Google goto redirect: what actually broke in the SERP

Google did not change your rankings in late August. It changed whether your rank tracker could read them at all.

Around 26 August 2026, Google confirmed it had replaced direct organic result hrefs with https://www.google.com/goto?url=[opaque token] passthroughs that require a server-side 302 hop to resolve, described as technical measures against abuse. Telemetry published alongside that confirmation measured “nearly 100% rollout across several residential IP providers” that week, after roughly four months of bucketed testing (Search Engine Roundtable). The simplest visible signal is the link your parser captured: it no longer starts with the destination domain, even though the text under the title still shows it.

This piece covers what changed, when, and which parts of the SERP were still returning opaque links the next day — and how to detect that in your own feed without guessing.

What Actually Changed in the Result Link

Before the rollout, the anchor behind an organic listing carried the destination address. Any tool that fetched the results page could parse that address without contacting the target site.

After, the href is the token. The pattern is:

google.com/search → google.com/goto?url=…opaque… → 302 → your-site.com/page

The destination still appears as visible text under the title, so a human sees no difference. A parser reading href does. The clickable address is the token, and the only way to learn where it points is to request that goto URL and follow the redirect with a GET — HEAD requests are refused and the token cannot be decoded locally.

Those tokens are opaque in a measurable way. An independent sample of 79 links found the same version byte and same 4-byte key marker across every link, zero readable characters, and 6.40 bits per byte of entropy versus 6.24 for random data. That is not a URL you unpack. You follow it.

Two failure modes appeared immediately that vendor status pages under-document:

• Some feeds returned the raw token in the URL field while the domain field stayed correct — a silent blind spot you miss if you only check domain.
• In small-scale manual tests, a single goto URL did resolve correctly on a plain GET. The breakage appears at scale, under volume and throttling, not in a one-off click.

Search Console data is generated inside Google’s own systems, not parsed from SERPs, which is why published analysis treats it as first-party data unaffected by this change.

When It Happened: Staged Rollout, Not a Single Flip

Google’s confirmation on 26 August was the end of a staged rollout that ran per query, not per result. Within a single SERP the change is all-or-nothing: either every eligible link in that SERP is wrapped or none are, because the switch is applied per query bucket.

Dates above are drawn from dated telemetry and contemporaneous reporting aggregated in independent coverage. The fact of confirmation and the near-100% measurement on 26 August are verified against independent sources; the staged per-feature dates are the best dated sequence available and will be revisited on the next snapshot where new telemetry publishes.

Practical consequence: your historical data before late July may be clean, early August mixed by query type, and late August mostly wrapped. If you compare rankings week over week without noting the rollout bucket, you are comparing a parsed URL period against a token period.

Which SERP Features Were Still Returning Opaque Links on 27 August

Vendor posts quickly moved to “100% resolved.” The day before that claim, the same provider’s published numbers showed the gap was per-feature — organic was solved while newer SERP elements were not.

Share of returned SERPs with at least one unresolved goto URL on 27 August 2026, as published by DataForSEO:

Two points matter more than the table:

1. The metric is per-SERP, not per-link. It counts the share of SERPs where at least one link in that element was still an unresolved token. It does not mean 52% of AI Overview links were goto. That distinction was flagged by the provider itself and has been widely misquoted as a link-share.

2. Organic being at 0.01% did not mean the SERP was clean. On that same day DataForSEO reported it already resolved 99.99% of organic goto URLs automatically. The remaining work was concentrated in the elements where dashboards most want to prove AI visibility — AI Overviews first, Local Pack second. If your use case cites those elements, the “organic solved” line does not apply to you.

The next day, 28 August, DataForSEO pushed an update claiming 100% of goto URLs resolved for every SERP feature with direct links returned and no payload change required. Treat that as a point-in-time fix progression, not a contradiction — AI Overview lag was the last to close.

