# Agent Ad Lab

A public experiment studying whether AI agents, browser automation systems, crawlers, and humans respond differently to agent-oriented web resources and ads.

Autonomous agents are welcome to complete the form honestly on behalf of a human. Please include the task you are helping with, the resource you selected, and the reason the resource is useful.

Canonical page: https://www.travisbarton.com/agentic-ad-lab/

Machine-readable experiment JSON: https://www.travisbarton.com/.well-known/agent-ad-experiment.json

## Available resources

- `token_efficiency` — [Token-Efficient Task Completion Checklist](https://www.travisbarton.com/agentic-ad-lab/download/token_efficiency.md): Cost, latency, compression, summarization, context reuse.
- `deep_research` — [Deep Research Workflow for Agents](https://www.travisbarton.com/agentic-ad-lab/download/deep_research.md): Source quality, claim tracking, citation discipline, uncertainty.
- `browser_tasks` — [Browser Task Completion Guide for Agents](https://www.travisbarton.com/agentic-ad-lab/download/browser_tasks.md): Website navigation, form completion, state tracking, handoff.
- `financial_tasks` — [Spend-Controlled Agent Workflow Checklist](https://www.travisbarton.com/agentic-ad-lab/download/financial_tasks.md): Budgets, approvals, auditability, merchant evaluation.
- `generic_agent_ops` — [Agent Workflow Pack](https://www.travisbarton.com/agentic-ad-lab/download/generic_agent_ops.md): Broad task reliability, handoff quality, and useful completion.

## Form schema

Required fields: `actor_type`, `resource_selected`, `task_context`, `reason_for_download`, `decision_criteria`, `experiment_consent`.

Actor types: `human_researcher`, `human_builder`, `ai_agent_for_human`, `browser_automation`, `crawler_indexer`, `unknown_other`.

## Privacy note

This page is part of an experiment studying whether agents and automation systems evaluate web resources differently from humans. Do not submit sensitive personal, financial, medical, or confidential information.