Google no longer links search results directly, but through an intermediate address on google.com/goto. On 26.08.2026 a Google spokesperson confirmed the rollout. The destination address is no longer written in the link, it is encoded.
- The format was first spotted on 23.06.2026 and confirmed two months later.
- Google’s statement speaks of technical measures against abuse. It does not name the goto parameter, and there is no documentation for it.
- For rank trackers, measuring becomes more expensive rather than impossible: Nozzle puts it at 500 to 1,000 requests to resolve a five-page ranking. Sistrix is already resolving.
- AI Overviews were hit hardest: on 27.08.2026 at least one URL stayed unresolved in 52.26 percent of those SERPs, against 0.01 percent for organic ones. A day later the provider announced an update that resolves those cases too.
- The context Google’s statement leaves out: in July a US court dismissed Google’s DMCA claims in so far as they concerned search results without copyright-protected components.
On 31.08.2026 I typed „seo kreativ” into Google and hovered over my own result, the way I do before every snippet check. The bottom left did not say seo-kreativ.de. It said google.com/goto, followed by a string that looked like nothing at all.
For two months that was an observation individual people reported on X. On 26.08.2026 Google confirmed technical measures against abuse, which the trade press attributes to this change. Since then, when you hover over a result in Google’s search results, the link no longer points at the website but at an address on Google.
For you as a user this changes almost nothing. You click, you land where you wanted to land. For every tool that reads search results, it changes the arithmetic. And on the question of why Google is doing this, Google’s own answer is the least interesting source.
This article separates three things that often run into each other in the reporting: what happens technically, how much of it is documented, and what follows from it for your numbers.
What Google changed technically
The pattern has been stable since the earliest publicly documented sighting. Alex Greenland described it on X on 23.06.2026 like this: “Result link URLs are now being rewritten to google•com/goto?url=[custom base64-like encoding of URL or Google ID]. Upon click, they redirect to their proper destination.”
Two details in there are the actual news.
First: the address is encoded, not written. Previously the target URL sat in the link and a program could simply read it. Now there is an encoded string in its place. Anyone who wants to know where a result leads has to follow the redirect and ask Google.
Second: in the available tests it is a 302. That is, a temporary redirect. This matters technically, because a 302 can signal to crawlers that the intermediate address is not meant to permanently replace the destination address. How Google processes the address internally does not follow from it. The goto URL is built as a passthrough, not as a new address for your page. How Google handles redirects in general is covered in the overview of the search algorithm.
I looked at what is inside that parameter. The token from this capture is 116 characters long and decodes to 87 bytes. Those 87 bytes carry a protobuf structure with exactly two fields: a number with the value 1, presumably a version marker, and a block of 83 bytes. Nothing readable sits in that block. No domain name, no path, no plain text.
That makes the common phrasing, that the address is „encrypted”, imprecise. The token examined contained no readable destination address, but a reference that could not be resolved without Google’s redirect. That is a single finding on one token and not a statement about every goto link. This is why no clever decoding helps. Whether an extra request becomes necessary, however, depends on the feature: where the destination address is still in the page source, it can be read there. Where it is not, only the route via the redirect remains. What Derek Perkins means by „can’t be decoded” is therefore not a complaint about effort, it is a description of the construction.
Is a token like that risky to show? I redacted it at first, because I could not determine its contents. That was a guess, not a measurement, and the measurement comes out differently. I requested the token from the capture above from a foreign server, without cookies, from a different IP: answer 302, destination https://www.seo-kreativ.de/. Three requests in a row return the same. So it is not bound to my session, it is portable.
It cannot be altered, though. A single different character in the payload block, a different final character, or ten characters fewer: every time HTTP 400 instead of a redirect. Those 83 bytes carry an integrity check.
Two things follow. In the token I tested, nothing personal was discernible and the request was not bound to my session; showing it therefore gives away only the destination that the screenshot shows anyway. That is not a statement about every goto link. And this is why resolving costs exactly one request and not one less: you cannot read the token, you cannot build one either, but you may redeem it as often as you like.
Location header holds your destination. That single request is exactly what every rank tracker has to spend per link, and it is what adds up to the 500 to 1,000 requests per ranking that Nozzle quotes.The path from sighting to confirmation was long. Publicly documented on 23.06.2026, picked up as a test on 08.07.2026, confirmed on 26.08.2026. An earlier start is not ruled out: in July Derek Perkins wrote that they had been seeing such redirects for months. In between, Brodie Clark documented the same behaviour in the SERP Alert newsletter, among others.
What Google says about it and what it does not
A Google spokesperson told Search Engine Roundtable: “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.”
