Dense array of identical brass funnel mouths in a charcoal plate, every opening leading into darkness

By Mark Buraga, Independent SEO Consultant, Growth Engine PH. Last updated 3 September 2026.

On 26 August 2026, Google confirmed it is rewriting the clickable links in its search results to pass through a Google-owned redirect at google.com/goto. The change had been in testing for roughly two months before the confirmation and reached near-complete rollout across residential IP ranges by late August. It is a link-layer change, not a ranking change: no algorithm moved, no site gained or lost position, and Search Console is unaffected.

What it does affect is measurement. On our own AI visibility tracking, the change turned four weeks of stable data into an apparent total collapse in one Monday, and the collapse was not real. This post covers what changed, what we recorded first-hand, and what practitioners should do about it. It also corrects the causal story now circulating about why Google did this, which the public timeline does not support.

What is the google.com/goto redirect?

section-1-what-is-the-goto-redirect

The google.com/goto redirect is an intermediary URL that Google now places between a search result link and its destination. Where a result link previously pointed at the destination site, the clickable link is now a Google URL carrying an encoded token, and Google resolves that token to the real page when the link is clicked.

Google’s confirmation, given to Search Engine Land on 26 August 2026, was brief: “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” (Search Engine Land, 2026-08-26).

For a person using Google, almost nothing changes. The redirect resolves in a fraction of a second and lands on the correct page. On classic results pages the destination domain still appears as text beneath the result title, so a reader can still see where a link goes before clicking it. The one visible difference is the browser status bar, which now shows the Google redirect rather than the destination on hover.

For anything reading the page programmatically, the change is significant, and the reason is the shape of the token.

The destination URL is not inside the link

This is the detail that determines how much work the change creates, and it is the part most coverage has skipped.

Google has used redirect links before. The older google.com/url?q= format carried the destination in plain text as a query parameter, so anything reading the page could strip the wrapper and recover the target with a line of string handling.

The /goto format does not work that way. Captured from a live Google AI Mode result on 3 September 2026, the links take this shape:

https://www.google.com/goto?url=CAESUwHrOzAVzZrDUgUAnV5h2VixmvHQCdNi7-iykW8nCNebU4pawL3mS32Ce2J6jtrS35pwoIJ2_SFfG1mjj6ZUHhKhBRIOXgtGbyR7pi1KpZa89InZ

The value after url= is an opaque encoded token, not an encoded URL. There is no destination in it to extract. The only ways to learn where that link goes are to ask Google by following the redirect, which is one network request per link and is reported to block HEAD requests, or to read the destination domain from the visible text on the page instead of from the link.

That distinction is the whole story for tooling. Recovering a plaintext target is a trivial fix. Resolving an opaque token is a per-link network round trip, at a reported cost of 500 to 1,000 additional requests for a five-page ranking report, against an endpoint Google controls, can rate-limit, and can fingerprint.

What the change does not do

Precision matters here, because the change has been over-read in places.

It does not alter rankings, and no site’s position changed because of it. It does not affect Search Console, which reports from Google’s own systems rather than from the results page. It does not stop a human from seeing a destination before clicking on classic results pages, where the domain is still printed under the title. It is not a penalty, not an update in the core or spam sense, and there is nothing on a website to fix in response to it.

Did Google deploy /goto because it lost the SerpApi case?

section-2-two-parallel-arms

No, and the timeline is the reason. This causal story is circulating widely in SEO commentary, and it does not survive a check of the dates.

The case is real. Google sued SerpApi, which scrapes Google results and resells them through an API, on 19 December 2024. The claims ran under Section 1201 of the DMCA, the anti-circumvention provision, alleging SerpApi had bypassed an anti-scraping system Google calls SearchGuard. On 20 July 2026, Judge Yvonne Gonzalez Rogers of the U.S. District Court for the Northern District of California dismissed the complaint in full (Search Engine Land, 2026-07-21).

The reasoning is the part worth understanding. Section 1201 protects technological measures that control access to a copyrighted work. The court found that plain and aggregated search results, meaning URLs, snippets and factual index data, are publicly accessible facts rather than works protected under the Copyright Act. Where there is no protected work, there is nothing for an access control to protect, and the anti-circumvention claim has nothing to attach to. Those claims were dismissed permanently. Claims touching genuinely copyrighted components, such as licensed images inside Knowledge Panels, were dismissed with leave to amend, on the condition that Google plead facts showing the copyright owners had authorized SearchGuard’s deployment (Search Engine Journal, 2026-07-21).

