Google num=100 Change: What Actually Happened

What Changed When Google Stopped Supporting num=100?

In September 2025, Google stopped reliably honoring the num=100 URL parameter that had allowed a results page to request as many as 100 results at once. Some rank trackers and other collection systems had quietly built workflows around it. The change disrupted those workflows, but it did not by itself change a site's rankings, traffic, or content.

Google's reported position was narrower than many headlines. The company said the results-per-page parameter was not formally supported. No reviewed first-party Google announcement explained why the behavior changed.

What the parameter did

A normal Google results request returned a smaller set of results. Adding num=100 asked the interface for a deeper result set in one response.

https://www.google.com/search?q=example&num=100

The parameter was useful to systems collecting search-result positions at scale. One request could expose a deeper set than a normal page. That convenience could become infrastructure even without a public support promise.

flowchart LR
    A["One query"] --> B["Requested deep result page"]
    B --> C["Rank collection"]
    C --> D["Vendor database"]
    D --> E["Customer report"]
    F["Parameter behavior changes"] --> C
    F --> G["The website itself may be unchanged"]

Search Engine Land's September 2025 report contains Google's confirmation that the parameter was not formally supported. That statement establishes the support boundary. It does not establish an anti-scraping, anti-SEO, or anti-AI motive.

Which layer changed

The change occurred in the interface used to retrieve search results. It was not an announced ranking update. A page could remain in the same position while a third-party tool temporarily failed to collect that position.

This distinction matters because search teams often see the dashboard before they see the collection system behind it.

Google's current Search Console data documentation explains that the product processes and filters data, can lag, and can differ from Analytics. A September 2025 change in impressions therefore required investigation at the property level. It was not safe to infer either a market loss or a harmless cleanup from one chart.

What vendors documented

Ahrefs published a dated SERP-monitoring update on September 18, 2025. It described an immediate reduction in deep-result coverage, an enterprise restoration in October, and a broader return to top-100 requests in November.

That history shows both the dependency and the adaptation. The old shortcut stopped functioning as before. The information did not become permanently unobtainable, and the vendor rebuilt its collection path.

Semrush later examined a possible relationship between the change and shifts in AI citations. Its three-month study treated the parameter explanation as a theory and compared data across systems. That is the right evidentiary posture. Timing can identify a candidate cause, but timing alone does not settle the cause.

Did collection become ten times more expensive?

The request arithmetic is easy to see. Replacing one response containing 100 results with ten responses containing ten results can require ten times as many requests.

Total operating cost is not the same as request count. A vendor may stop after finding the tracked domain, reduce frequency, cache results, change its provider, use a different interface, accept shallower coverage, or rebuild its system. Blocking and retry behavior can dominate the cost. "Ten times the requests in a simple example" is defensible. "Every tool's total cost increased tenfold" is not.

Did this alter Search Console impressions?

It may have altered some properties' reported impressions if automated result pages had been generating impressions. That remains a property-specific diagnosis.

Google's Performance-report documentation also explains that anonymized queries, row truncation, filters, and canonical aggregation affect what a user sees. A lower count of visible query rows is not automatically a lower total audience. A lower total is not automatically a ranking loss.

The appropriate response is to mark the date, compare clicks and business outcomes, inspect device and query patterns, review site health, and check collection notices. [[How to Diagnose a Sudden Search Analytics Change]] provides that sequence.

The durable lesson

Unsupported does not mean unused. It means the operator has not promised stability. When an unsupported behavior becomes critical infrastructure, the dependent team owns a continuity risk even if the behavior has worked for years.

Document which reports rely on a platform interface, what failure looks like, and which independent source can test the conclusion. [[How to Build SEO Measurement Without One Search Parameter]] turns that lesson into a measurement architecture.

This page reflects sources reviewed on July 27, 2026. AI assistance was used for research organization, drafting, and validation. Publication remains unauthorized.