From Feed to Agent

Independent product-strategy research on how people move from discovery feeds and reviews into AI-assisted decisions.

Explore the interactive publication: valerianxxx.github.io/from-feed-to-agent

Research in one view

The study compares four consumer-research surfaces across the same frozen task lattice. The central product opportunity is not another destination; it is an evidence handoff that carries intent, source origin, uncertainty, and proof between tools.

Evidence layer	Scope
Formal observations	192
Research surfaces	4
Collection rounds	2
Survey submissions preserved	149
Primary / expanded survey layers	73 / 122
Metric-state records	3,264
Named validation checks	136
Stratified coding-audit rows	48

What the evidence supports

• TikTok functioned as an origin-rich inspiration surface inside the captured search-page boundary.
• YouTube provided routes to deeper reviews, while the study did not watch or score video content.
• Google AI Mode offered source-linked synthesis but exposed generation-state reliability as a product concern.
• ChatGPT consistently produced structured initial answers in the captured account, browser, and time window.
• Survey preferences favored visible control over automatic cross-product handoff in both eligibility layers.

These are descriptive, task-scoped findings. They are not population estimates, causal claims, universal platform rankings, or evidence of product-market fit.

Repository map

data/                 Privacy-safe row-level and aggregate evidence
methodology/          Protocol, frozen method, limitations, and accessible appendix
public/downloads/     Report, workbook, appendix, social deck, and public-data bundle
src/                  Interactive publication source
.github/workflows/    Reproducible quality and GitHub Pages deployment

Run locally

Requirements: Node.js 22+ and pnpm 11.19+.

pnpm install --frozen-lockfile
pnpm dev

Run the same checks used in continuous integration:

pnpm check

The check performs TypeScript validation, a production build, publication-trace scanning, download inventory checks, and row-count verification for the four principal public datasets.

Public artifacts

• Research report — PDF
• Research report — editable PPTX
• Method appendix — PDF
• Auditable analysis workbook
• Editable social carousel
• Share-ready social carousel — PDF
• Privacy-safe public data bundle

Verify any downloaded artifact against SHA256SUMS.txt.

Privacy and evidence boundary

The repository excludes participant-level survey responses, birth fields, exact response timestamps, account identifiers, raw logged-in browser captures, local file paths, and exact private-session prompts. Public rows are derived research outputs with explicit missing-state handling.

See LIMITATIONS.md, RESEARCH_CHARTER.md, and the data dictionary before reusing or interpreting the results.

NeedRadar method relationship

This research contributes origin-preserving evidence, claim-to-proof, verification-state, and consent patterns to the NeedRadar evidence-to-execution ecosystem.

The relationship is method-level and metadata-only. NeedRadar must treat this publication as one research artifact rather than converting study rows or respondents into independent demand observations. See needradar-project.json.

Citation

Use the repository's CITATION.cff metadata or cite:

Xu, S. (2026). From Feed to Agent: The Evidence Handoff. Independent product-strategy research. https://github.com/ValerianXXX/from-feed-to-agent

Rights and contributions

Copyright © 2026 Sihan Xu. All rights reserved. Research corrections and reproducibility improvements are welcome through GitHub issues or pull requests; see CONTRIBUTING.md.