---
description: "The last layer of discovery: AI explains a strong relationship and proposes links worth making — interpretation rather than computation, needing Pro and your own API key."
---
<!--
  Markdown representation of https://docs.jotaid.com/en/guide/ai-suggestions
  Generated by scripts/gen-markdown-variants.mjs — do not edit by hand.
  Served from the canonical URL under `Accept: text/markdown`.
-->

# AI Interpretation & Suggestions

Everything else under Discover hands you **the relationship itself** — A relates to B, this strongly, through these shared concepts.

This layer is different: it hands you an **interpretation**. Plain language for what a relationship means, or an outright proposal about which links to make.

Which is why only this layer actually calls a language model, needing [Pro](https://docs.jotaid.com/en/guide/pricing) and your own API key.

## Explaining a Relationship

Select a bright cell in the [co-occurrence matrix](https://docs.jotaid.com/en/guide/matrix) and have the AI explain *why* those two concepts are strongly related — turning a number into a sentence you can read.

The matrix can tell you the two are strongly related. It cannot tell you "because whenever you discuss interview methods you end up on sampling bias". The second is what you will remember and use.

## Backlink Suggestions

The **SUGGESTIONS** section of a node's detail view: having read your notes, the AI proposes which semantically related notes — **worded differently** — belong linked to this node.

It deliberately **excludes** anything already covered by [unlinked mentions](https://docs.jotaid.com/en/guide/unlinked-mentions): that layer finds literal matches accurately and for free, so there is no point paying a model to repeat it. What appears here is what literal matching cannot catch.

## How Much to Trust Each Layer

| Layer | Based on | Confidence |
|-------|----------|-----------|
| [Unlinked mentions](https://docs.jotaid.com/en/guide/unlinked-mentions) | Exact literal match | Certain |
| [Co-occurrence](https://docs.jotaid.com/en/guide/matrix) | Appearing in the same note | Fact |
| [Link prediction](https://docs.jotaid.com/en/guide/predictions) | Your link structure | Explainable (you see the shared concepts) |
| [Semantic relations](https://docs.jotaid.com/en/guide/semantic-relations) | Closeness in meaning | The model's judgement |
| **AI interpretation** | A language model | **An interpretation** |

The further down, the more suggestive — and the more it needs your own review. The value of an AI suggestion is not in accepting it wholesale; it is that it proposes combinations you would not have reached. The judgement stays yours.

## When It Works Best

A handful of notes reveals nothing. It takes dozens of already-linked notes to give a model enough signal. So this layer becomes genuinely useful once a library has grown, while everything else under Discover works from day one.

## Privacy

Content goes to **the provider you configured**, only when you trigger a feature, never through a Jotaid server and never into training. See [AI Setup](https://docs.jotaid.com/en/guide/ai-setup).

## Next Steps

- [AI Features](https://docs.jotaid.com/en/guide/ai-features) — Everything else the AI does
- [AI Setup](https://docs.jotaid.com/en/guide/ai-setup) — Providers and keys
- [Semantic Relations](https://docs.jotaid.com/en/guide/semantic-relations) — Relatedness by meaning, at no cost