Attributes & accuracy
Is AI describing the right product?
Compare attributes identified in saved answers with the catalog evidence you confirmed. Separate wrong, missing and unknown information.
Confirmed catalog evidence requiredAttribute accuracy
The problem
Start with the question your team needs to answer — not a dashboard full of ungrounded claims.
Specs differ across variants
A missing value looks like an error
A brand citation disguises a missed product
How it works
01
Confirm your reference listing
AnalysedAurora Pro 14
Notebooks · 3 stores
Recommended in
9 / 15
Avg. position
2.3
Store suggested
Mercado Livre
in 6 of 9 answers
Spec accuracy
4 / 5
battery stated wrong
Illustrative data · Aurora, TechPrime, Northwind 02
Read extracted attributes and name variations
Attribute accuracy
RAM16 GBSSD512 GBBatteryTo verifyVariantPro 14Illustrative data · Aurora, TechPrime, Northwind 03
Review mismatches against evidence
SummaryAnswersSourcesPrompt 2 · repetition 3 · model API · saved answer
“For students who want long battery life, the TechPrime X5 is a strong choice. The Aurora Pro 14 is lighter but…”
- • techprime.example/x5
- • No source provided by this model API
Illustrative data · Aurora, TechPrime, Northwind
01 · Attributes & accuracy
Check model and variant first
Never compare a 16 GB model against another variant as if they were identical.
Attribute accuracy
02 · Attributes & accuracy
Unknown does not mean wrong
Missing content has a reason and extraction stage; unsupported values do not become facts.
Sales readiness
- P1Publish a structured spec tableSeller
- P1Add battery hours to the titleSeller
- P2Add product structured dataSite
03 · Attributes & accuracy
Site cited, product not recognized
The product hub distinguishes site evidence from an actual match to the product identity.
Prompt 2 · repetition 3 · model API · saved answer
“For students who want long battery life, the TechPrime X5 is a strong choice. The Aurora Pro 14 is lighter but…”
- • techprime.example/x5
- • No source provided by this model API
What you measure
Measurements come from saved catalog, audit, answer or manually imported rows. No visibility-to-revenue inference.
- Confirmed, conflicting and unknown attributes
- Recognized product names and site-only citations
Transparent methodology
An observation, not a promise
Configured model APIs. Saved original answers. Incomplete responses excluded. Unknown is never treated as failure.
Prompt
Buyer-style, in Portuguese
Provider
Model + version recorded
Repetition
Same prompt, repeated
1 observation
Saved verbatim, never rewritten
Continue the workflow
Review before changing content
Action plan
Prioritize listing corrections, assign owners and track status. Keep the source finding close to each action.
Aurora Pro 14
Prioritized action plan
Clarify the exact model and variant
Pending · saved listing evidence
Add a structured specification table
Pending · saved listing evidence
Review product structured data
Pending · saved listing evidence
CSV available · APIs require setup
Automations & marketplaces
Import CSV, inspect partial results and re-audit stored content. Partner API connections remain setup-required until verified.
Aurora Pro 14
Import & connection review
Manual data · no inferred conversion
Sales performance
Import manual sales aggregates and compare them with saved visibility. Keep association separate from causality.
Aurora Pro 14
Your imported sales, beside visibility
Observed recommendation
9 / 15
Sales source
CSV · Aurora
Read side by side. Visibility does not establish conversion or causality.
FAQ
Straight answers
Can AI invent catalog values?
Assisted extraction is opt-in and only retains values and literal quotes present in captured content.
What is the reference?
Your confirmed catalog and captured listing, not a guessed specification.
Is unknown a failure?
No. Unknown evidence stays distinct from failed rules.
Does this reproduce the consumer chat app?
No. Diagnostics use configured model APIs and recorded prompts in Brazilian Portuguese. Results can differ from consumer apps.
Your next decision starts with evidence
Confirm your product, review the diagnostic and keep every answer.
Run a free diagnostic