Read that sentence again. It says that Google takes technical measures against abuse. It does not say what this is about, since when, for whom and at what scale. The word goto does not appear in it.
Attributing this statement to the goto change comes from those reporting, not from Google. It is plausible, because the question put to Google was precisely about that. Documented in the sense of a Google statement it is not.
And this statement exists exactly once. It appears at Search Engine Roundtable and, seven hours later, at Search Engine Land. Both times it is the same author, Barry Schwartz, and both times the wording is „told us”. These are not two independent confirmations but one inquiry in two publications. Reading the news in two places therefore does not double the evidence.
What this costs the measurement
Derek Perkins of Nozzle has been watching the rollout since the test and describes it as “nearly a 100% rollout across several residential ip providers”. On the effort he says: according to Nozzle, resolving all links of a single five-page ranking costs between 500 and 1,000 requests.
The obvious shortcut does not work. From the reply thread under Perkins’ post, PPC Land reports that Mic King asked whether HEAD requests would be enough. A HEAD request retrieves only the headers without the page body and would return the redirect target cheaply. Perkins is reported to have answered that Google does not let the HEAD request through. Both are PPC Land’s account; the individual replies are not linked there, and I have not seen them in the original.
So the problem is neither bandwidth nor storage, it is rate limiting. Anyone who has to send a few hundred extra requests to Google per keyword eventually hits a limit that Google sets rather than their own server.
How quickly that limit arrives I happened to see myself. For this article I wanted to load a google.de results page in an automated browser to look at the links in the source. The attempt did not end at the results but immediately at google.com/sorry, the CAPTCHA page. A single request from an automated browser, no series, no tooling.
That is not a measurement of the rollout and is not meant to be one. It is the cheapest illustration of what Perkins and Beus are talking about: the route to the results is not closed, but it costs, and Google sets the price. Anyone reading the goto resolution as an isolated problem misses that it sits on top of a barrier that was already there.
What the tool providers are doing
Johannes Beus of Sistrix commented a day after the confirmation and first places the matter in context: Google has been making the collection of search results harder for about a year, he writes, and the loudest public echo back then came from the end of the num=100 parameter.
On their own approach he writes that they are already resolving these addresses to the real destinations and are scaling that up, with every resolved URL costing one additional request. Measuring search results becomes „a bit more complex and more expensive again, but it is not prevented”. Update cycles will be slower for a while, while „nothing changes about the completeness and comparability of the data”.
More interesting is who he sees as the target. The measures are aimed „primarily at competitors of Google such as ChatGPT, which run no search index of their own and still use Google’s results”. Tools like Sistrix are „collateral damage and not the actual target”. That is his reading and not a Google statement, but it explains the direction better than the reference to „abuse” does.
Not every provider reports the same. On 28.08.2026 Glenn Gabe pointed to a status message from Ahrefs speaking of a „Google keyword ranking data disruption”; in his assessment they have no solid solution yet. The message itself is an in-product notice and was not retrievable for me; it is documented through Gabe’s post. That leaves three positions side by side: Sistrix is scaling the resolution, DataForSEO reports it as largely solved, Ahrefs reports a disruption.
Both sets of figures, those from Nozzle and those from Sistrix, are provider statements about their own infrastructure. That does not make them wrong, quite the opposite: nobody else can measure this. It only means there is no independent cross-check for them.
Where the measurement actually breaks
The common narrative is “rank tracking in danger”. The numbers say something else, and they say it rather clearly. On 27.08.2026 DataForSEO disclosed for how many SERPs at least one goto URL remained unresolved after the change, broken down by SERP element.
| SERP element | Share of SERPs with at least one unresolved goto URL |
|---|---|
| Organic results | 0.01 % |
| Featured snippet | 0.39 % |
| Local pack | 25.1 % |
| AI Overview | 52.26 % |
Source: DataForSEO status update of 27.08.2026. Share of SERPs with at least one unresolved goto URL per element. Provider statement about its own infrastructure, not an independent measurement. As of 31.08.2026.
And these figures are already superseded. The same page carries an addendum: „On August 28, we released an update that enables us to return direct destination URLs across all affected SERP elements, including AI Overviews.” The table above is therefore the state of 27.08.2026 and not the current condition. It stays instructive nonetheless, because it shows which elements took longest.
The finding in one sentence: for classic rank measurement the gap was small from the outset according to the provider, for AI Overviews it was large and was closed a day later. Note that these are shares of SERPs and not of individual links: a SERP counts as soon as a single URL in it stays unresolved.
That weighs more than it sounds. For organic rankings you have years of comparison values and notice immediately when a series tips over. For visibility in AI Overviews you have months. A gap in a young time series looks like a fluctuation, and nobody has an experience value that contradicts it.
