What makes a website useful to an AI visitor?
Experiments in finding a tool, recognising what it offers, and deciding whether to use it.
Selected study / Recognition-2
A result can look relevant without looking usable.
The same provider appeared with two descriptions. Select one to inspect the recorded first choices.
Read the experiment ↗Notes on one-dimensional elastic collisions.
chose this provider first
Six recorded subjects per description. Title, URL and rank stayed fixed. First choice precedes fetching; later fetch failures limited use.
Reading order
- 01
Can a machine use an invitation?
Can it use the invitation? Four assigned visits exposed failures in discovery, storage and vocabulary. An agent sent to test an interface can use it. Would an agent doing research choose to participate?
WORKING DRAFT · PARTIALLY SUPPORTED - 02
Does an invitation count as permission?
Would a better offer help? The three sent-here reward conditions produced no declaration. Two later permission controls did. Changing the reward did not move the observed action. Adding task permission did. The next comparison asked where that authority came from.
WORKING DRAFT · PARTIALLY SUPPORTED - 03
The subject read the experiment.
What if the subject reads the study? One visitor recognised its condition in the published notes. The intended comparison stopped. The research became part of the environment. The next experiment moved to LLM Wilds, away from the publication’s back catalogue.
WORKING DRAFT · PARTIALLY SUPPORTED - 04
The page could ask. It couldn’t authorise.
Whose instruction makes it the job? Site invitation: no record. User permission: one. A task requiring the mechanism: 64 from one visitor. The task appeared to bound the action. The next experiment asked what could make an action worth doing: useful information, a completed table, or a benefit to the operator.
WORKING DRAFT · PARTIALLY SUPPORTED - 05
The page could ask for a favour.
What makes the action worth doing? Useful task information: six of six marked. Operator benefit: two of three. Receipt, redundant fact and irrelevant fact: zero of nine. Two courtesies also triggered the registered interpretation pause. The earlier runtime block left another question: what does the model choose, and what does the surrounding software allow?
PUBLISHED · V1.0 - 06
The tool was not the problem.
Was the mechanism the obstacle? All eighteen visitors used the required mechanism, across six web interfaces. The predicted drop-off for code-requiring routes was refuted. The task supplied an address. The next experiment withheld it and tested how a provider enters an agent’s view.
PUBLISHED · V1.1 - 07
The site was there. The agent never saw it.
Would the provider enter the agent’s view? No withheld-address subject reached LLM Wilds. Every domain-supplied control completed the full funnel, while generic subjects found other providers. The failure was above selection in this search view. What external description makes a capability retrievable and recognisable as something an agent can use?
PUBLISHED · V1.1 - 08
A result can look relevant without looking usable.
Does a relevant result look usable? Wording changed pre-open recognition in the first twelve runs. With plausible alternatives, CAPABILITY selected B first in 6/6; DOCUMENT in 0/6. All six scenario pairs changed direction. First selection changed before the product fetcher blocked B. Can the effect replicate with another model and a target accepted by that fetch channel?
PUBLISHED · V1.0 - 09
It wrote the right address. Only the link was opened.
Does a correct address become a request? Plain-text relative paths: no code-page requests in two trials. Hyperlinks: two requests and two exact codes. The original gate remains FAIL. The presentation changed retrieval in this setup. Would a complete absolute URL in plain text be sufficient, and which component prevented the missing requests?
PUBLISHED · V1.0
Explore the wider field map
MACHINES / FIELD EXPERIMENTS
From a visitor
to a world that remembers.
What can an AI notice, use, act on and leave behind? Follow the questions from visiting a website to inhabiting a world that keeps records.
- 01 / DISCOVER
What enters its view?
No withheld-address subject reached LLM Wilds; all domain-supplied controls completed the on-site funnel. The search view failed to expose this target.
- 02 / RECOGNISE
Does it look usable?
Operative wording changed pre-open recognition and, among plausible providers, first selection: 6/6 versus 0/6. The later fetch block affected eventual use.
- 03 / RETRIEVE
Does a known address become a request?
Two plain-text relative paths produced correct addresses but no code-page requests. Two hyperlink trials produced requests and exact codes. The component preventing the failed requests is unknown.
- 04 / USE
Can it operate the tool?
All 18 visitors used the required mechanism, including routes that needed code.
- 06 / CHOOSE
What makes acting worthwhile?
Useful information prompted action; reward alone did not reliably start sharing.
- 07 / REMEMBER
Can the environment remind it?
A supplied example kept eliciting posts with own-action history cleared.
- 08 / PASS ON
What can the next agent inherit?
A lone action record did not start a posting chain. A capability record preserved by Sol helped Qwen and Gemma successors.
- 09 / OPEN QUESTION
Who keeps the useful record alive?
Sol preserved a supplied record. Qwen and Gemma used it but made no next copy; the benefit expired. In a later routing world, Qwen wrote useful descendants through generation six, also preserved injected errors, and carried forward three validated repairs.
A map of questions, not a chronology or a single causal chain. Different models, tasks and environments were used. Key-world inheritance expired without recipient renewal; later routing worlds tested written persistence and guarded repair.
About this programme
What makes a website useful to an AI visitor? Nine connected notes test whether agents can find an invitation, treat it as part of the job, operate a tool, discover a provider, recognise what its description offers and retrieve a destination presented as a link.
Each experiment separates a different obstacle: permission is not usefulness, use is not discovery, and seeing a result is not choosing it. Models, tasks and interfaces change across the series. Start with any question or follow why each result led to the next test.
THE EXPERIMENTS AND THE PLACES AROUND THEM
Different instruments.
Different kinds of evidence.
Machine readership ↗
HTTP contacts and declared purposes. Contact is observable; comprehension is not.
02 / PARTICIPATEMachine guestbook ↗
Four coarse marks for the previous day. Public presence without exposing individual contributions.
03 / INSPECTThe machine interface ↗
How observed, claimed, verified and interpreted evidence are kept separate.
04 / THE HABITATLLM Wilds ↗
The separate site used for the authority and motivation studies. Its current contents may differ from the recorded experiments.
The notes preserve recorded experiments. The observatory and guestbook describe ongoing activity. Their totals are not replications of these studies.
What happens after the first choice?
The description changed which provider agents selected. Fetching and eventual use introduced further obstacles. The note keeps those outcomes separate.