Claims you cannot make
A draft that promises “clinically proven” or “100% recyclable” without proof is a legal risk, not a conversion win.
Under Settings → Brand you set brand voice, approved and prohibited claims, legal requirements and verified product facts. Every AI proposal is checked against them when it is written and again when it is applied.
Brand guardrails are the rules every AI draft in Sphen.ai has to follow. Under Settings → Brand you record brand voice, description, audience and brand knowledge, list approved and prohibited claims with legal requirements and required evidence, and keep verified product facts. Each proposal is checked against these rules when it is generated and again when it is applied.
Updated:
A draft that promises “clinically proven” or “100% recyclable” without proof is a legal risk, not a conversion win.
Materials, compatibility, shipping and returns are not in the product title. Without a source, the model fills the gap.
The brand brief sits in a document or in someone's head, and every tool gets a different version of it.
Understand your brand holds Brand voice, Brand description, Target audience and Brand knowledge — the brief you would hand a copywriter. Import folder or files reads .md, .txt and .csv files into it.
Control claims lists Approved claims and Prohibited claims, plus Legal requirements and Required evidence. A claim cannot be approved and prohibited at the same time.
Reliable product facts holds one fact per row with an attribute, a source and an approval state. Only approved facts are given to the model.
The Brand rules card under Settings → Brand has three sections: Understand your brand, Control claims and Reliable product facts. Save guardrails stores all of them at once; Discard drops unsaved edits. After saving, new AI proposals use the updated guardrails.
Each fact carries an attribute — Audiences, Use cases, Benefits, Materials, Compatibility, Proof or Shipping & returns — a source (Merchant, Catalog or Derived) and an approval state (Approved, Pending or Rejected). The attribute is suggested from the text of the fact; your choice always wins.
Catalog drafts, FAQ answers and Digital PR drafts are checked against the guardrails. Brand knowledge also feeds blog generation and the agents.md draft, so the same brief reaches every surface.
Included in every plan, with no separate quota.
It is rejected. Sphen.ai checks every AI output for prohibited claims and for product facts that are pending or rejected, both when the draft is generated and when you apply it. A rejected draft is never written to Shopify.
The brand profile from onboarding holds business name, category, audience, markets and tone. Guardrails add what the profile cannot: claims you allow or forbid, legal requirements, required evidence and verified product facts. Where guardrail fields are empty, the confirmed profile values are used.
Yes. Import folder or files under Brand knowledge reads .md, .txt and .csv files up to 200 KB each and appends their text for review. The import happens in your browser, and you edit the result before saving.
No. Saving guardrails affects new proposals only. Content you applied earlier stays as it is until you regenerate or edit it.
You add them yourself as Merchant facts, or they come from the catalog or are derived from it. Facts from the catalog can be used while they are still pending; everything else needs your approval first.
Install Sphen.ai, let it read your store, and decide exactly what goes live.