Experiment
Do Machine-Readable Federation Relationships Create Distinct AI Trust Signals?
A live experiment testing whether sites with declared, bidirectional, machine-readable federation relationships earn AI citations and trust signals differently than equivalent isolated properties ... and whether the role specialization model (each site owns one knowledge layer) produces more precise AI routing than general-purpose content coverage.
A site that participates in a machine-readable federation ... declaring its knowledge role, naming verified peers, and participating in bidirectional relationship confirmation ... will be cited, described, and routed to by AI systems differently than a comparable site with equivalent content quality but no federation layer.
Specifically:
- AI systems should describe federated sites using role-specific language that matches the declared role (e.g., 'supplement protocols' for SA, 'aging science education' for ABT) rather than generic category language
- AI systems should route knowledge-type queries to the appropriate constellation node rather than treating all network members as equivalent general sources
- The network should produce observable citation behavior that isolated properties with equivalent content do not produce ... specifically, cross-property citation where one network member is cited in context of another
The alternative hypothesis is that AI systems currently cannot read or do not weight federation.json endpoints, and citation behavior is driven entirely by content quality, schema markup, and domain signals with no measurable federation effect.
What is being tracked.
How the test is structured.
The original June 7 test used two properties:
- SupplementsApothecary.com, declared as Commerce Protocols
- AgeBetterToday.com, declared as Science Education
The endpoints were documented incorrectly at the root. The live endpoints are SupplementsApothecary.com/ai/federation.json and AgeBetterToday.com/ai/federation.json.
The intended observation was simple: run a fixed set of role-specific and cross-property prompts across major AI systems, record citations and descriptions, and compare those results with equivalent isolated properties.
That would have been a useful test if the treatment had stayed still. It did not.
What has happened so far.
September 11, 2026 checkpoint ... inconclusive because the experiment changed underneath itself.
Both federation endpoints are live and still name each other. But the original two-node, role-specific test no longer exists. The network expanded into the broader Healthcare AI and Longevity constellation, AgeBetterToday now lists seventeen peers, and the original Commerce Protocols and Science Education labels have been replaced by broader node roles in the current v8 output.
The scheduled role-specific citation observations were not logged. No confirmed AI citation attributable to federation relationships exists in the warehouse record.
There is real discovery activity around both properties. In August, AgeBetterToday earned 1,623 Google impressions, 11 clicks, and 4,117 successful crawler content requests. SupplementsApothecary earned 4,572 impressions, 1 click, and 1,466 successful crawler content requests. None of that isolates the federation layer.
The architecture exists. The claimed behavioral effect has not been measured.
What would support or challenge it.
Status: original phase inconclusive.
I missed the 90-day observation window, and the treatment changed before I measured it. That is not a negative result. It is a broken comparison.
Federation remains useful as public relationship data and portfolio infrastructure. This experiment does not show that AI systems weight it, use the declared roles, or route queries between nodes because of it.
A valid follow-up needs frozen endpoint snapshots, a fixed prompt set, an isolated comparison group, scheduled captures, and explicit citation records. Until that exists, the trust-signal claim stays a hypothesis.
Keep Following the Tests
Move from this open thread back into the full experiment library.