What the masking really looks like was measured by Hanns Kronenberg of Chefkoch on 27.08.2026: 1,040 keywords on google.de through the DataForSEO interface, then the same queries cross-checked in ordinary browser sessions, 44,072 classified URLs in total.
His first finding contradicts the assumption that this hits everyone alike. For „girokonto vergleich” he saw 116 external links in the browser and not a single goto address. In the HTML the same query returned through the interface, there was not one direct result link. For „best checking accounts” on google.com, the same picture. His summary: what matters is not the market but how Google treats the client.
That is why my capture above and his measurement can both be right. I saw the goto address in an ordinary browser, he did not see it in his browser sessions. Why one client gets the redirects and another does not, he explicitly says he does not know.
His second finding explains why some tools stay clean and others do not. The real destination address has not disappeared, it sits elsewhere on the page, in an embedded data block next to the redirect. Where that block is missing, only the Google address remains. Broken down by feature:
| Feature | Destination address readable |
|---|---|
| Organic results | 17,347 of 17,347 |
| Videos | 1,204 of 1,204 |
| Comparison boxes | none, mobile and desktop alike |
| Recipe galleries mobile | 12 of 1,956 · all of them on desktop |
| Local pack desktop | 9 of 276 · all of them on mobile |
Source: own measurement series by Hanns Kronenberg, published 27.08.2026. The basis is 1,040 keywords and 44,072 classified URLs. The column states for how many results the real destination address was still present in the page source.
google.com. Anyone filtering their SERP data by domain therefore loses those results silently. There is no error, no gap, no warning. The analysis looks valid and still degrades quietly. If you work with SERP exports, that is the point to check in your filters.The context the statement leaves out
Google filed suit against the provider SerpApi on 19.12.2025. The claim rested on Section 1201 of the US Digital Millennium Copyright Act: SerpApi had circumvented Google’s anti-bot barrier named SearchGuard.
On 20.07.2026 US District Judge Yvonne Gonzalez Rogers granted SerpApi’s motion to dismiss (order as PDF). The reasoning is more instructive for our question than the outcome, so here it is in three steps:
- The alleged acts of circumvention were sufficiently pleaded. At this stage the court takes the complaint as true and only tests whether it states a claim. It found it sufficiently pleaded that SerpApi circumvented SearchGuard by making automated queries look like human ones, misrepresenting the device, software or location they came from, and by passing an authorisation obtained by one browser on to browsers that had not been cleared. Nothing is established by this; what was tested is the sufficiency of the pleading.
- For results without protected components the case ended on this pleading. Section 1201 requires the technological measure to control access to a copyright-protected work. In so far as SearchGuard controlled access to search results containing no copyrighted content, the court dismissed the claims without leave to amend.
- For results with protected components it stayed open. There the court objected above all that Google had not shown it was authorised by the respective rights holders to deploy SearchGuard. On that point it gave Google 21 days to amend.
Several trade publications draw a connection between this ruling and the goto change. That is an obvious reading and it is not my invention. But it is not a Google statement either. What can be evidenced is the sequence alone: one route to preventing the reading did not hold up in court on this pleading. Another route, making it more expensive, was rolled out shortly afterwards. Whether one has anything to do with the other is a hypothesis and stays one.
What changes for you and what does not
I am sorting this by certainty, not by excitement.
There is no indication of an effect on Search Console. It counts impressions and clicks on Google’s side and not based on what arrives in your log. An impairment of those figures is not documented so far, and there is no explicit technical statement from Google about it. If you want to judge your own visibility, it is therefore the most stable source you have. How I read my own numbers there is covered in Ranking Volatility August 2026.
Affected are tools that read SERPs. Rank trackers, SERP APIs, in-house scripts. Not in the sense that they stop working, but that they need more requests and become slower or more expensive as a result.
Open is the effect on referrer and attribution. And this is where many articles get sloppy. The redirect sits on a Google domain, so there is nothing to suggest a foreign source will suddenly show up in your analytics. Except: Google has documented nothing about this. There is no provider statement you could check it against.
What I do not recommend: rebuilding your tracking now. The change is a few days old, undocumented, and the tool providers are currently absorbing it. A rebuild aimed at a state that is still moving is at best ineffective.
Frequently asked questions (FAQ)
What is a google.com/goto link?
An intermediate address at Google that a search result link runs through before you land on the target page. The destination address is no longer readable in the link but encoded in the parameter. On click, Google answers with a 302 redirect to the actual destination.
Since when has this existed?
The format was first publicly spotted on 23.06.2026 and reported as a test on 08.07.2026. A Google spokesperson confirmed the rollout on 26.08.2026. Google has not named an exact start date for the full rollout.
