How Pastefront uses AI

• Claude, from Anthropic: building a store, chat edits, A/B test ideas, ad copy, accessory ideas and checking market prices.
• Saver: a lower-cost model reached through OpenRouter, for chat edits and ad copy when the owner picks it.

• A store's name ideas, headline, product descriptions, FAQ, colours and fonts, from the supplier's listing and the owner's brief.
• Changes the owner asks for in plain words, as a preview with a list of what changed.
• Ad copy for Meta and TikTok, and ideas for A/B tests.

Prices start from the supplier's real cost and shipping quote. The AI doesn't decide what the supplier charges, and a sale price follows Pastefront's fixed rules for honest sales.

• Change a store on its own. A chat edit waits as a preview until the owner presses Publish, a rebuild runs only when the owner starts one, and an earlier version can always be restored.
• Send an email, post or ad on its own.
• Invent reviews, ratings, customer counts or urgency, or promise delivery dates or health results. Everything it writes goes through a check for those claims, and flagged text is rewritten before the owner sees it.
• See buyers' names, addresses or order details.

Pastefront doesn't train AI models. Requests to Anthropic go through its commercial API, whose terms say it doesn't train its models on them by default. Saver requests through OpenRouter ask for zero data retention and no training.

AI can be wrong. The store owner is the seller, reads what the store says before it goes live, and signs that they have. Before a store goes live, Pastefront checks its whole text again and asks the owner to fix anything flagged or confirm it's true. See the terms and the subprocessors page.