Crossave shows one number: the real total cost of a purchase. You are entitled to know which part of it was worked out by a rule and which part by an AI model — and where the model can be wrong. This page says so plainly.
Prices and shipping costs come from the shops and marketplaces themselves. Duty and VAT come from a rate table written by hand, with the source and the verification date next to each figure. No language model decides how much you pay: that part is arithmetic, and you can check it on how we calculate.
A model reads your search and picks the product category (electronics, collectibles, footwear…), which then selects the right duty rate. It also cleans up the query — a misspelling, a model number written in an unusual way.
When one search returns items that look alike but are not comparable — a graded card and an ungraded one, a first edition and a reprint — a model splits them into groups with a label. The label is written by the AI; the prices inside it are not.
Lotteries dressed up as single items, multi-item lots, replicas, spare parts sold under the name of the whole product. Some of these checks are word lists, some use a model.
The guides in the guides section are written with the help of AI. Every rule, date and figure in them is checked against an official source, which is linked at the bottom of each guide, and the numbers in the examples are produced by the same calculator the site uses — if the calculation changes and a guide does not, an automated test fails. A person reads and approves every guide before it is published, and holds editorial responsibility for it.
Category detection can pick the wrong one, and with it the wrong duty rate. A group label can be imprecise. A suspicious-listing check can be too strict or too loose. These are estimates and heuristics, never certainties — which is why, whenever a piece of data is missing, the site writes “n/a” instead of guessing. If something looks wrong to you, there is a report button on every result: it goes straight to us.
There is no chatbot on this site, and nothing here talks to you as if it were a person. We do not generate product photos or fake reviews. No result moves up because someone paid for it: the order is decided by the lowest total cost, never by affiliate commission. Only two things push a row down, and you should know which: an incomplete total, when the seller does not state shipping, and a suspicious price warning, which is one of the checks described above. In both cases the row stays visible, with its warning. We do not use your searches to train any model and we do not sell them; which services a search is forwarded to in order to find prices is set out in the privacy policy.
The EU AI Act (Regulation (EU) 2024/1689) has applied its transparency obligations since 2 August 2026. Nothing we do falls into its high-risk categories, and none of our systems interacts directly with people. This page exists anyway, because the reason to say how a number was produced is not that a law requires it.
Last updated: 2 August 2026.
Compare a product now