AIreviews: Every review. One answer.

The starting point

Review sites hand you a wall of stars and leave you to do the reading. The consumer product had to earn trust with every sentence, and the business model had to live on the B2B side: an AI agent each business owner builds on their own operating truth.

What TDAT built

The consumer engine, live today: natural-language search that verifies every sentence of its answer against the reviews it cites, drops anything unsupported, and shows its confidence honestly, down to "not confirmed" labels on claims like dietary and accessibility. Alongside it, the B2B platform: business onboarding, AI research that auto-fills a business's operating facts from the public web, competitor intelligence, and a versioned Superagent builder.

Current status

The consumer app is live and code-reviewed through its third build stage. The B2B platform has its foundation shipped, with the agent runtime and live data ingestion as the current build front.

One feature: an answer that exposes its evidence

The consumer search experience connects a natural-language question to review-based answers and citations. Its claim-grounding workflow checks generated statements against cited reviews and uses not-confirmed labels for claims the available material cannot support.

The lesson for your build

The useful design decision is to let the reader inspect the basis of an answer and see uncertainty. For a similar search product, evaluate unsupported, conflicting and missing evidence as deliberately as a clear match. A citation is something to inspect, not a guarantee of correctness. The business-agent runtime and live ingestion remain a separate build stage.

What comes next

Wiring the Superagent runtime and live review ingestion, then billing: turning free consumer distribution into B2B revenue.