How FlowProofs uses AI
Last updated: October 24, 2026
FlowProofs uses AI for a few optional writing and drafting features. Its core, explaining why a number changed, is arithmetic, not AI. This page says exactly where AI is used, what it sees, and how to switch it off.
1. What is AI and what is not
- Not AI: every number, formula and chart; Explain’s breakdown of a change into what each input contributed (a calculation called the Shapley method, which always adds up exactly); the bridge chart; whether a change is unusual; and the sentence Explain writes by default, which is filled in from the numbers.
- AI: the features below, and only when you use one.
2. Features that use AI
| Feature | What it does | Runs when |
|---|---|---|
| Phrase with AI (Explain panel) | Rewrites Explain's computed sentence more naturally. A check rejects any sentence with a number that is not in the explanation | You click Phrase with AI |
| AI assistant (canvas) | Answers questions and draws shapes, arrows and variables on the board for you to edit | You send it a message |
| Meeting answers and summary | Writes a plain-language answer to a question, or a summary at the end of a meeting | You ask, or open the summary |
| Story draft | Drafts the stops and narration for a walkthrough of your canvas | You choose Generate summary in the story panel |
| Formula from a description | Suggests a formula for a data table column, or a variable, from a sentence | You describe one and ask |
| Report import | Reads the text of a PDF or Word report and suggests the numbers and structure to import | You import a report |
Nothing runs in the background, and nothing is sent to the AI provider unless you use one of these features.
3. The provider, and what is sent
These features run on Claude models from Anthropic PBC (United States), called from our servers; your browser never talks to Anthropic. Each request carries only what that feature needs: for example the names, values and formulas of the blocks on a canvas, or the text of a report you import. Never your whole account, and never table rows unless the feature needs them.
4. Your data is not used for training
Customer data is not used to train AI models. We do not train models on it, and under Anthropic’s commercial terms Anthropic does not train its models on content sent through its API by business customers like us. Anthropic keeps API requests only for a limited time, for safety and abuse monitoring, as its terms and privacy center describe.
5. How AI text is labelled
- Every piece of AI-written text shows an AI-generated label. Its tooltip says: “Written by AI from your data. Check important numbers.”
- It is also marked for software: a
data-ai-generated="true"attribute on the page,"generatedBy": "ai"in saved stories, a note inside exported PNG images, and a line at the end of printed pages that contain it. - The assistant says plainly that it is an AI, not a person.
- If you rewrite an AI-drafted story stop yourself, the label goes, because the words are then yours.
6. Turning AI off
Go to Settings → Data and privacy → AI features and turn it off. Every AI feature then refuses before anything is sent, for your account. Everything else, Explain’s numbers and breakdowns included, keeps working. Organisations that want AI off for every account can write to hello@flowproofs.com.
7. You stay responsible for decisions
AI text can be wrong, even when it sounds sure. FlowProofs is analysis, not financial, accounting, tax, investment or legal advice, and AI output does not decide anything. Check the numbers that matter before you act on them; the Terms of Use say more. Rules for using the AI features are in the Acceptable Use Policy.
8. Questions
Write to privacy@flowproofs.com. The Privacy Policy and the sub-processors list cover the rest.