Guides
How to review the work AI handed you.
Written for the person who has to sign off, not for the person who generated it. Each guide is a procedure you can run this afternoon.
Same rule, every unit.
Engineering7 min
A practical checklist for reviewing AI-generated code
The failure patterns that show up again and again in assistant-written code, and how to catch each one before it reaches review.
Read the guideEditorial6 min
The signals in AI-written copy that readers actually notice
Which tells matter for comprehension and trust, which are noise, and the five-minute editorial pass that fixes the ones that count.
Read the guideMedia7 min
Image provenance, C2PA and what metadata can actually prove
What Content Credentials are, what they are not, and the offline signals worth checking on any image before you publish it.
Read the guideOperations6 min
A QA workflow for teams shipping AI-assisted documents
Where generated documents break, a fifteen-minute review pass that catches it, and the handoff rules that keep the record straight.
Read the guideTool guides
The tool behind each check, explained in full.
What the check catches, how to read what it returns, and the part of the job it cannot do.
- Prose & editingThe inline AI highlighter: editing against marks instead of reportsHow to use an editor that marks AI tells as you type: what each colour asks you to do, which density bands matter, and the five-minute pass that fixes a draft.Read
- Prose & editingThe read-only word highlighter: seeing a draft before you touch itWhy a read-only pass beats an editor for judging a draft, what five pattern sets catch, and how to read a density figure without turning it into a verdict.Read
- Prose & editingThe rewrite diff viewer: what humanising a draft actually changesWord-level diffs between an original and its rewrite, the four things worth checking in the result, and how to keep an edit history a client can audit.Read
- Prose & editingThe markup pad: a plain box that marks text in placeWhen a bare contentEditable pad beats a full editor, why marking on demand is calmer than live marks, and the one thing to remember before copying text out.Read
- Claims & sourcesComparing an AI summary with its source, line by lineSummary drift is quieter than invention. How a word-level diff plus a figure-and-quote trace catches the edits that survive a fluent read.Read
- Claims & sourcesThe claim review editor: attaching a verdict to the sentenceWhy check-the-numbers notes fail, how three verdicts turn a review into something countable, and what to do with the claims nobody has ruled on.Read
- Claims & sourcesChecking the markdown a model hands you before it becomes a pageRendered markdown is the only view that counts. The broken citations, dead relative links and collapsed tables that survive a copy-paste into production.Read
- Claims & sourcesThe JSON inspector: catching schema drift before your code doesA model returns plausible JSON that fails quietly downstream. How to check required fields, types and undeclared keys, and where a structural check stops.Read
- Visual assetsThe 400% rule: inspecting generated charts and mockups properlyWhy a chart that passes at page size fails under magnification, the six details worth checking at 400%, and what zoom can never tell you about the numbers.Read
- Visual assetsThe wipe test: what changed when you asked for one editImage models regenerate the whole frame. How a wipe between original and edit exposes moved objects, rewritten packaging and reflections of things that no longer exist.Read
- Visual assetsRunning generated React before you trust a word of itWhy reading generated components proves nothing, the four failures a live mount exposes, and the one tool in the box that is not offline.Read
- Inline annotationBuild your own phrase list, then make it earn its placeGeneric tell lists stop teaching after a week. How to write literal and regex rules for the repeats only your drafts have, and when to delete a rule.Read
- Inline annotationClichés, dead citations and unsourced figures as badges you can openTurning the flag into a component inside the sentence: what the badge explains, which fixes a machine can apply, and the ones it must leave to a human.Read
- Inline annotationConnectives are the structural tell, and they are easy to deleteWhy the clearest signal of generated prose sits at the start of a paragraph, how paragraph-level scoring works, and what an opener share does not measure.Read
- Inline annotationSentence diagnostics that stay where the sentence isTooltips instead of a side panel, the four checks worth interrupting a read for, and how burstiness describes rhythm without judging authorship.Read
- Records & rulesTurning a draft into a claim ledger you can actually finishReading for accuracy fails for a boring reason: it is unstructured. How to work a matrix of every figure, name and link, and what a verified status owes.Read
- Records & rulesGuardrails: the rules a draft must pass before it publishesSix checks that stop the most common AI publishing failures, why measuring beats declaring, and how much a passing validation is really worth.Read
- Records & rulesWhat an AI validation certificate is actually worthA timestamped, numbered PDF that records the reviewer, the rules and the claims with their sources, and names exactly what it does not attest.Read
- Documents & dataTracing a summary's quotations back to the page they came fromPage references are the cheapest thing to get wrong and the hardest to notice. How a side-by-side viewer and a quote trace make attribution checkable.Read
- Documents & dataChecking an entire CSV export for duplicate and thin copyEvery row reads fine on its own. How four-word shingle comparison, cross-row phrase counts and a word floor expose scaled content in an export.Read
- Documents & dataTurning a schema into a form, then validating output against itA prose rulebook becomes executable when it is JSON Schema. How generated forms catch the field-level failures a pipeline otherwise inherits silently.Read
- Audio & videoListening to an AI voiceover with measurements, not impressionsPitch range, breath gaps and cadence turn a voiceover review into three numbers you can compare, plus the traits no measurement captures.Read
- Audio & videoStepping through a generated video, frame by frameGenerated video hides its failures between the frames you notice. How timecoded notes, frame stepping and a review log turn a watch into an inspection.Read
- Audio & videoReading a file's own account of where it came fromC2PA manifests, EXIF tool strings and container blocks, read offline: what the bytes can prove, what they merely suggest, and how metadata gets stripped.Read
- Audio & videoCropping in on the detail that gives the generation awayArtifacts cluster in hands, ears, text and edges. How to crop and magnify a region to a verdict, and why a zoomed crop beats squinting at the full frame.Read
Put it to work
Reading about it is slower than running it.
The console applies the same rules these guides describe, in under a second, without uploading anything.