Case Study 28.1 — "(not provided)": the day analytics went blind to the organic keyword

A real, public event. Facts are drawn from Google's own announcements and the well-documented industry response; where the historical record is approximate, this study says so and avoids inventing precise figures.

Background

For the first decade of SEO, web analytics tools showed you the exact keywords that brought each organic visit. When someone searched Google and clicked your result, the search terms rode along in the browser's referrer data, and analytics dutifully listed them: you could see that "emergency plumber" sent 40 visits and "how to bleed a radiator" sent 200. Keyword-level organic reporting was the bedrock of the craft. Practitioners optimized pages against the exact terms they saw converting, and a whole generation of SEO tooling was built on the assumption that this data would always be there.

Then, in October 2011, Google announced a change that would quietly demolish that assumption. Framing it as a privacy and security measure, Google began encrypting search — moving signed-in users to HTTPS (secure) search — so that a searcher's query would no longer be passed to the destination website in the referrer. In analytics reports, those visits stopped showing a keyword and started showing a now-infamous two-word placeholder: (not provided).

The measurement issue

At first, Google positioned the impact as limited — it would affect only signed-in users, and Google suggested the share of affected traffic would be modest. That framing did not hold. Over the next two to three years, Google progressively extended encrypted search to signed-out users and to searches made from the browser's address bar, until, by roughly 2013–2014, essentially all organic keyword data had collapsed into (not provided). The exact trajectory varied by site and region, and this study deliberately does not quote a precise percentage curve — the durable, well-documented fact is the direction: from a small slice to nearly the entire organic keyword picture, gone.

The change landed as a genuine shock to the industry, and it carried a controversy worth stating plainly as a matter of public record: while organic keyword data disappeared, advertisers buying Google's paid search (then AdWords) continued to receive keyword data for their paid clicks. To many SEOs this looked like a double standard — privacy invoked to withhold data from the unpaid channel while the paid channel kept it — and Google was publicly criticized for it. Google maintained the change was about protecting users' search data in transit. Both the criticism and Google's rationale are part of the documented history; this study takes no position beyond recording them.

What it shows

Three lessons from this event sit at the center of Chapter 28:

  • The platform decides what you can measure, and it can change the rules. Site owners did not lose keyword data through any failure of their own; a platform they don't control altered what it would share. Every measurement setup you build sits on top of decisions made by Google, Apple, browser makers, and regulators — and those decisions move. Measurement is contingent, not permanent.
  • There is a hard wall between the query and on-site behavior. After (not provided), no analytics tool could tie the exact words a person searched to what that specific person then did on your site. This is the precise limit Chapter 28 keeps returning to: GA4 knows a visit came from organic search and everything it did afterward — but not what was searched. That wall was built in 2011 and has never come down.
  • The answer was a second tool, not a workaround. Google's response to the outcry was to route aggregate query data — impressions, clicks, average position, per query — into Google Webmaster Tools, later renamed Google Search Console. The industry adapted not by recovering the old per-visit keyword in analytics (impossible) but by learning to reason across two tools: Search Console for the query and the click, analytics for the behavior and the conversion. That two-tool division of labor — the seam between Chapters 27 and 28 — is a direct descendant of this event.

Outcome

(not provided) is now simply how organic search works. A generation of SEOs has never known anything else, and the tooling ecosystem reorganized permanently around the split: Search Console became the canonical source of organic query data (aggregate, sampled, and capped, but query-level), while Google Analytics and its successors settled into their role as the post-click behavior-and-conversion tool. Paid tools (Chapter 30) built businesses partly on estimating the organic keywords GSC and analytics won't hand over — estimates, not the ground truth that vanished in 2011.

The lesson

Measurement in search is a privilege extended by platforms, not a property you own — and the single most important structural fact about SEO analytics, the wall between the query and the visitor's behavior, is the result of a deliberate platform decision made over a decade ago. The professional response is not to chase the lost data with tools that claim to have it back, but to build an honest, two-sided measurement practice (GSC + GA4) that respects the wall and reasons across it. It is the same posture the rest of the chapter teaches toward consent, modeling, and thresholding: know exactly what your instruments can and cannot see, and never report a number as if the limit weren't there.

Discussion questions

  1. Google justified (not provided) on privacy grounds while paid search kept keyword data. Evaluate that justification: is "protecting the query" consistent with continuing to share it with advertisers? What would an honest defense of Google's position look like, and an honest critique?
  2. The industry adapted by leaning on Search Console for queries. In what specific ways is aggregate, per-query GSC data worse than the old per-visit keyword — and in what ways might it be better or safer?
  3. This case shows a platform changing the measurement rules unilaterally. Identify two more recent changes (from Chapter 28) that similarly shrank what analytics can see, and explain what they have in common with (not provided).
  4. If you had been an SEO in 2011 who optimized entirely against the organic keyword report, what would you have had to change about your workflow the week (not provided) began spreading?