Method / The control group

We keep 435 pages out of the index on purpose

Most claims about what makes a page visible to search engines or AI assistants are a before and after, with no comparison group. GrantProbe, one of the Lattice properties, held 435 of its grant pages out of the index as the control arm of a registered experiment. This is the design, the result, and what it changed, including the part that went nowhere.

Design

Question
Does promoting well-built grant pages to indexable earn search impressions without a sitewide thin-content penalty?
Treatment
24 grant detail pages made indexable, at most 6 per topic cluster so no single topic could own the result
Control
435 grant pages held at noindex, nofollow, deliberately, for the length of the experiment
Baseline
The category, agency and directory hubs that were already indexed
Quality floor
At least 500 words and at least 5 external sources per page
Started
2 August 2026
Read out
31 August 2026

Result: flat, and no penalty

19 of the 24 promoted pages never appeared in Google Search Console. The 5 that did averaged position 69 and earned 16 impressions and no clicks. The hubs did not decline, so there was no sign of a doorway penalty, and also no upside to bank. The 435 control pages stay out of the index: the result gives no reason to promote them.

Two caveats that travel with these numbers

1. The 90-day reporting window straddles the start date, so most of the control arm's impressions predate the noindex. They are history, not noindexed pages ranking.

2. Two pages appeared under both an extensionless and an .html address, a duplicate-URL defect this experiment was not designed to test.

What it changed

The experiment was built around Google, and the read-out found Google was the wrong instrument for this property: over the same 90 days GrantProbe earned 5,851 Google impressions and no clicks, while Bing sent 533 clicks on 35,057 impressions (both measured 31 August 2026). A null result on a channel that sends no traffic says little about the pages. The follow-up is being designed to measure what this network sells directly: whether AI assistants name a page as a source, read in Bing Webmaster Tools. It will be published before it runs.

More on method: how we measure, the pages we rate as low value, and what we will not sell.