Now the dates. The /goto format was first observed in the wild in late June 2026 and, per Search Engine Land’s reporting, had been in testing for approximately two months before the 26 August confirmation. The dismissal landed on 20 July. The deployment predates the ruling it is supposed to be a response to.

The more defensible reading is that these are two arms of one long-running strategy rather than cause and effect. Google has been working the legal route and the technical route in parallel, and the technical route was already in the field when the legal route stalled.

The case is not over, and it is now about Google’s contracts

The other half of the circulating story is that Google lost and that was that. It refiled.

On 10 August 2026, Google filed a 15-page amended complaint reasserting two DMCA claims under 17 U.S.C. Sections 1201(a)(1)(A) and 1201(a)(2). To meet the court’s condition, it pleaded its confidential content licensing agreements as the factual core: a licensing agreement with Reddit, an unnamed partner holding copyrights to millions of works under an agreement dating to 2017, and another content provider that requires download prevention. SerpApi moved to dismiss a second time on 25 August 2026 (Search Engine Roundtable, 2026-08-11; PPC Land, 2026).

The strategic cost there is real, if not the one usually described. To keep the case alive, Google has had to identify licensing counterparties and put confidential agreements on a path toward discovery. That is a meaningful concession to make in order to preserve a claim, and it tells you how much Google wants this precedent.

For practitioners the practical position is simple. Scraping Google is not settled as unlawful, the public-facts holding stands, and the live dispute has narrowed to licensed content inside results rather than results as a whole. Meanwhile the technical friction arrived regardless of how the case ends. That asymmetry is the lesson: the legal question was slow, contested and partial, and the engineering answer shipped globally in about two months.

What did the Google goto redirect break in our own tracking?

section-3-what-it-broke-in-our-tracking

We track AI visibility weekly for every retainer client and for ourselves, using a fixed set of 15 prompts captured on the same weekday through Google AI Mode. The prompt set is locked at the start of an engagement and never edited, which is what makes week-over-week comparison meaningful. That discipline is why we can date this precisely.

Here is our own capture record across five consecutive Mondays, recomputed from the stored capture files on 3 September 2026.

Capture weekSource URLs recordedPrompts citing our siteSource count reported
3 August 20261976 of 15230
10 August 20262156 of 15243
17 August 20261966 of 15189
24 August 20262096 of 15236
31 August 202600 of 1548

Read as a report, that last row says our AI citations went to zero in a week. Every prompt, no sources, no brand presence, after four weeks holding steady at six of fifteen.

It is not true. Our citations did not move. Note the shape of the four clean weeks: the source count moved between 189 and 243 while our own presence held at exactly six of fifteen every week, which is the same pattern we measured across the summer when we published our own error bar. The base is noisy. Our presence was not. That is what makes the fifth row so obviously an instrument fault rather than a result.

The 31 August captures are stored with screenshots, per our standing method, and the screenshots show what actually happened. The pages rendered normally. None was flagged as blocked, none hit a consent or CAPTCHA wall, all fifteen held the AI Mode surface, and each stored a real page title. On the first prompt, the source panel is present, reads “5 sites,” and lists our own domain as the second source. We were cited. Our instrument recorded that we were not.

The cause is the one described above. Our capture reads the links on the rendered page and discards anything hosted on a Google domain, on the reasonable assumption that Google-hosted links are navigation and policy chrome rather than citations. Once every citation became a google.com/goto link, every citation looked like chrome, and the filter threw all of them away. The source count in the final column survived only because it is read from the answer’s own “N sites” text rather than from the links, which is why it degraded rather than zeroing.

A live probe on 3 September confirmed the mechanism rather than leaving it inferred. On a single AI Mode result, 28 links were present, 20 of them on www.google.com, and 11 carried the /goto wrapper with the opaque token shown earlier. No citation on the page exposed its destination in its link.

Two things about this are worth sitting with.

First, it failed silently. Nothing errored. No alert fired. The weekly job ran, completed, wrote a well-formed record, and reported a number that was wrong in the most alarming possible direction. Had we read that number as performance, the honest conclusion available from the data alone would have been that something catastrophic happened to our AI visibility in the last week of August.

Second, the only reason we caught it is a rule that exists for an unrelated reason. Our method requires a screenshot of every captured answer, which we adopted so that a strange week could be inspected rather than assumed broken. That rule, written for a different problem, is what separated a tooling fault from a client-facing fiction here.

