RecoScopeRecoScope
PlatformMethodologyResearchTrackerAbout

RecoScope

An independent AI commerce benchmark designed, built, and operated by Robert Hu.

Explore

  • Platform
  • Tracker
  • Methodology
  • Research
  • About
  • Subscribe

For brands

  • Request a private analysis
@recoscope·Terms·Privacy

Built and operated by Robert Hu.

About the Builder

Why I Built RecoScope

AI systems were becoming product discovery engines, but brands and operators had no reliable way to observe how those systems made recommendations. I built RecoScope to turn that emerging behavior into structured, longitudinal evidence.

The problem I saw

Product discovery was moving beyond search and marketplaces. When someone asks an AI assistant what to buy, the answer shapes consideration before a shopper ever reaches a product page.

Conventional rankings did not explain those answers. Marketplace best-seller position and AI recommendation visibility turned out to be different things, and often disagreed.

Most of the industry conversation ran on screenshots and anecdotes. There was no repeatable way to measure what AI actually recommends, or how it changes. That gap is what RecoScope was built to close.

What I built

I designed the product concept and the evaluation methodology: a prompt framework of repeated, commercial-intent questions, run across ChatGPT, Claude, Gemini, and Perplexity, with each model classified by commercial interest.

I structured the normalized data model, from categories through runs, agent responses, brand mentions, and run insights, so raw evidence is preserved separately from the structured data derived from it, and every finding is traceable back to a specific run.

I built the publication workflow and the public experience: a draft, reviewed, published status lifecycle with published-only filtering, the category tracker, the research findings, prompt-level pages, and a private per-brand analysis output, all generated from one dataset. I operate the benchmark on a recurring cadence and interpret the results that get published.

How I think

Measure before advising

RecoScope started as a measurement system, not an opinion. The tracker exists so claims about AI recommendations can be checked against repeated, structured data.

Preserve raw evidence

Every model response is stored in full, separate from the normalized data derived from it, so findings stay auditable and re-derivable.

Structure before scaling

A normalized data model, from categories through runs, responses, mentions, and insights, came first, so new categories and models plug into a consistent shape.

Keep human judgment in the loop

Runs move through a review and publication workflow. Nothing reaches a public surface until it has been through that gate.

Build reusable capability, not one-time output

The same dataset produces public tracker reports, research findings, prompt-level pages, and private brand analyses. One system, many outputs.

Relevant background

More than 20 years in ecommerce, across brands, marketplaces, merchandising, digital transformation, and technology selection.

My current focus is AI commerce: product discovery, evaluation systems, and how enterprises adopt AI in the path to purchase. RecoScope is where that focus is applied and made concrete.

What this project demonstrates

AI product thinking

Turned an emerging behavior, AI recommending products, into a defined product with a measurement model and public outputs.

Multi-model evaluation design

Structured a repeatable comparison across ChatGPT, Claude, Gemini, and Perplexity, and classified each model by commercial interest.

Data normalization

Designed a schema that separates raw brand strings from a normalized brand name, so counts are consistent across models and runs.

Longitudinal research

Built a recurring benchmark cadence so the data shows how recommendations move over time, not just a single snapshot.

Operational workflows

Implemented a draft, reviewed, published status lifecycle with published-only filtering on every public query.

Quality controls

Separated public and private outputs, and gated publication, so private work never leaks and only reviewed data goes live.

Product discovery expertise

Applied two decades of ecommerce and merchandising experience to a discovery channel that is moving from search and marketplaces toward AI answers.

Translating emerging technology into usable systems

Converted an unstructured, anecdote-driven topic into structured evidence people can query, cite, and act on.

I am interested in roles where AI, commerce, product, data, and transformation intersect.

The clearest picture of how I work is the platform itself. Explore it, then get in touch.

Explore the platformGet in touch