Trust & Governance

How FirstEdit handles your content

FirstEdit is a controlled first-pass editing layer in your workflow. It's not an unconstrained generative tool. Every part of how it handles documents and applies changes, reflects that human reviewers are in control and remain the irreplaceable part of the review process. This page shows every step of what that means in practice.

Jump to section

Control and oversight

FirstEdit is a tool that supports review teams. It removes mechanical work so that reviewers can focus their attention where it matters. Control stays with your people.

FirstEdit is designed to prepare documents for human review. It works within boundaries that you set. The only changes it makes are the ones that you instruct it to do.

You set FirstEdit's rules and boundaries.

FirstEdit applies changes according to the house style, terminology lists, and usage guidance you provide. The tool does not impose its own preferences — it applies yours.

FirstEdit does not replace editorial judgement.

FirstEdit does not rewrite, restructure, or make editorial judgements beyond the scope of the rules you set.

Outputs do not bypass review.

The workflow is designed so that FirstEdit feeds into human review, not past it. Questions of tone, clarity, structure, meaning, and audience remain with your editors and reviewers.

Under-editing

When FirstEdit is uncertain, it holds back. Caution is a feature, not a limitation.

Over-editing is a real risk with any AI tool. A model that applies changes confidently — even when it shouldn't — can introduce errorsThat require additional human review to find and fix.

How FirstEdit handles uncertainty
Apply Change is high-confidence, rule-governed, and independently verified. Applied as a tracked change for reviewer approval.
Query A potential issue is identified but context is ambiguous. Flagged as a reviewer query — no change made.
Skip FirstEdit is uncertain. No change is made. The text is left as written for the human reviewer to assess.
High-confidence changes only

FirstEdit targets changes it can make with high confidence. Every suggested edit is verified by a second AI model before it reaches your document. If verification fails, the change is dropped.

Limited scope for maximum accuracy

Unnecessary changes can introduce errors. Just the danger of AI introducing errors forces an extra human review. FirstEdit under-edits but it does not underperform. Its deliberately smaller scope makes its changes accurate and reliable.

Visibility and auditability

AI tools that makes changes silently create risk. Reviewers cannot manage what they cannot see. FirstEdit can return every suggested edit as a tracked change: auditable, reviewable, and reversible.

Example changes
Change "health care" → "healthcare" — applied per preferred terminology list
Query Acronym "EHR" used before definition — flagged for reviewer attention, no change made
Skipped Sentence structure ambiguous — no change applied, left for human reviewer
Change "e-mail" → "email" — applied per house style preference
Visible and Auditable

If you send FirstEdit a word document then every change is tracked in your document. If you access via browser then a review screen shows you each change. If you work with plain text then FirstEdit keeps a review log. No matter how you access it, every change is visible and auditable.

Clear explanations

Unlike other AI tools, FirstEdit does not operate as a black box. Edits can be linked to the rule that triggered them — so reviewers understand not just what changed, but why a change was made.

Rules and governance

Generic AI writing tools guess. They approximate rules from training data, apply broad conventions, and produce outputs that may or may not match your organization's requirements. FirstEdit is different: its behaviour is shaped by rules you define, not by probabilistic approximation.

House style enforcement

Your organization's style guide is the authority. FirstEdit reliably applies the specific rules your style guide specifies.

Preferred terminology

Defined terminology lists tell FirstEdit which terms are preferred, which are prohibited, and which are context-dependent. It applies these consistently across the document.

Organization-level configuration

Rules are configured at the organizational level. Different teams, document types, or client accounts can operate under different rule sets.

Summary of the approach

What this means in practice

When you deploy FirstEdit, you are not introducing an unconstrained AI agent into your document workflows. You are deploying a governed editing tool whose behaviour is defined by your rules, verified by a second AI model, reviewed by your team, and attributable at every step. That is the architecture of trust that serious document environments require — and it is how FirstEdit is built.