We are reporting one site’s break honestly rather than generalizing from it. We cannot tell you how any given commercial tool handles this, and we are not going to guess. What we can tell you is the failure signature to look for, because it will look the same wherever it happens: counts that fall toward zero rather than moving, arriving in one step on a single date, on a set of pages that render perfectly when you look at them.

What should SEO consultants do about the goto redirect?

Where this lands in your work is the gap between a tool’s output and a client’s inbox. You are the last check before a number becomes a claim, and this change manufactures a number that reads as a crisis.

The lesson is that an instrument change and a performance change are indistinguishable in a chart. They separate only against a date and a screenshot. A drop that begins precisely on a known platform change date, arrives in a single step rather than a slope, and hits every tracked query at once is a measurement artifact until proven otherwise. Real performance loss is ragged, partial, and rarely punctual.

Do this now:

  • Pull your AI visibility and citation data for the weeks either side of 31 August 2026 and look specifically for a one-step fall to zero or near-zero. Treat that shape as suspect before you treat it as a finding.
  • If you run any in-house capture that reads links off a results page, check whether it filters Google-hosted domains. If it does, it is discarding every citation right now.
  • Fix it by reading the destination domain from the visible text of the source panel rather than from the link. The domain is still printed on the page. Resolving the redirect per link works too, but it costs a network request each and can be rate-limited, so treat it as the fallback.
  • Add the capture date of any platform change to your reporting notes, so that the next person reading the chart has the instrument history beside the numbers.

What should agencies do about the goto redirect?

The portfolio problem is different from the single-client problem: the artifact hits every client at once, on the same date, and it will reach some of your clients through their own dashboards before it reaches you.

The lesson is that a synchronized drop across an entire book is diagnostic in itself. Independent client sites do not lose AI visibility simultaneously on a Monday. When every account moves together, the common factor is the instrument, not the accounts.

Do this now:

  • Run the one-step-to-zero check across the whole portfolio at once rather than client by client. If it is present everywhere, you have your answer in one pass and you have it before the first client email.
  • Decide the client communication line before you are asked, and keep it short: a Google link-format change on 26 August affected how tracking tools read results pages, it did not affect the client’s site, rankings or Search Console, and here is the corrected figure.
  • Hold the corrected number, not the raw one, in any report covering the affected weeks, and say in the report that the week was re-measured. A footnote costs you nothing. A retracted headline number costs trust.
  • If a report already went out with the artifact in it, correct it proactively. The client will eventually read about /goto somewhere, and a correction you sent first is a competence signal while one you sent second is not.
  • Ask your vendors directly how they are resolving /goto and what it does to their refresh cadence and cost. This is a reasonable question and the answer tells you what to expect at renewal.

What should in-house marketing managers change?

Your exposure is upward. You reported an AI visibility figure to people who will not read this post, and the version they remember is the one in the deck.

The lesson is that the reliability of a metric is part of the metric. A number sourced from a third party reading a platform’s public pages carries a dependency that a number sourced from your own property does not, and that difference belongs in how you present it.

Do this now:

  • Check any AI visibility or citation figure you reported for late August and early September against the dates above before it gets quoted again in a quarterly review.
  • Put Search Console at the center of your AI reporting rather than at the edge. It reports from Google’s own systems, no redirect sits between you and it, and it is the one source in this stack that this change cannot touch.
  • When you present a scraped metric, state its source and its known failure mode in one line. This is what protects you when the number moves for a reason that has nothing to do with your work.
  • Do not commission remediation work off a single-week drop. Require the confirmation rule below before anyone spends budget responding to a fall.

The action checklist

  • Pull AI visibility and citation data for the weeks either side of 31 August 2026 and look for a one-step fall to zero or near-zero.
  • Treat any drop that is punctual, total and synchronized across unrelated properties as a measurement artifact until a screenshot proves otherwise.
  • Audit any in-house capture script for a filter that discards Google-hosted links. That filter now discards every citation.
  • Read citation domains from the visible text of the source panel, not from the link, because the link no longer contains the destination.
  • Use redirect resolution only as a fallback, and budget one network request per link when you do.
  • Screenshot every captured answer, so a strange week can be checked instead of assumed.
  • Apply a confirmation rule: a single-run move is not a finding. Require the same result on a second capture before it enters a report.
  • Record platform change dates alongside the data, so the instrument history travels with the numbers.
  • Re-issue any client report covering the affected weeks with a corrected figure and a one-line note.
  • Ask SERP data vendors how they resolve /goto and what it does to cost and refresh cadence.
  • Weight Search Console more heavily in AI reporting, as the one source in this stack the change cannot reach.
  • Change nothing on the website. There is no on-page fix, because nothing on your site is implicated.

