What Google actually confirmed
A Google result link used to point at the destination page. Hover over one now and you will often see google.com/goto?url=… instead, with the real address encoded inside. Click it and Google passes you through to the same page you were always going to reach.
Barry Schwartz reported Google's confirmation on August 26 at both Search Engine Roundtable and Search Engine Land. Google's statement was one sentence, and it named nobody:
"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."
The rollout was not sudden. It surfaced as a single post on X on June 23, was documented by Search Engine Roundtable on July 8 — when Schwartz noted he could not reproduce it himself — and reached near-total coverage by the day of the confirmation.
Timeline compiled from PPC Land, Search Engine Roundtable and Search Engine Land.
One technical detail carries most of the consequences: the encoded destination cannot be read without asking Google. SerpApi describes the payload as protobuf-based, Google's own compact serialisation format. Perkins reports the links cannot be decoded locally, and that Google blocks HEAD requests, so a tool cannot cheaply peek at where a link goes. It has to follow it, once per link, with a full GET.
Three things that did not change
Most of the panic this week came from collapsing "Google changed the links" into "Google changed search." Before the useful analysis, it is worth being precise about what is still exactly where you left it.
Rankings and indexing
This rewrites an href on the results page. It does not touch retrieval, ranking or the index. No credible report has tied /goto to position movement.
Search Console
GSC counts clicks and impressions inside Google, before any redirect matters. Search Engine Land reported it as unimpacted.
The searcher's journey
People land on the same page as before. The visible URL still appears under the snippet title; only the browser's hover preview changes.
The one-line version
Nothing about your website changed. Something about the instruments pointed at Google changed. Those are different problems with different owners, and only one of them is yours.
What did change: the price of looking
A rank tracker's job used to be simple arithmetic. Fetch a results page, read the URLs written in the HTML, match them against the domains you care about, record positions. The destination was right there in the markup, free.
Now it is not. To know whether position four belongs to you or a competitor, a tool has to follow position four's redirect. Then position five's. Then every SERP feature link on the page. A results page carries hundreds of links.
Perkins put a number on it: resolving all links for a single five-page ranking takes between 500 and 1,000 requests. That is the cost of one keyword, on one day, in one location.
Sources: PPC Land on the /goto request cost; Search Engine Land on the num=100 removal.
Vendors are already showing their hands, and their answers differ. DemandSphere published a status note on August 27 saying organic ranking data was unaffected and that fixes for SERP feature URLs were rolling out. SerpApi called resolution "in progress" across its Google Search, Search Light, News Light, Videos and AI Overview APIs, telling customers no integration changes were required. Nozzle discarded affected pages during the testing phase — a workable tactic at partial coverage, and an untenable one at 100%.
The quiet risk in your reports
DemandSphere spelled out what happens to a tool that does nothing: every destination URL resolves to google.com, domain matching stops recognising your own pages, share-of-voice metrics compute against the wrong denominator, and historical data breaks at the rollout date. That failure is not loud. It looks like a chart.
Which means the dangerous scenario is not your tracker going down. It is your tracker staying up, returning numbers that look plausible, while quietly measuring something different than it did on August 25.
Strike three in twelve months
Read /goto on its own and it looks like a curiosity. Read it in sequence and it looks like a policy.
| When | Change | Effect on outside measurement |
|---|---|---|
| Jan 2025 | JavaScript required to render results | Cheap HTML fetching stops working; scrapers need a browser runtime |
| Sep 2025 | &num=100 parameter removed | One request for 100 results becomes ten; reported API costs rise about 10x |
| Aug 2026 | google.com/goto passthrough links | Every destination URL now costs a request; 500–1,000 per five-page ranking |
The middle entry is the one with settled consequences. When Google removed &num=100 in September 2025 — calling it "not an officially supported feature" — SEO Tyler Gargula's analysis of 319 properties found 87.7% lost Search Console impressions and 77.6% lost unique ranking terms. Those impressions had been rank-tracking bots all along. A year of reporting had been partly measuring the measurers.
The two changes are structurally different, and that difference matters. Removing &num=100 multiplied the number of pages a tool needed per query. The /goto redirect multiplies the number of requests per page. They compound.
There is also a legal shadow over the timing. On July 20, 2026, Chief US District Judge Yvonne Gonzalez Rogers dismissed Google's DMCA claims against SerpApi, holding that URLs, snippets and factual index data on a results page are not works protected by copyright. Google was given 21 days to amend the narrower claims involving copyrighted content, and said it would. Three weeks later, /goto went from intermittent test to near-total rollout.
Correlation, not causation
Google has not linked the rollout to the litigation, and the first sighting predates the ruling by four weeks. But the strategic reading is hard to avoid: a court said Google could not make reading its results unlawful, and Google can still make it expensive. Engineering has fewer procedural constraints than litigation.
The attribution question, answered properly
The loudest claim circulating this week is that /goto will strip referrers and dump your organic traffic into (direct) / (none) in GA4. It is the most consequential version of the story and the least supported one. Here is the actual mechanism.
GA4 classifies a session as organic search when the HTTP referrer matches a known search engine and no paid parameters are present. Google Search has long sent only the origin — https://www.google.com/ — rather than the full query URL. The /goto hop lives on that same origin. A redirect through it still arrives at your server carrying a google.com referrer, and still maps to google / organic.
