Understanding Quality Scores
Plans: Starter ✓ · Growth ✓ · Pro ✓ · Agency ✓
The Quality Scores on a product — Durability, Quality Perception, Value for Money, plus a price positioning — are generated by the AI model from your catalog text. They are not backed by reviews, returns or sales data unless you provide that evidence yourself.
What the scores are
Every enriched product gets:
| Score | Range | Meaning |
|---|---|---|
| Durability | 1–10 | Inferred build quality and longevity. |
| Quality Perception | 1–10 | Inferred perceived quality. |
| Value for Money | 1–10 | Inferred price-to-quality ratio. |
| Price positioning | budget / mid-range / premium / luxury | Inferred market segment. |
The three scores appear in the Quality Scores block on the product's first tab. The price positioning is shown as a pill in the product facts card, next to the availability, above the price.
Categories and collections are also given a perceived quality, value for money and price range, and pages a perceived quality, but the dashboard shows the Quality Scores block on products only.
What the model uses
- Title, description, price, vendor and categories.
- Your store-level context from AI Training.
- Your answers to the product's Teach the AI questionnaire, if you filled it.
It does not see real customer reviews, returns, sales or conversion data, or external catalogs — unless you put that information in the questionnaire.
What this means for you
- Scores are a synthetic opinion, not verified facts.
- The same product can score differently between runs, because the model doesn't return identical output every time.
- Useful relatively, not absolutely. The model has common sense about markets, so scores help you compare products within your own catalog and category. They don't benchmark you against competitors, and a "high quality" pair of socks is not the same bar as a "high quality" mattress.
- Typical competitors have the same limitation — plausible inferences, not market data. Name the real ones in Teach the AI → Positioning & context.
How grounded is a score? The badge
Next to the Quality Scores, a badge tells you how much evidence backs them:
| Badge | Meaning | What triggers it |
|---|---|---|
| ✓ Data-grounded | Backed by evidence you provided. | An enrichment run with 3 or more evidence answers filled. |
| ◐ AI + owner hints | Informed by your answers, not yet backed by enough evidence. | Any saved Teach the AI answer. |
| ⚠ AI-inferred | The model's opinion from title, description, vendor and price only. | No answers. |
The evidence answers are the five questions in Teach the AI → Verifiable evidence: warranty, certifications, returns rate, reviews and manufacturing origin. A warranty of "None" or a returns rate of "Don't know" doesn't count as evidence.
Saving Teach the AI answers moves the badge from ⚠ AI-inferred to ◐ AI + owner hints straight away. ✓ Data-grounded needs the next enrichment after you add the evidence — filling the questionnaire doesn't change an existing enrichment. Click ✦ Enrich with AI afterwards.
Where scores reach the storefront
Scores are not shown to shoppers on the page. Once you publish an enriched product, its JSON-LD carries them as schema.org additionalProperty values — qualityScore, durabilityScore and valueForMoney (each 1–10), plus pricePositioning. Crawlers and AI agents read that JSON-LD. You can see it on the product's </> tab.
When to re-enrich
Three triggers warrant a re-enrich:
- You added evidence in Teach the AI — so the model uses it and the badge can move up.
- You substantially changed the description — sync brings in the new text, but the scores stay on the old text until you re-enrich.
- You named the real competitors — so the model stops guessing them.
Don't re-enrich just because you don't like a score: it uses a credit, and scores rarely swing much unless the inputs change.
How to use scores
Good uses:
- Rank products within a category to surface your strongest items.
- Find low-scored products whose descriptions need work.
- Check whether a description change shifts the score after a re-enrich.
Bad uses:
- Marketing claims ("rated 9/10 by AI!") — see Reviews setup for why Clione never fabricates ratings.
- Comparisons across categories.
- Pricing decisions — the model doesn't see real demand.
Troubleshooting
A premium product shows a budget price positioning — The model weighs the price against the description language. If the description doesn't mention craftsmanship, materials or brand heritage, it may anchor on the product type rather than the brand. Expand the description, or add it in Teach the AI, and re-enrich.
Same product, different scores on re-enrich — Expected. Fill Teach the AI → Verifiable evidence to anchor them.
The badge stays ◐ AI + owner hints after I added evidence — ✓ Data-grounded comes with the next enrichment. Click ✦ Enrich with AI.
One product's scores look off — Check its description in your store first: sparse text, or text written for a different SKU, gives the model little to work with. Fix it, sync, fill Teach the AI, re-enrich, and compare with the previous version in the History tab.