Google's goto links tax every SERP copy

Google's goto links are server-side redirects that replace the direct destination URL of a search result with an encoded google.com/goto address, so any program that reads a results page has to ask Google again to learn where each result points.

For a person clicking a result, nothing visible changes. For anything that copies Google's results at scale, the cost of each copy just went up. That group includes the rank trackers and AI Overview trackers most SEO teams report from, and some of the AI products that have borrowed Google's ranking to answer questions.

#What changed

Google confirmed the rollout to Search Engine Roundtable on August 26, 2026, with a general statement: "We have a long history of deploying technical measures against evolving forms of abuse." Derek Perkins of Nozzle, who spotted it, reported "nearly a 100% rollout across several residential ip providers" after a test period.

The mechanics matter more than the announcement. According to Autom's technical write-up, the old format was google.com/url?q= followed by the destination in plain URL encoding, which any parser could read offline. The new format is google.com/goto?url= followed by a custom Google encoding that cannot be decoded without asking Google. The destination only appears in the Location header of Google's response. Autom observed the pattern consistently for logged-out and private browsing sessions.

A results page therefore stops being a self-contained document. Each organic result costs one more request to Google, and a client resolving hundreds of links in sequence gives Google a clean signal of what it is.

#The fourth step in the same direction

The goto change is easier to read as part of a sequence than on its own.

• On January 15, 2025, Google began requiring JavaScript to load search results, a move Search Engine Journal confirmed was aimed at bots and scrapers, including SEO tools.
• In September 2025, Google stopped honouring the num=100 parameter that let tools pull 100 results in one request. A Google spokesperson told Search Engine Land that it was "not something that we formally support."
• On December 19, 2025, Google sued SerpApi, a company that scrapes and resells Google results. The complaint describes SearchGuard, the anti-bot system Google launched in January 2025, and claims SerpApi's automated requests grew by "as much as 25,000%" over two years, according to IPWatchdog's summary.
• In August 2026, goto links made every scraped result more expensive to resolve.

Each step raised the price of reading Google automatically. None of them changed what a signed-in or ordinary searcher sees.

#Why this reaches AI answers

In 2025, The Information reported that OpenAI had partly used Google results scraped through SerpApi for ChatGPT answers on current events such as news and sports. Search Engine Roundtable's summary of that report also names Meta, Apple and Perplexity as reported SerpApi customers. None of this was confirmed as a permanent architecture, and it is reporting rather than documentation, but it describes a real path: a page ranks in Google, a scraper copies the ranking, and an AI answer inherits it.

That path is the one Google keeps making more expensive. Our inference, not a documented fact: as borrowing Google's ranking gets costlier and more legally exposed, AI engines lean harder on indexes they own or license. We have covered two of those already. Claude's web search appears to run on Brave, and ChatGPT's link to Bing is partly documented. The practical reading is that a strong Google position transfers to third-party AI answers less automatically than it did, and visibility in the other indexes has to be earned directly.

#Why this reaches your reporting

Rank trackers and AI Overview trackers are scrapers. The goto change hits them the same way it hits anyone else.

The num=100 removal showed what that looks like in practice. Search Console impressions fell sharply for many sites while clicks held and average position improved, because tools had been generating impressions deep in the results. Locomotive's analysis walks through the arithmetic: bot impressions at low positions had been dragging the average down. A change in how data was collected looked like a change in performance.

Goto links carry the same risk in a different form. A vendor that resolves each link adds cost and request volume, and may respond by tracking fewer keywords, checking less often or sampling. A vendor that fails to resolve some links may record google.com/goto as the URL, or drop results it cannot attribute. In an AI Overview tracker, that can show up as a citation that seems to vanish, or a competitor that seems to appear, when only the parsing changed.

#What to do this quarter

Ask your vendors one question. How do you resolve goto links, and did your keyword volume, check frequency or sampling change after August 2026? A vendor that has handled it will answer quickly. Autom, for example, says it now resolves them and returns destination URLs in the existing fields.

Check your exports for google.com/goto. Search tracked URL columns for the string. Any hit means a result was stored without its real destination and your position or citation data for that row is wrong.

Annotate late August 2026 in every dashboard. Treat a sudden shift in tracked rankings or AI Overview citations near that date as a collection question before a performance question, the same way the num=100 week should be annotated.

Lean on first-party Google data where it exists. Search Console is measured by Google itself rather than copied from results pages, so a vendor's parsing problem cannot enter it. It is not immune to scraping, as the num=100 week proved, and it will not tell you which AI Overview cited you. It is still the Google series least exposed to a collection change like this one.

Measure the other indexes on their own terms. If Google's ranking is becoming harder for others to copy, then a ChatGPT, Claude or Perplexity citation says less about your Google position than it used to. Track each engine as its own series, and check your standing in Bing and Brave for the queries that matter rather than assuming Google carries over.

#The short version

Google has spent twenty months making its results expensive to copy. The goto redirect is the latest step and probably not the last. Anyone whose reporting or visibility depends on a copy of Google, whether a rank tracker or an AI answer built on scraped results, should expect that copy to get thinner, slower or more expensive, and plan measurement that does not depend on it.