Experiment

AI-Crawler Visibility to Search Visibility

A live experiment testing whether properties that AI crawlers already read heavily, but that barely show up in Search Console, will move into real search visibility once matching content ships on the exact gap the data identified.

Krisada Eaton Updated September 11, 2026 Ongoing, first checkpoint reconstructed 2026-09-11 192 views
Hypothesis

A site cluster with strong, sustained AI-crawler traffic and weak-to-nonexistent click-based search visibility is not an underperforming cluster. It is an undiscovered one. Publishing content directly on the query gap the warehouse identifies will move GSC impressions and position meaningfully within weeks, because the sites were already legible enough to earn the AI-crawler attention in the first place.

Measurement Frame

What is being tracked.

Status Ongoing
Started August 4, 2026
Updated September 11, 2026
Duration Ongoing, first checkpoint reconstructed 2026-09-11
Branches 8
Setup

How the test is structured.

The test covered nine sites in the Digital Karma portfolio:

  • AssetClassLeverage.com
  • AISymantix.com
  • SmartInternetInvesting.com
  • DatasetSEO.com
  • AIAssetLeverage.com
  • BuySEOWebsites.com
  • AIWebsiteSystems.com
  • DigitalPropertyStore.com
  • SmartDigitalInvesting.com

Together, those sites were logging a combined 13,781 AI-crawler requests over 90 days, with ClaudeBot and GPTBot leading, while the 'digital asset' query cluster across the same sites sat at 9,903 impressions and only 28 clicks, average position in the 40s to 70s.

That gap, real AI-crawler attention against near-zero click-based visibility, is the variable under test.

Thirteen pieces of content were published across all nine sites in a single day, each one targeting a specific query in the digital asset cluster, cross-linked back to a central hub on AssetClassLeverage.com. Nothing else about the sites changed. No new AI-layer files, no backlink push, no paid promotion.

The test is whether content alone, aimed precisely at a gap the warehouse already proved existed, is enough to move the click side of the numbers.

Observed Signals

What has happened so far.

September 11, 2026 checkpoint ... late, but finally measured.

The August 18 checkpoint never made it into the warehouse. That is part of the result. A scheduled checkpoint that does not run is not a checkpoint, so I rebuilt it from the live page and crawler tables instead of pretending the experiment had been monitored on time.

The launch receipt stays exactly where it was: 9,903 impressions, 28 clicks, and 13,781 crawler requests across the original pre-publish windows. Later warehouse backfills and crawler-classification repairs produce slightly different reconstructed totals, which is why the dated receipt matters.

The cleanest result is at the thirteen pages actually published. Through the latest Search Console date, September 8, all thirteen had received at least one successful crawler request. Eight had earned Google impressions. Together they produced 79 impressions, zero clicks, and 158 successful crawler content requests. Five pages had no Google impression at all.

So the bots found the work. Google tested some of it. Real search visibility did not follow.

The portfolio also handed me the opposite side of the same separation. The AS400System family lost 85.8 percent of its Google impressions in one week while recognized crawler activity initially held. Machine attention can arrive before Google visibility, and it can remain after Google visibility disappears. Those are not the same signal wearing different clothes.

Decision Thread

What would support or challenge it.

The original short-window hypothesis is not supported by the launch pages. Crawler discovery happened. A ranking win did not.

That does not make crawler data useless. It changes the job I can honestly give it. Crawler activity is evidence that a machine requested a resource. It is not evidence that the resource was indexed, selected, cited, or ranked.

What started as a Bots First ranking question is now an Exposure Velocity question: whether crawler requests, query expansion, impressions, human activity, and citations form a useful sequence before clicks make the change obvious. One signal alone is not enough. That is the better experiment.

Experiment Lab

Keep Following the Tests

Move from this open thread back into the full experiment library.