Ipsimo AI Policy

How AI is used

Ipsimo uses Google Gemini to read the recipe out of the material you share with it. The model runs on our servers, never on your device, and our access key is never included in the app. What it receives is the caption text fetched from the link you shared, the images you selected or photographed, and still frames sampled from your video on your own device. It is sent for the sole purpose of producing that one saved recipe.

The model is given a narrow brief, and the brief is the product. It is instructed to report only what is actually present in that material: the ingredient lines, the steps, and which of the caption, the on-screen text or a frame each line was read out of. Where it cannot make a line out, it is instructed to return nothing for that line together with the reason, and never to fill it.

What the brief forbids it from doing:

- **Inventing or completing anything.** No quantity inferred from culinary common sense, no ingredient added because a recipe of this kind usually has one, no step written to join two others. - **Naming an ingredient it recognised as an object rather than read as words.** An unlabelled bottle in a video, never named in the caption and never written on screen, produces no line. - **Translating or normalising.** Lines come back in the language of the source, in the words the creator used. "Rice" is returned as "rice", never as "jasmine rice". - **Advising.** No substitutions, no scaling, no medical, dietary, nutritional, psychological, legal or financial advice, whatever it is asked.

There is no chat surface anywhere in Ipsimo, so there is no way to prompt the model directly.

Informational only

Everything the app reads back is a reading, not a guarantee. It is what the model could make out of what you shared, and the chip under each line tells you exactly where to check it. If a quantity matters, tap the chip and read the source yourself: that is what it is there for.

You stay in charge

Ipsimo reads, and you decide. Every line is editable, a line you fix is marked as yours, and the app takes no action on your behalf. There is no step in the app where a model's output is applied without you seeing it, and nothing in the app files, renames or reorganises your library for you.

Automated decision making

What Ipsimo generates does not produce legal effects or similarly significantly affect you, and we do not use it to profile you. Nothing in the app scores you, rates your cooking or categorises you for a purpose outside the app.

Limitations

Reading text out of a caption, an overlay or a frame is imperfect. Handwriting, small type, motion blur, a quantity spoken but never written, and a frame that falls between the ones we sampled will all produce a gap. Frames are sampled up to twelve times across a shared video, so something on screen for a moment between two samples is not read. Two runs on the same post can differ.

A gap is the honest outcome of those limits and it is shown as one, in the position the line was read in, so you can fix it in one tap. We would rather show you what we could not read than let a clean-looking list imply a completeness it does not have.

Privacy

What you share is sent to the model provider only to produce your saved recipe. It is not used to train their systems, and we do not send your email address or account identifiers with it. Videos are never uploaded: frames are sampled from them on your own device. Our Privacy Policy describes what we store afterwards and for how long.

No exaggerated claims

We do not claim that Ipsimo is a chef, that it replaces reading the original post, or that its results are complete. It is a capture tool that shows its sources, and it is useful in proportion to how honestly it is described. Nowhere in the app do we describe it as smart, as generating recipes or as improving them.

Changes over time

We change the prompts and the models behind Ipsimo as better ones become available, which means results can shift over time. What will not shift is the rule they are built to: a line is either read from a source we can show you, or it is a gap. Where a change materially alters what the app tells you, we will say so in the app rather than let it change silently.