If your rank tracker suddenly started returning Google addresses instead of the sites it was supposed to be tracking, you were not imagining it.
On September 5, 2026, search data provider SerpApi published a post saying it had shipped a fix for Google’s google.com/goto redirect links, and that its Google Search API is once again returning the addresses of actual destination sites rather than a Google-owned wrapper.
What broke, exactly
Google widened its use of goto redirect addresses across search results. In practice, the anchor in the results markup no longer contains the website it points to. It points at Google, and Google forwards the visitor on.
SerpApi, whose post carries the byline of Hilman Ramadhan, its VP of marketing and developer relations, says the change briefly affected some of the URLs its service returned. Engineers identified the behaviour, built a fix and deployed it across affected result types.
Crucially, only one thing broke: the link field. According to SerpApi, titles, snippets, rankings and other parsed data were untouched. That makes sense. Titles and snippets are text on the page. Position is a property of ordering. The URL is the only element that identifies what is actually ranking.
Who felt it
SerpApi listed five categories of work potentially hit by the change:
- Rank tracking and search analysis
- Citation and source extraction
- AI agents and retrieval-augmented generation pipelines
- Link analysis and content discovery
- Any app that stores or compares destination addresses
Note the third item. A RAG pipeline building citations for a generated answer needs the source URL. Hand it a wrapper that only resolves on click and your chatbot footnote says “google.com”. The same field underwrites a keyword report and an AI answer’s sourcing. That is new territory for a SERP parsing bug.
Why one parser cannot fix everything
SerpApi offered one genuine engineering detail: the destination address can be encoded differently depending on result type, so a single parsing method could not resolve every variation.
Think about what a modern SERP contains. Text results, sitelinks, image blocks, video carousels, top stories, product units, paid listings. Each is rendered by a different subsystem. If encoding varies across those, every variant has to be taught to the parser separately. A fix announced as complete is a claim about the result types the vendor checked, not a proof about all of them.
The timeline Google didn’t advertise
The goto behaviour was not new in September. Alex Greenland reported the first public sighting on June 23, 2026. Brodie Clark documented the pattern across organic and paid listings on July 2, with a 302 carrying visitors to the destination. Barry Schwartz could not reproduce it on July 8, which reads like a bucketed test. Google confirmed the rollout on August 26, 2026, citing abuse rather than scraping, with no documentation, timetable or geographic breakdown.
On the same day, Nozzle CEO Derek Perkins published roughly four months of testing showing near total coverage across several residential IP providers. Two of his findings matter: encoded links cannot be decoded locally, and HEAD requests are refused. A GET that stops at the redirect works, but resolving every link across a five-page ranking runs to 500 to 1,000 requests. At that volume, rate limiting is the binding constraint, not bandwidth.
Which leaves a gap in SerpApi’s post. If local decoding is impossible, restoring direct addresses implies network calls that were not needed in June. The post says nothing about that cost.
What this means for your budget
A restored field is not a restored cost base. SISTRIX, the German visibility platform, has described the goto links as a change requiring an added-cost workaround, with founder Johannes Beus saying he expects more measures of the same kind.
This stacks on top of last year’s hit. Google eliminated the num=100 parameter on September 14, 2025; Semrush confirmed a tenfold operational cost increase within a day. AccuRanker capped tracking at the top twenty. SISTRIX dropped desktop SERP updates. Ahrefs began restoring top 100 tracking on October 29, 2025 without saying how.
The two changes compound. Removing num=100 multiplied the pages a provider must fetch. The goto wrapper multiplies the requests needed to interpret each page after fetching.
Three things to do this quarter:
- Ask your rank tracking vendor directly how it resolves goto links and whether pricing will move.
- Treat ranking data from before June 2026, during the test window, and after August 26 as three different datasets. Nozzle discarded goto-only pages during testing; others kept them. Do not draw trend lines across that seam.
- Audit any RAG or agent pipeline that stores source URLs from search output.
The legal backdrop
Context matters here: Google sued SerpApi on December 19, 2025 under DMCA Section 1201 over its SearchGuard challenge system. A judge dismissed every claim on July 20, 2026. Google filed an amended complaint on August 10 built on three licensing contracts. SerpApi moved to end the case with prejudice on August 24 — two days before Google confirmed the goto rollout. A hearing is set for September 29, 2026.
SerpApi’s own precedent is instructive. Its Google Light Fast API workaround after num=100 was restricted to three organic results within weeks. A fix that works on September 5 is a statement about September 5.
Source: PPC Land