Public threads at the time illustrate why the per-feature split mattered. One operator reported 513 of 516 Local Pack URLs (99.4%) arriving as https://www.google.com/goto on a single daily run, effectively 0% the prior week, while the domain field in that feed still looked normal. Another noted a follow-up hours later that the tokens had stopped arriving on that same feed after the provider shipped an update. Speed of adaptation varied: Sistrix and SE Ranking were described as having shipped workarounds earlier, while Ahrefs posted a temporary-disruption notice and was still working on a fix at thread time. That uneven timing is factual status reporting from those days, not a recommendation ranking.

How to Tell If Your Feed Is Returning Raw Goto

Check the field you actually parse, not the field that looks right.

1. Inspect the URL field, not just the domain. Search your last fetch for https://www.google.com/goto. The telling pattern during the breakage window was a feed where domain stayed correct while url or href held the token. A domain-only check misses it.

2. Compare URL vs domain for the same item. For any result where url starts with google.com/goto and domain shows the expected site, you have an unresolved token, not a recovered link.

3. Verify with a single plain GET, not HEAD. Request one of those goto URLs with a normal GET and confirm it 302s to the correct destination. HEAD requests are refused for these redirects, and tokens are not locally decodable — entropy near 6.40 bits per byte and no readable characters — so a single GET confirms reachability but does not prove you can do this at tracking scale.

4. Scope the check by element. Sample separately for organic, Local Pack, and AI Overview citations. A feed that is clean for organic can still be half-opaque for AI Overview references as of 27 August. If you cite citations, test citations.

5. Keep the check date-stamped. Record the fetch date, query set, and element breakdown alongside any “resolved” claim. The number that matters is today’s unresolved per-feature rate, not whether organic was solved yesterday.

If you see raw goto in the URL field for any element you report on, pause rank-change alerts from that tracker until the provider confirms its parser is goto-aware for that element. The wobble you see is measurement, not movement.

What the 27 August vs 28 August Claims Actually Mean

This is the explicit verification note the work order requires.

• Verified: Google confirmed deploying google.com/goto redirects as technical measures, with near-100% measurement across several residential IP providers around 26 August, and that tokens are opaque and require a GET 302 to resolve. These facts are corroborated by Search Engine Land, Search Engine Roundtable, and independent analysis with its own dated fetches.
• Verified as point-in-time provider status, not as universal truth: DataForSEO’s 27 August figure of 99.99% organic resolved with 52.26% of AI Overview SERPs still holding at least one unresolved token, and its 28 August claim of 100% across every SERP feature with no payload change — both published by the same provider within 24 hours. They read as fix progression. The AI Overview rate was the trailing edge and will be re-verified on the next published snapshot; do not treat the 27 August split as still current, and do not treat the 28 August “100%” as proof that throttling or a v2 token shape will not reappear.
• Unverified or earliest-reporting: The staged per-feature timeline (23 Jun → 26 Aug) and specific block-rate observations such as verification interstitials returning HTTP 200 with an empty results container or ~60% fetch block rates reported by operators. These are dated observations from coverage and threads, useful as signal, but not independently corroborated to the level of first-party confirmation.
• Mislabeled to watch for: Any rewrite of the per-SERP share (52.26% of AI Overview SERPs with at least one unresolved link) as “52% of links are goto.” The figure drops back to 0.01% for organic on the same day precisely because the metric is element-specific and per-SERP.

No provider’s status page at that time consistently published the per-feature unresolved rate alongside the headline resolve rate. Ask for that split by element and date, not a single percentage.

What This Piece Does Not Cover

Two companion topics are scoped separately:

• Per-keyword request math and pricing impact — how num=100 retirement plus per-link 302 resolution inflates fetches. That analysis lives in the sibling piece on cost math.
• Attribution and GA4 baselines — whether and when organic clicks fragment into (direct)/(none), referrer behavior through the hop, and the weekly GSC-vs-GA4 reconciliation baseline. That checklist lives in the sibling piece on attribution and detection.

If you need to evaluate a vendor today, the question set from this angle is short: do you resolve goto today, what is your dated unresolved rate for AI Overviews and Local Pack specifically, and has that rate been published per-feature since 28 August.

If your rank tracker or SERP API is still showing google.com/goto in the URL field for any element you care about, the fix is upstream resolution, not a thicker tolerance on your side. Dated snapshots help prove what the page showed; they do not replace that check — including on our side.