Sometime around 23 June 2026, Google started replacing the destination addresses in its search results with a passthrough link. By late August it was confirmed and close to fully rolled out. The result link you see in the page markup is no longer https://example.com/page. It is google.com/goto?url= followed by an encoded parameter, and clicking it issues a 302 redirect to the real address.
If you are a person using Google, nothing has happened to you. You click, you arrive, and the extra hop is invisible.
If you are anything that reads search results without being a person, quite a lot has happened to you.
What the change actually does
The mechanism matters more than the announcement, so it is worth being precise about it.
Previously, a tracker could fetch a results page once and read every destination address straight out of the HTML. One request produced a full picture of who ranked where and what URL they ranked with.
Now the destination is encoded inside the goto parameter, and — this is the load-bearing detail — it cannot be decoded locally. Derek Perkins of the rank-tracking firm Nozzle put it plainly: the goto links cannot be decoded locally, which means providers have to follow the redirect rather than read the address.
Following a redirect is a network request. Do it for every result on every page of a ranking and the arithmetic turns ugly fast. Perkins reported that resolving all links for a single five-page ranking now takes between 500 and 1,000 requests. Google also blocks HEAD requests — the cheap method that retrieves only headers — which forces a full GET for each link.
Scraping cost used to scale with the number of queries you tracked. It now scales with the number of links on every page of every query you track. That is a different economic curve, and it was almost certainly chosen because it is a different economic curve.
What Google says, and what the timing says
Google’s statement is brief. A spokesperson said the company has a long history of deploying technical measures against evolving forms of abuse, and that it regularly takes steps to protect its services and users.
That is not nothing, and it is not a complete account either. The context is a live dispute with SerpAPI, which Google sued in December 2025. In July 2026 Judge Yvonne Gonzalez Rogers granted SerpAPI’s motion to dismiss the initial DMCA claims; in August, as the goto rollout intensified, Google refiled an amended case.
You are free to read that sequence as coincidence. We do not think you have to.
It also sits in a line of changes with the same direction of travel. In January 2025 a JavaScript requirement disrupted several tools — Similarweb, Rank Ranger, SE Ranking and ZipTie were all affected, while Ahrefs and SISTRIX were not. In September 2025 Google removed support for the num=100 parameter, which had let a single request retrieve a hundred results. Each of these individually is defensible as anti-abuse. Together they describe a policy.
The part that should worry you
Here is the consequence that matters for anyone reading a monthly report.
Rank tracking beyond position ten or twenty is now materially less trustworthy than it was two years ago. Not wrong, necessarily — less trustworthy, which is a different and more awkward problem, because a number that is sometimes right does not announce which times.
The clearest illustration in the coverage is a Reddit case study: keyword visibility collapsed 25% month over month between May and June 2026 while actual traffic stayed flat. A 25% visibility drop with no traffic change is not a ranking event. It is a measurement event. If you had seen that chart in a report without the traffic line beside it, you would have spent a quarter solving a problem that did not exist.
This is the same failure mode we wrote about in Production Got Cheap. Verification Got Expensive. — the instrument moved, the business did not, and the report cannot tell you which.
What this has to do with AI search
More than it first appears.
The tools that estimate AI visibility — share of answer, citation counts, prompt-level brand tracking — are built on the same foundations as rank tracking: automated retrieval of a results surface that the platform does not provide as data. Google’s Search Console reports AI impressions, and only impressions. Everything richer than that is inference from scraped surfaces.
So when the cost of reading a results page multiplies, the cost of every derived AI-visibility estimate multiplies with it. Sampling gets thinner. Refresh intervals get longer. Nobody sends an email about it, because coverage reduction does not look like an outage — it looks like a chart with slightly fewer data points, which looks like nothing at all.
Meanwhile AI Mode has started requiring sign-in for some queries. Reading a logged-in surface at scale is harder again.
The measurement layer is not failing. It is closing, deliberately, from several directions at once, and mostly for defensible reasons.
What to do this week
Four things, none of which take long.
Ask your provider a direct question in writing. Do you resolve goto links, and what changed in your coverage or sampling frequency after August 2026? SerpAPI has stated it updated its service to resolve goto URLs correctly. Your vendor may have done the same, may have quietly reduced how often it refreshes, or may not have noticed. All three are survivable. Not knowing which is not.
Re-baseline, and say so out loud. Draw a line at August 2026 in every visibility chart you report against. Treat what is on either side of it as two series, not one. A trend that crosses an instrumentation change is not a trend.
Stop reading deep-position movement as signal. Positions one to ten remain reasonably observable. Below that, treat movement as noise unless your provider has explained how it is still produced.
Move weight onto first-party data. Search Console, server logs, and your own analytics are unaffected by any of this, because none of them require reading Google’s results page. They are narrower than the dashboards. They are also still true next quarter.
What we will and will not claim
We will not tell you your rank tracking is now worthless. Position-level data on page one still works, and the vendors are not sitting still.
We will tell you that the honest version of a September 2026 visibility report has a footnote in it, and that any agency showing you smooth deep-position trend lines across the last quarter without mentioning the goto change either has not checked, or is hoping you will not.
Checking takes one email to your provider. We would send it before the next report lands.