The breakage scenario requires Google to actively suppress the referrer on that hop. That would also blind Google Analytics itself, which is not an obviously attractive outcome for Google. As of August 31 there is no confirmed report of a wave of organic-to-direct reclassification, and Search Console — which never depended on the referrer — is reported unaffected.
That is a reason to verify, not to relax. The check takes ten minutes and it beats an argument on X.
| Check | Where | A real problem looks like |
|---|---|---|
| Organic vs direct ratio | GA4, 90 days daily | A step change on or after Aug 26, with direct rising as organic falls by a matching amount |
| GSC clicks vs GA4 organic sessions | Both, same date range | The two series diverging in one direction, GSC holding while GA4 organic drops |
| Referrer values in server logs | Access logs or your CDN | Empty referrers on search-landing hits, or a google.com/goto referrer you do not have a rule for |
| Redirect behaviour on a live result | A redirect checker or header checker | A referrer policy on the hop that would prevent your analytics from seeing google.com |
Do not pre-emptively "fix" this
Adding referral exclusions, rewriting channel groupings or bolting UTM parameters onto organic listings in anticipation of a problem you have not observed will corrupt clean data to solve a hypothetical. Baseline first. Change only what your own numbers tell you to change.
What to actually do this week
None of this requires a site change. All of it requires being honest about which of your numbers you can still defend in a meeting.
- 1
Annotate August 26 on every rank-tracking chart
Treat it as a methodology boundary, the same way you would a tracking migration. Comparing September positions against July positions across a vendor's patch window is comparing two different measurements with one label.
- 2
Ask your vendor three specific questions
When did you start resolving /goto links? Which SERP features are still unresolved? Did you drop, estimate or backfill any data between June and August? Vague reassurance is not an answer — DemandSphere and SerpApi both published specifics, so the bar exists.
- 3
Move your source of truth to first-party data
Search Console clicks, impressions and average position come from inside Google and survive every change on this list. Third-party rank data is directional context. If your monthly report leads with a tracked position average, you have the hierarchy backwards. Our guide to finding opportunities in existing Search Console data is a better starting point than a position chart.
- 4
Spot-check positions by hand for the queries that matter
For the ten or twenty keywords that actually drive revenue, a manual or single-query check is cheap and unmediated. A SERP checker gives you a live read on what a results page holds today, which is worth more than a dashboard average built on a pipeline in flux.
- 5
Re-baseline your analytics before anyone panics
Pull the 90-day organic-versus-direct series now and save it. If a genuine attribution shift appears in September, you will want a pre-rollout baseline you captured yourself rather than a screenshot from someone else's thread.
What is not on this list
Changing your content, your internal links, your schema or your technical setup. There is no on-site optimisation for a change to how Google writes its own hrefs. Anyone selling you a "/goto readiness audit" is selling you a spreadsheet.
The part the AI visibility crowd should be reading
AEO and GEO tooling grew fast on a comfortable assumption: that search results and AI answers would stay cheaply readable from the outside. A large share of AI visibility products are, underneath the dashboard, SERP scraping plus prompt sampling.
SerpApi's own status note listed its AI Overview API among the affected products. That is the pipe a lot of "are we cited in AI Overviews?" reporting runs through. If citation URLs inside AI surfaces need resolving one redirect at a time, the same cost multiplier lands on the newest category in the industry — the one whose pricing was set before the multiplier existed.
Expect the same three vendor responses seen after the num=100 removal: raise prices, absorb margin, or quietly reduce sampling depth and frequency. The third is the one to watch for, because it degrades your data without appearing on an invoice.
The strategic read
Every measurement layer you do not own can be repriced by a platform without warning. The parts of your reporting that survived January 2025, September 2025 and August 2026 have one thing in common: they come from your own logs, your own analytics, and your own Search Console property. Weight them accordingly.
What to watch next
Surface expansion
Coverage has already spread past organic results into Videos, Lens and AI Overviews. Watch whether News, Images and Discover follow.
Google's amended complaint
Google said it would refile the narrower claims against SerpApi. The legal and technical tracks are running in parallel, and both shape what SERP data costs.
Vendor pricing notices
After num=100, price and depth changes followed within weeks. Read your tracker's changelog and billing emails through September.
A genuine referrer change
If Google ever tightens the referrer policy on the /goto hop, that would be a real attribution event. It has not happened. Watch your own data, not the discourse.
Bottom line
Google's /goto rollout is a real change and a badly framed one. It does not touch rankings, indexing, Search Console or the page a searcher reaches. It makes reading Google from the outside dramatically more expensive, and it is the third such change in twelve months.
The lesson is not that rank tracking is dead. It is that rank tracking is a leased instrument, and the landlord has raised the rent three times. Plan your reporting around the data you own.
The move this week
- Annotate August 26 on every rank chart, and stop comparing straight across it.
- Save a 90-day organic-versus-direct baseline today, before you need it.
- Send your rank vendor the three questions above and keep the reply.
- Change nothing on your site. There is nothing here to optimise for.
Frequently asked questions
Sources and next checks
Google confirms the /goto rollout
Search Engine Land, August 26: the confirmation, the statement, and the Search Console note.
The 500–1,000 request figure
PPC Land, with Derek Perkins on local decoding, blocked HEAD requests and the rollout timeline.
A vendor status note worth copying
DemandSphere on what breaks without countermeasures, and what they fixed.
The SerpApi ruling
July 20: why URLs, snippets and index facts were held not to be copyrighted works.