Our take: the moat moved from the index to the door

The strategic shape of this is more interesting than the mechanics.

For two decades the argument over search data was about ownership, and it was fought on whether results are a creative work or a set of facts. On 20 July a court gave the answer most observers expected: the plain, aggregated part is facts, and facts do not get copyright protection. That is a durable holding and it is a real constraint on Google.

It also turned out not to be the binding one. While the ownership question was being litigated, Google spent about two months shipping a change that made the data more expensive to reach, and shipped it worldwide. Nobody needed to establish who owns a search result to make it cost 500 requests to read a ranking report. The ownership question was slow and partial. The access question was an engineering ticket.

This is the pattern worth carrying forward: a platform that cannot make something unlawful can still make it uneconomic, and the second lever is faster, cheaper and beyond a court’s reach. Every measurement layer that sits on someone else’s public pages inherits that risk, and it is the same structural exposure whether the page belongs to Google, an AI answer engine, or a marketplace.

The practical consequence for AI search work is a reordering of what to trust. First-party data just became more valuable relative to scraped data, not because scraped data got less accurate, but because it got less certain. Search Console is a permissioned channel from Google to you. A scraped results page is a permissioned channel too, and the permission is the part that just changed.

None of this touches the underlying discipline. Being cited by an AI answer still comes from structural retrievability, and being named inside one still comes from consensus across the surfaces where opinion is formed. Those mechanisms are unchanged by a link format, and they are what we have argued throughout this cluster: that most dashboards still report the wrong half of AI search, that Reddit and LinkedIn dominate the sources AI answers actually reach for, and that rank and citation have come apart.

What changed on 26 August is not the work. It is the confidence interval on the reporting of the work, and that is worth more attention than it has had.

Frequently asked questions

What is the google.com/goto redirect?

It is an intermediary URL Google now places between a search result link and its destination. The clickable link points at Google and carries an encoded token, which Google resolves to the real page on click. Google confirmed the rollout on 26 August 2026 and described it as a technical measure against abuse.

Does the /goto redirect affect my rankings?

No. It is a change to how links are formatted on the results page, not to how results are ordered. No site gained or lost position because of it, and there is no on-page change to make in response. Search Console reporting is also unaffected, since it comes from Google’s systems rather than from the results page.

Why did Google deploy /goto redirects?

Google’s stated reason is protection against evolving forms of abuse. The practical effect is to make automated reading of search results more expensive, since the destination is no longer present in the link and must be resolved one request at a time. Google did not name scraping specifically in its confirmation.

Was /goto a response to losing the SerpApi lawsuit?

The timeline does not support that. The format was first observed in late June 2026 and had been in testing for roughly two months before the 26 August confirmation, while the dismissal came on 20 July 2026. The deployment predates the ruling, which suggests parallel strategies rather than retaliation.

Did Google lose the SerpApi case permanently?

No. Claims over results without copyrighted content were dismissed permanently on the basis that search results are public facts. Claims over copyrighted components were dismissed with leave to amend, and Google filed an amended complaint on 10 August 2026 pleading its content licensing agreements. SerpApi moved to dismiss again on 25 August 2026, so the case is live.

How do I know if my AI citation tracking broke?

Look for a fall to zero or near-zero that arrives in a single step on one date rather than as a slope, and that hits every tracked query at once. Then open a stored screenshot of an affected capture. If the page rendered normally and sources are visible on it, the instrument broke and the performance did not.

How do I fix a capture script that /goto broke?

Read the destination domain from the visible text of the source panel rather than from the link, because the link no longer contains a recoverable destination. Following the redirect also works but costs one network request per link and can be rate-limited, so use it as a fallback rather than a default.

Does /goto affect analytics attribution?

Google has not stated any change to how organic traffic is attributed, and we have not measured one, so we are not asserting either way. It is worth checking your own analytics for unexplained movement between organic, referral and direct across late August, and treating anything you find as a question rather than a conclusion until it is confirmed.

Sources

This post is part of our Google Updates desk, where every confirmed Google change gets one practitioner-grade analysis. Growth Engine PH tracks AI citations and mentions weekly as a standard deliverable on every retainer tier. If you want to know what your own error bar looks like before you act on a number, let’s talk or see how we work.