; deleted by 20 Sep; still 404 on 28 Sep.
INFERENCE (we label it)
- A partner agreement with a dedicated client ID implies commercial terms — a paid tier, not a free preview.
- A last edit on 19 Sep followed by deletion on 20 Sep reads like an accidental early publish, pulled fast — abandoned pages do not get edited the day before they vanish.
- REST+gRPC plumbing suggests modern serving infrastructure — consistent with groundwork for an eventual successor to the legacy Custom Search line, though Google has connected the two nowhere in public.
UNKNOWN (we will not guess)
- Pricing, quotas, eligibility, and whether any partner’s traffic actually ran through these endpoints.
- Why the removal was a hard 404 instead of a redirect or deprecation note. The industry’s known precedent for partner search arrangements is Apple-style — we note the precedent; we do not claim it for this document.
- Whether republication is planned, and under what name.
Why the 404 matters — and why the 19-Sep stamp matters more
Anyone can delete a page. Google kept this one alive for roughly a month, edited it a day before removing it, and removed it without a redirect. If the document had been abandoned infrastructure, nobody would have touched it on 19 Sep. If it had been a deliberate launch, there would have been an announcement. The observed shape — maintain until yesterday, vanish today, say nothing — matches an accidental publication pulled in a hurry, by every editorial standard we know how to apply. That is an inference, labelled as one; but it fits the dated sequence better than any announced-sunset story, and the sunset-story version is the one Google did not tell.
Why anyone outside Google should care: programmatic access to full-web search results is repricing right now — a legacy API walking toward a 1 Jan 2027 cliff on one side, AI-grounding bundles on the other. A partner-priced Google “search as a service” would set the floor under RAG pipelines, answer engines, media monitoring and compliance tooling. The vanished document confirms that on 19 Sep 2026 Google was actively writing partner-facing terms for exactly that. That is now a dated public fact with an archived copy. It no longer depends on anyone’s word — including ours.