Find
Find products that match explicit price, feature and delivery constraints.
E-commerce
Make commerce journeys reliable for AI agents.
Verify product facts, variant selection, delivery and cart state with replayable evidence.
For retailers, marketplaces, DTC teams, commerce platforms and agencies.
Test the decisions that shape a commerce journey.
Tasks cover discovery, selection, conversion and recovery without making a production payment.
Compare
Compare variants and specifications without mixing facts between products.
Verify
Confirm current price, discount context, availability and delivery promise.
Select
Choose the intended size, color or configuration and preserve the selection.
Reach cart
Add an item in sandbox or reversible mode and verify the cart state.
Recover
Handle unavailable inventory, validation errors and replacement options.
Follow the facts from product page to safe stop gate.
The run stays linked to assertions at every consequential transition.
Product match
Visible facts and structured data identify the same item and variant.
Delivery choice
The selector remains named, stable and actionable after inventory loads.
Cart assertion
The intended item, quantity, variant and price appear in the reversible cart.
Typical findings connect technical behavior to commerce risk.
Conflicting price
Visible, structured and cart prices describe different states.
Ambiguous variant
Custom controls hide selection state from the accessibility tree.
Moving target
Inventory hydration shifts the delivery or cart control during action.
Duplicate action
Delayed feedback causes the agent to add the same item twice.
Measure reliable completion, not a lucky click.
A successful outcome can still carry friction that makes it fragile across agents.
Session recording
Replay discovery, selection, recovery and the final assertion.
Completion time
Separate navigation, rendering, inventory waits and retries.
CLS and target movement
Capture page CLS and movement around the intended control.
Result consistency
Repeat standard tasks three times and critical tasks five times.
Production payment remains outside the public audit.
Public runs stop before purchase. Full checkout, refunds or order changes require an approved sandbox, fixtures and explicit authorization.
Technical signals are inventoried without becoming arbitrary requirements.
The audit checks visible and machine-readable commerce facts, interaction semantics and optional agent-native surfaces.
- Product and Offer structured data
- Canonical URLs, sitemaps and merchant signals
- Variant, quantity, cart and delivery semantics
- Accessibility tree and keyboard interaction
- Hydration, network errors, CLS and DOM stability
- OpenAPI, MCP or WebMCP when present
Your team gets a traceable remediation package.
- Task by agent result matrix
- Recordings, screenshots and interaction traces
- Product fact and structured-data conflicts
- Prioritized findings with acceptance criteria
- Baseline and retest comparison
E-commerce audit questions
Do public tests place real orders?
No. They stop before payment or another irreversible transaction.
Is missing WebMCP a failure?
No. It is reported as capability state. Browser and search readiness are scored independently.
Do you estimate revenue impact?
Only when a partner supplies suitable data. The baseline audit reports observed task and conversion friction.
Practical testing guide
Turn this commerce journey into a repeatable test.
Use the canonical task, fixture, stop-gate and evidence contracts for product discovery, variants, delivery and cart review.
Read the e-commerce testing guideTest a commerce journey before agents meet it at scale.
Apply with one product discovery, delivery or cart workflow for a private partner review.