Privacy and data retention

1. Your file is not stored

The Pre-Flight Report is written from your upload, and the upload is deleted in the same request that created it. There is no copy on our servers to leak, to hand over, or to forget about.

You can see this from the outside: the file is uploaded again for each step rather than held between screens. That is not an oversight in the design. It is what not keeping it looks like.

2. What we do keep

Two things, and this is the complete list.

Your Supplier Profiles — column names and the field each one was mapped to. No cell of your data is in them. A profile records that your column called Wholesale holds cost prices; it does not record a cost price. This is the saved work that makes the second file from a supplier one click, and it is yours to delete.

One row per free check, so we can count how many files get checked. It holds a random identifier stored in a first-party cookie, a SHA-256 fingerprint of the bytes, and three numbers: rows read, products found, blockers found. The fingerprint is a hash and does not reverse into a file; it is there so that one person re-running one spreadsheet counts once instead of fifty times. No filename, no cell, no column name, no IP address and no user agent is recorded, and there is no third-party analytics or session-replay script anywhere on this site.

3. What reaches an AI model, and what never does

Only these fields are ever sent, and only for the cells our deterministic checks could not resolve:

Title · Vendor · Type · Tags · Option names and values · Description · SEO title · SEO description · Image alt text

Your cost prices, SKUs, barcodes, supplier identifiers and prices of any kind are never sent to any model. That is enforced by an allowlist in code, which refuses to send a field that is not on the list — rather than by a filter that has to think of everything in advance. The difference matters: an allowlist fails closed on a field nobody thought about.

What does leave is your product titles, descriptions and option vocabulary. We are not going to pretend that is nothing. It is why the rest of this page exists, and why section 5 is a switch rather than a promise.

4. Our model provider is Google, on paid terms

We use the Gemini API on Google’s Paid Services. Three things follow, and the third is the one most policy pages leave out.

Google does not train on it. Under the paid terms Google does not use prompts or responses to improve its products, and no human reviewer reads them.

Google does log it, for a period Google has not published. The terms say, in full:

“For Paid Services, Google logs prompts and responses for a limited period of time, solely for detecting and preventing violations of the Prohibited Use Policy to maintain the safety and security of the Services, and any required legal or regulatory disclosures. This data may be stored transiently or cached in any country in which Google or its agents maintain facilities.”

“A limited period of time” is not a number, and we are not going to invent one. Google sets that period and does not state it; we do not control it and cannot tell you what it is. What we can tell you is what it is for — abuse detection and legal disclosure — and that it is not training.

It is not confined to the EU. If you are asking whether your catalogue stays inside the European Economic Area, the honest answer is no. Google may hold this data in any country where it operates. That is true of the paid terms we are on, and it would be true of any tier.

5. If that is not acceptable for your catalogue, switch it off

AI enrichment is a single control on your account page, and it takes effect immediately. With it off, nothing from your workspace is sent to any model — the request stops on our side, before a provider is contacted.

Everything else keeps working exactly as it did. The Pre-Flight Report, the column mapping, the validator and the export never called a model in the first place. Fields that enrichment would have written are simply left blank. You can switch it back on whenever you like.

The free Pre-Flight Report has no path to a model at all. There is nothing to switch off there, because there is nothing to switch.

6. References

Read on 2026-09-06. Terms carry effective dates and these will change; where they do, this page is wrong until it is corrected, and correcting it is our job.

• Gemini API Additional Terms of ServiceEffective March 23, 2026. The Paid Services paragraph quoted above is in section 4.
• Generative AI Prohibited Use PolicyLast modified December 17, 2024. This is the policy the abuse logging exists to enforce.
• Cloud Data Processing Addendum (Customers)The processor terms the Gemini API terms point to for paid use.