Execution
Burnish reads your whole catalogue, works out what each product is missing, and writes the fixes back.
Every one of the forty-one named fix types is one or the other — and the gate is the default.
29 of 41 · nothing ships without you
12 of 41 · deterministic writes
The quiet overwrite is the danger. A field is rarely just a field.
Six dependency checks
Smart-collection rules, Merchant Center feeds, Facebook catalogs, foreign metafield namespaces, occupied custom labels, structured data already on the page.
A tripped check queues
It moves to the approval queue, where you see what it would have overwritten.
Two fixes carry a confidence floor
Variant consistency will not fire below 0.90, product-type normalisation below 0.95. Under it, a fix is proposed.
Your theme is never edited
Storefront work ships through a Theme App Extension, never your theme files.
What shoppers and AI assistants read — generated to be quotable and specific.
The hero image AI shopping surfaces match on — a precise, factual description so assistants can read and cite what the product looks like.
Want these FAQs on your product pages?
Turn on FAQ display so AI assistants can cite your answers — and shoppers get clear answers on the spot.
Make this product sound like you.
Your brand's source of truth — who you serve, what sets you apart, your voice, and what you'll never do. Your brand's brain: Burnish reads it to shape every product description, blog and research hunt.
With the three limits on that stated, rather than left for you to discover.
Every write saves the previous state first. Evidence rows are kept as long as your store is — there is no window to run out of.
Thirty-seven of the forty-one types have a registered undo, which reads the saved before-state.
No registered strategy; still pending; already rolled back; or the state exceeded 256 KB, which Burnish reports rather than hides.
Four levels, set by you. Every paid plan can reach the top one.
Reads and scores. Zero mutations, ever.
Proposes everything, writes nothing until you approve. Stores without a plan are capped here.
The auto-safe twelve apply themselves; anything riskier goes to the queue.
The widest lane, and it needs your store’s explicit consent before it turns on.
Each has a published weight and a score ceiling. A product counts as AI-ready at 80.
| Component | Weight | Ceiling |
|---|---|---|
| Description quality | 0.22 | 95 |
| FAQ presence | 0.14 | 100 |
| Structured data | 0.12 | 100 |
| Title optimisation | 0.12 | 95 |
| Alt-text coverage | 0.10 | 95 |
| Meta description | 0.10 | 90 |
| Meta title | 0.10 | 90 |
| Manifesto | 0.10 | 100 |
Projected lift = Σ max(0, (ceiling − score) × weight). It is what a publish and a re-scan actually realise, which is why it is never the rounder, larger number.
Your plan governs how much of it is rewritten.
| Presence | Authority | Omnipresence | |
|---|---|---|---|
| Products optimised each month | 300 | 1,000 | 2,000 |
The scan is not the allowance. Burnish scores your catalogue in full — whole-catalogue on every plan, up to a stated ceiling of 7,000 products — then spends your monthly budget where the projected lift is largest.
Twelve of forty-one, in five groups: 18 product · 9 content generation · 7 content · 4 technical · 3 collection.
The free audit scores your products against the same rubric and shows the gap, before you decide whether it is worth fixing.
