Google Search is replacing the direct links in its results with encoded redirects through google.com/goto, so software reading a results page can no longer see where each result leads without asking Google one link at a time. Google confirmed the change in late August as a defence against "evolving forms of abuse," and it raises the cost for the SEO tools, search APIs and AI agents that scrape Google results, even though the biggest vendors say they have already worked around it.
Key facts
- 493 of 493: a study by the marketing agency Growtika ran eight searches and found that every one of the 493 result links it collected hid its destination.
- When: the data firm Decodo traces first sightings to June 2026, Google confirmed the rollout on 26 August, and the story went viral on Hacker News on 12 September.
- Who: Google; SERP API vendors including SerpApi and Autom are among the tools affected.
- Primary source: Google's statement to Search Engine Land.
What changed
For most of Google's history a results page contained, somewhere in its code, the plain web address of each result. That made it easy for software to read. SEO rank trackers, price monitors, and the growing crowd of AI agents and retrieval pipelines that search the web on a user's behalf all rely on it.
Now each result points to a google.com/goto address instead. According to Autom, "the url parameter uses a custom, Google-specific encoding. It is not a plain base64 of the target URL." SerpApi found that the destination is present "in the form of ciphertext binary, packed into Protocol Buffers." A person clicking a link notices nothing, because Google's server answers with a redirect to the real site.
It is like a library that replaces the shelf numbers in its catalogue with ticket numbers: anyone can still get a book, but only by queuing at the desk once per title.
In a check by Ground Truth on 12 September, a Google results page loaded in a real browser engine contained no direct links to outside sites. Each result went through a goto address that returned the real destination only when requested individually.
What Google says
Google has published no documentation of the change. Its statement to Search Engine Land's Barry Schwartz was general: "We have a long history of deploying technical measures against evolving forms of abuse, and we regularly take steps to protect our services and users." Derek Perkins of the rank-tracking company Nozzle told Search Engine Roundtable he saw "nearly a 100% rollout across several residential ip providers," and predicted each results page would carry "hundreds of links to decode."
Search Engine Land drew the connection to Google's legal fight with scrapers. In July a court rejected Google's DMCA claim against SerpApi. A technical barrier does what that lawsuit did not.
Broken, or just more expensive?
The viral framing on 12 September was that Google had broken the scrapers AI tools run on. The vendors' own posts describe something narrower. SerpApi wrote on 5 September that it "is now returning direct destination URLs again," while listing "AI agents and RAG pipelines" among the workflows that had been affected. The data company Decodo estimated that resolving the links adds "500 to 1K requests per five-page ranking report." The real change is cost and visibility: every lookup is now a request Google can count, slow down or block.
The post that took the story to Hacker News, where it drew more than 450 comments, came from Autom, which sells a Google search API. It was published on 27 August, not this week. One of the top comments made the point directly: "It's relevant to know when the source of information is biased due to a conflict of interest."
Why it matters
Search results are the raw material for a growing share of AI products, from deep-research agents to citation features. If the dominant search engine meters access link by link, independent builders pay more, or they move to the handful of paid and alternative indexes. At least one developer in the thread said they had already switched to an alternative index for their agents. It also sits alongside Cloudflare's move to split AI crawlers into separate permission switches: the open web's plumbing is being rebuilt around who is allowed to read it automatically.
The caveat
Nobody at Google has said the change is aimed at AI, and the evidence that it targets any particular group is inference from vendors with a commercial stake. Reports also suggest it may not yet apply everywhere: Growtika tested only from US connections, and at least one European reader said they did not see goto links.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)