Google Is Degrading Third-Party Search Data. Here's How We Kept Ours Accurate.

Since mid-September, a meaningful share of the search data the SEO industry relies on has been unreliable. Google started returning altered results pages to automated requests, and it now routes clicks on organic results through a redirect. Neither change affects how sites rank. Both affect whether anyone outside Google can measure it accurately.

For an agency, this isn't an abstract problem. Our research, deliverables, and client reporting all depend on knowing what actually ranks. If that input is wrong, every recommendation built on it carries the error forward.

What changed

The first change is what the industry has started calling poisoned results. Google serves complete, normal-looking results pages with the listings swapped, and nothing on the page signals that anything is wrong. Nozzle measured about 8% of the pages it collected after September 13 as poisoned, compared with 0.17% before, with a peak near 27% at the end of September. AccuRanker, using a different method, measured much lower rates of around 1.5% in early October. Semrush and SE Ranking have both acknowledged the issue to their customers. Google hasn't commented publicly.

The second is the /goto redirect. Clicks on organic results now pass through a google.com/goto address before reaching the destination site. Google confirmed the rollout in late August, describing it only as part of "a long history of deploying technical measures against evolving forms of abuse." Each link now has to be resolved individually. Since late October, Google has also throttled clients that open many of these links in quick succession, and some tools have begun reporting google.com as the ranking domain.

There's no evidence that either change affects actual rankings or Search Console data. What they affect is visibility into those rankings from outside Google (AHREFs, SEMRush, etc).

Our view

Google has legitimate reasons to limit automated access to its results. Large-scale scraping to train models or build competing products is a real concern, and scraping falls outside Google's terms of service. Rate limits, IP blocks and CAPTCHAs are reasonable responses, and they share one important property: the requester knows it has been refused.

Serving altered results without any signal is different. It doesn't stop data collection. It corrupts it, and the cost falls mostly on people using that data for ordinary work: in-house teams, agencies and the businesses they advise. A false ranking drop can lead to unnecessary rewrites or misdirected budget, and the error may never be traced back to its source.

The redirect is less concerning. It adds cost and latency to research, and searchers can no longer preview a destination by hovering over a result. But it doesn't misrepresent anything, and that makes it a technical constraint we can engineer around.

What this means for brands

Any ranking report built on scraped data now carries more noise than it did a few months ago. Before acting on a reported drop, confirm it in Search Console.

It's also reasonable to ask your agency or tool provider where its search data comes from and how it's validated. The answer should be specific. If your reports have shifted unexpectedly since September, that review is worth doing before any content or strategy changes are made on the strength of them.

How we responded

We chose to build our own research tooling instead of waiting for vendors to adapt.

Much of our content work relies on correlation analysis. We measure what the top-ranking pages for a keyword have in common and turn that into specific targets for writers. The desktop software we had used for this was affected by both changes, and it had long been resource-intensive enough to disrupt the machines it ran on.

We replaced it with an internal tool. Without going into specifics, it resolves each result to the actual ranking page, draws on SERP data sources we've validated for accuracy, and measures every page under identical conditions. It also corrects several long-standing measurement issues in the software it replaced. It's built directly into our content outline and content refresh workflows, so the research arrives inside the deliverable rather than as a separate step.

Our reporting follows the same principle. Search Console is the reference point, and third-party ranking data is checked against it before it appears in a client report.

Continuous testing

We hold our own tools to the same standard of skepticism. Google's countermeasures are still evolving: AccuRanker reported a further tightening of detection in mid-October, primarily on mobile. A data source that's accurate today may not be next month.

We regularly benchmark SERP data providers and rank trackers against one another and against Search Console, and we revise our tooling when the results shift. Every release is reviewed by someone other than its author before it reaches the wider team.

I expect this to be an ongoing discipline rather than a problem that gets solved once, and we've organized our work around that assumption.

‍

Sources

• Google feeds rank trackers swapped results on 8% of pages, Nozzle finds (PPC Land)
• google.com/goto links in Google results: what changes (Outrank)

‍