Does this hurt my rankings?
There is no indication of that. As things stand, the change concerns the technical linking in the result list, not the evaluation of your page. A ranking effect is not evidenced by the available sources. What changes is the effort with which tools read results.
Is my Search Console data affected?
No. Search Console counts impressions and clicks on Google’s side and not based on the link that arrives at your end. Search Engine Land explicitly notes that the data should not be affected.
Do rank trackers still work now?
Yes, but with more effort. According to its own statement, Sistrix is already resolving the addresses and scaling that up; update cycles will be slower for a while, while completeness and comparability are said to remain unaffected. Nozzle puts the effort at 500 to 1,000 requests for a five-page ranking.
Why is Google doing this?
Google itself only says that it takes technical measures against abuse. Trade publications draw a connection to the legal dispute with the provider SerpApi, in which a US court dismissed part of Google’s DMCA claims for good in July 2026. That connection is a reading of the reporting, not a Google statement and not an evidenced causal link.
Conclusion: made expensive, not closed off
The sober version of this news is unspectacular: Google has introduced an intermediate address that makes machine reading of the results more laborious. The providers are absorbing it. A ranking effect is not evidenced so far. Nor is there any indication of an impairment of your Search Console data.
Anyone turning this into a crisis of rank measurement is looking at the wrong number. For organic results, an URL stays unresolved in 0.01 percent of SERPs according to the provider. The gap sits with the AI Overviews, and that is precisely the measurement for which nobody yet has a reliable time series.
The story is interesting anyway, and it is the sequence that makes it so. In December Google sues a scraping provider. In July a court dismisses the claims for good in so far as they concern results without protected components. In August a technical measure rolls out that does not forbid the reading but makes it expensive. Google does not connect these three points, and I therefore do not assert the connection as fact either. I merely note that the chronology runs like this.
What I take from it: when a route through the law does not hold, the route through cost remains. That is not a headline, but it is the movement worth keeping an eye on if your work depends on data from someone else’s systems. The development around AI crawlers and how they are blocked is of a similar kind.
Sources
What is linked in the text appears here again with date and role. Added to it are the trade articles whose citation in the text has since been replaced by the primary source.
- Primary sources: Alex Greenland on X, 23.06.2026 (first sighting) · SERP Alert on X, 08.07.2026 (independent observation) · Glenn Gabe on X, 28.08.2026 (Ahrefs status message) · Derek Perkins on X, 26.08.2026 (rollout and effort) · Order of the US District Court, Google v. SerpApi (PDF), 20.07.2026
- Search Engine Roundtable, Google Testing google.com/goto As Tracking URLs In Search Results, 08.07.2026
- Search Engine Roundtable, Confirmed: Google Search Rolling Out google.com/goto Tracking Parameters, 26.08.2026 (Google statement, Nozzle)
- Search Engine Land, Google confirms deploying goto URL redirects to search results links, 26.08.2026
- PPC Land, Google forces rank trackers into 500 to 1,000 requests per query, 27.08.2026 (302, HEAD requests, missing documentation)
- SEO Südwest, Sistrix on Google’s new goto links, 27.08.2026 (Johannes Beus)
- DataForSEO, Google’s Goto Rollout: We Already Resolve 99.99% of URLs, 27.08.2026
- OnlineMarketing.de, Google goto URLs erschweren Site Tracking, 27.08.2026 (measurement series by Hanns Kronenberg)
- Search Engine Roundtable, Google Lawsuit Against SerpApi Over Scraping Search Results Has Been Dismissed, 22.07.2026 (analysis of the decision of 20.07.2026)
- Hanns Kronenberg on LinkedIn, 27.08.2026 (own measurement series, 1,040 keywords and 44,072 URLs)
- Johannes Beus on LinkedIn, 27.08.2026 (Sistrix on their own approach)
- Not linkable: the reply thread under Derek Perkins’ post, from which PPC Land relays the contributions by Glenn Gabe and Mic King without linking them. Each of these is documented through the reporting article, not through the original.
As of: 31.08.2026. All information without guarantee. Despite careful research, no warranty is given as to topicality or completeness. Technical and procedural details reflect the state documented as of the date given and may change. Figures on effort and rollout come from the respective tool providers about their own infrastructure. The account of the court proceedings follows the order of 20.07.2026 in Google LLC v. SerpApi LLC, case no. 4:25-cv-10826-YGR, and is not legal advice. The remarks relate exclusively to the order of a US federal district court of 20.07.2026 and permit no general statement about the lawfulness of reading search results under US, German, Swiss or any other law. The providers named were selected editorially. All brand and product names mentioned are the property of their respective owners.
