AI Grammar and Clarity Review

Flag grammar, readability, and ambiguity problems without rewriting the author's voice.

moderate AI Content

What it adds

An inline review pass that marks suspected issues in a document with an explanation and a suggested fix the author applies one at a time.

What your agent is told to do

1
Return findings anchored to positions in the text, each with a category, a short reason, and a proposed replacement. Do not return a rewritten document; the author applies fixes individually.
2
Separate confident mechanical errors from judgement calls about clarity, and present them differently. Grouping a subject-verb disagreement with a suggestion to shorten a sentence trains the author to dismiss both.
3
Read from a workspace dictionary of product names, technical terms, and accepted spellings before flagging anything, and give the author a way to add a term from the finding itself.
4
Explain any suggestion that changes emphasis, structure, or meaning rather than offering a silent replacement, so the author can tell a correction from an opinion.
5
Wholesale restructuring belongs to AI Text Rewriting and register changes belong to AI Tone Adjustment. This feature owns detection and never applies a change the author did not accept.

Edge cases it handles

Deliberate style is not an error. Sentence fragments, informal address, and unconventional punctuation used consistently should be left alone or flagged at the lowest confidence, never corrected automatically.
Product names, code identifiers, industry jargon, and internal acronyms will be flagged as misspellings unless the workspace dictionary is consulted first and additions take effect immediately.
High-impact suggestions must carry a reason. A replacement offered without explanation is indistinguishable from a rewrite and gets accepted without thought.
Formatting must survive the review. Applying a fix must not flatten a link, break a list, drop emphasis, or disturb code blocks and quoted passages.
Mixed-language and non-English content must be detected and either reviewed in that language or skipped with a clear message, rather than being flagged wholesale as errors.
Positions drift as the author edits. Re-anchor or invalidate findings when the underlying text changes, so a fix is never applied to the wrong span.
Long documents exceed the token ceiling. Review in sections with a documented boundary, show progress, and never present a partial review as a complete one.
A refusal, a timeout, or malformed findings must leave the document unmarked and unmodified, with the failure stated rather than shown as a clean bill of health.

Definition of done

Findings are anchored to text positions and carry a category, a reason, and a suggested fix.
Mechanical errors and clarity judgements are visually and textually distinguished.
Terms in the workspace dictionary are never flagged, and terms can be added from a finding.
No change is applied to the document without the author accepting that specific finding.
Applying a fix preserves links, lists, emphasis, and code formatting.
Non-English or mixed-language content is handled explicitly rather than flagged as error.
Findings are invalidated or re-anchored when the underlying text is edited.
The feature matches the existing design system.
No existing functionality is broken.

Related features

AI Image Alt Text

AI FAQ Generator

AI Translation

How it works

1

Copy the link

Grab the Markdown instruction URL for this feature.
2

Give it to your AI

Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.
3

It inspects, then implements

Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.