AI Verification

AI Verification lets you ask the agent to assess how well an individual mindmap node is supported by its citations — and to generate alternative ways of phrasing the same claim in four different tones. The original node is never changed. The results are added as a new subtree directly underneath the node you verified.

What it does

Given a node and its attached citations, the verification agent:

Reads all citations attached to the node — the actual passage text from your documents.
Assesses citation backing — determines whether each claim in the node text is directly supported, partially supported, or unsupported by the evidence.
Assigns a support score — a number from 0 to 100 reflecting overall citation alignment.
Identifies discrepancies — specific claims or phrases the citations do not support.
Generates four alternative phrasings of the node text — one per tone — for you to consider and optionally adopt.

All output is written into your mindmap as nodes, so you can navigate, copy, or build on it like any other content.

Invoking AI Verification

Select one or more nodes on the mindmap, then right-click to open the context menu and choose AI Verify. You can also trigger it by pressing F3 on your keyboard.

To verify a single node, select it and invoke verification. To verify several nodes at once, multi-select them — the agent will process each node sequentially, one after another.

While verification is running, a progress overlay appears on the canvas: "AI Verifying node X of N...". The mindmap remains readable but edits are paused until the batch completes.

The result tree

The verification output appears as a new subtree attached as a child of the verified node:

Your original node
└── AI Verification Result          ← colour reflects the support score
    ├── Citation Backing
    │   ├── Score: 74/100
    │   └── Discrepancies
    │       └── The claim about X is not directly supported by any attached citation.
    └── Alternative Phrasing
        ├── Formal
        │   └── [formal rephrasing]
        ├── Concise
        │   └── [concise rephrasing]
        ├── Accessible
        │   └── [accessible rephrasing]
        └── Assertive
            └── [assertive rephrasing]

The AI Verification Result root node is colour-coded on a red→yellow→green scale matching the support score, so you can see at a glance whether the evidence is strong or weak before drilling in.

Understanding the support score

The score is a 0–100 integer:

Score	Meaning
80–100	Every claim is well-supported by the attached citations
50–79	Mostly supported, with minor gaps or hedged claims
30–49	Partially supported — notable claims lack direct citation backing
0–29	Poorly supported, or citations are only tangentially related

A high score does not mean the claims are objectively true — it means they are consistent with the passages your citations contain. The quality of verification is bounded by the quality of your citations.

Discrepancies

The Discrepancies node lists specific problems the agent found — claims that are stated more broadly than the evidence supports, conclusions not present in any attached citation, or wording that mischaracterises the source. Each discrepancy is a single child node with a short description (under 300 characters).

If the node is fully supported, the Discrepancies node will have no children.

Discrepancies are grounded in the citation content provided. The agent does not flag stylistic issues or speculate about what citations say; it only flags what it can directly observe from the text.

Alternative phrasings

The Alternative Phrasing branch always contains exactly four rephrasings of your original node text:

Formal — precise, scholarly register suitable for an academic paper; uses technical vocabulary and complex sentence structure.
Concise — maximum information density, minimum words; removes all redundancy while preserving every claim.
Accessible — plain language for a general educated audience; avoids jargon and uses concrete expressions.
Assertive — confident, direct voice; states findings and conclusions strongly without hedging.

Each phrasing is a leaf node in the tree. To adopt one, you can move it in place of the original node, or use it as a starting point for manual editing. The original node is never touched by the agent.

When a node has no citations

If the node you verify has no citations attached, the verification still runs but the result reflects the absence of evidence:

Support score: 0/100
A single discrepancy: "No citations are attached to this node — citation support cannot be verified."
All four alternative phrasings are still generated (based on improving clarity and grammar of the original text, without introducing any new claims).

This is useful as a reminder to go back and add supporting citations, and still gives you the four rephrasing options even without citation backing.

Credits

Each AI Verification call uses credits, but it is optimized to minimize usage, usually consuming around one credit per node. The charge appears in your credits history as AI Verification. When verifying multiple nodes in a batch, each node is charged separately.