Your rank tracker is quietly dropping pages

Google changed how a search result points at your site, and the damage landed somewhere nobody was watching. Since late August the outbound links in Google's results run through a passthrough address at google.com/goto rather than pointing at the destination directly. Users will never notice. Rank tracking data accuracy took the full hit, and the way it took the hit matters more than the fact that it did.

The failure mode here is not an outage. An outage is loud, and loud problems get fixed. This one shows up as a report that looks exactly like last month's report, built on a smaller and quieter sample, with no field in the export that tells you which pages were thrown away along the way.

What Google actually changed

Until this summer, a Google results page carried the destination URL for every listing in the markup. Anything reading that page could see where each result pointed without making a single additional network call. That is how rank trackers have worked for fifteen years, and it is why tracking a keyword cost roughly one request.

Now the href is an encoded passthrough on google.com/goto, and the encoding cannot be reversed on the machine doing the reading. To learn where a result actually points, a tool has to follow the link and let Google's 302 tell it. Alex Greenland reported the first public sighting on June 23, 2026. Brodie Clark documented the pattern across both organic and paid listings on July 2. By August 26 it was confirmed at near total rollout across residential IP providers, and Google confirmed the change the same day.

“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.”

That is the whole official explanation, given by a Google spokesperson. It is an anti-scraping measure aimed at commercial scrapers and at the AI companies buying from them. Read plainly, it is not a search feature and was never pitched as one. Your reporting stack is collateral, and collateral does not get a migration guide.

Why rank tracking data accuracy got worse overnight

Here is the part that should worry you, and it has nothing to do with cost. During testing, Nozzle discarded the pages it could not resolve. SERP Lens made the same call. Both decisions are defensible engineering: a result whose destination you cannot confirm is worse than no result, because a wrong attribution silently reassigns a ranking to the wrong domain.

But a discarded page does not arrive in your dashboard wearing a label. It arrives as a slightly shorter list. A keyword where you held position 8 last week now shows you absent, and absent reads as a loss. Your quarterly deck says visibility declined. The engineer at the vendor who wrote the discard rule is not in that meeting, and neither is the reason.

HEAD requests, the cheap way to resolve a redirect without downloading anything, are blocked. Only full GET requests that follow the redirect are honored. So the only path to an accurate destination is the expensive one, and the only path to a cheap answer is to guess or to drop. Vendors under margin pressure will make that tradeoff differently from each other, which means two tools tracking the same keyword set will now disagree in ways that have nothing to do with ranking.

The compounding cost nobody budgeted for

Derek Perkins, the chief executive at Nozzle, put a number on it: resolving all links for a single five page ranking takes between 500 and 1,000 requests. That is per keyword, per pull. It replaces a job that used to be one request and, since the num=100 retirement, has been closer to ten.

Run the illustrative math on a mid sized tracking set. Five thousand keywords pulled daily at the old rate is fifty thousand requests a day. The same five thousand keywords fully resolved at the low end of Perkins' range is two and a half million. That is not a pricing adjustment a vendor absorbs quietly for long, and the two ways out of it are both bad for you: charge more, or resolve less and say nothing.

Expect the second one more often than the first, because the second is invisible. Sampling is the default response to a cost spike in every measurement business, and sampling is fine as long as the sample rate ships with the number. Ask your vendor what theirs is. If the answer takes more than a day, that is also an answer.

August made the timing worse

None of this landed on a quiet month. Google ran a spam update from August 18 to August 21, with unconfirmed movement on either side of it. Practitioners logged ranking volatility spikes on August 5 and 6 and again on August 12 and 13. Search Console dropped performance data around August 12 and lost two days of crawl stats.

So a client asking why August looks strange has four candidate explanations available, and three of them are instrumentation. A spam update is a real ranking event and deserves a real answer. A vendor discard rule and a Search Console gap are not ranking events at all, and treating them as one is how a team spends a month rewriting content that was never the problem. We laid out the discipline for separating these in how to report on search visibility volatility without overreacting, and this month is the strongest argument for it we have seen.

The honest position with a client is that August is partially unmeasured and you can say precisely which parts. That is a better meeting than a confident narrative built on a sample nobody characterized. Clients forgive gaps they were told about. They do not forgive a story that quietly reverses next quarter.

How to audit your rank tracking data accuracy this week

This is a two hour job and it is worth doing before your next reporting cycle, not after it.

Do the same pass on paid reporting while you are in there. Brodie Clark documented the pattern across paid listings too, and competitive ad monitoring reads the results page the same way organic tracking does. Teams tend to audit the organic stack and assume the paid stack is fine because a different vendor supplies it. The mechanism does not care which team bought the tool.

What to measure instead

The strategic read is that position tracking has been getting less load bearing for two years and just got a hard shove. Zero click behavior already meant a ranking was a weaker proxy for traffic than it used to be. AI surfaces already meant a chunk of the visibility that matters never appears in a position report at all. Now the position report itself has a confirmed accuracy problem with no fix on the vendor side.

That does not make trackers useless. It makes them one input among several instead of the spine of the report. The spine should be first party: Search Console impressions and clicks, which come from Google's logs; on site conversions; and a citation measurement practice for the AI surfaces, which is where a growing share of the discovery you care about now happens. If you have not built that third leg yet, the readiness checklist for AI answer surfaces is the shortest route to a defensible baseline. We make the case for stating uncertainty in every one of those in the argument for error bars in AI search measurement, and the same rule applies to a position report that has quietly become an estimate.

One more thing worth saying plainly, because it will come up in a pitch soon. Vendors will start selling goto resolution as a premium feature, and the ones that resolve properly will be more expensive and will deserve to be. That is a reasonable thing to buy. What is not reasonable is paying the old price and assuming the old completeness, which is the position most reporting stacks are in right now by default. Reporting from a set of numbers you cannot characterize is the expensive option, whatever the invoice says.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.