A good editing checklist turns a vague "give it a once-over" into a sequence of focused passes, each hunting one class of problem. That structure matters because AI drafts fail in predictable ways, and the human eye can't catch facts, structure, voice, and typos all at once. Run the passes in order (accuracy first, polish last) and you fix substance before you sweat surface, so you never waste effort perfecting a sentence you're about to cut.

This is the operational companion to quality and editing: where that piece explains why editing is the highest-leverage skill, this one is the thing you actually run on every draft before it ships. Each section below is a pass. Each item has a reason. Work top to bottom.

1. Accuracy

Run this first, before touching a single sentence of prose. Verification is the highest-risk pass; everything else is wasted if the underlying claims are wrong. A fluent draft can be confidently false with no tell in the writing, so assume nothing and check everything that's checkable. (The full method is in how to fact-check AI-assisted content.)

Every statistic is traced to a primary source. Invented numbers read as more credible than the truth they replaced. Find the original dataset, not a secondary summary.
Every citation has been independently located. Models fabricate references that look completely real. If you can't find it, treat it as fabricated.
Quotes are verified for wording and attribution. Models paraphrase as if verbatim and misattribute. Confirm both.
Dates, names, and timelines are confirmed. Subtle errors (a year off, a reversed cause) survive a casual read precisely because they're plausible.
Time-sensitive facts are current as of today. Prices, versions, officeholders, and "the latest" go stale silently.
Unverifiable claims are cut. An absent fact beats a wrong one, every time.

2. Structure

With facts settled, make sure the reader reaches your best material. Structure decides whether a great point in paragraph nine ever gets read. Fix this before line-level work: there's no use polishing a paragraph you're about to move or delete.

The opening answers the core question in the first two sentences. Drafts default to throat-clearing intros. Lead with the answer; a buried point is a point most readers never see.
Each section owns one idea. One H2, one job. Sections that try to do three things should become three sections, or lose two.
Headings are phrased the way people ask things. Question-form headings map onto real queries and make the piece scannable.
The order follows a curious reader's questions. Sequence sections the way someone learning the topic would naturally ask, not the order the draft happened to generate them in.
Transitions earn their place. Cut any transition that only restates the last paragraph before introducing the next.

3. Voice

This is where a draft stops sounding like its generic default and starts sounding like a person who knows the subject. (The cornerstone explains why the default is so flat: a draft gravitates toward the average of everything written on a topic.) Voice is what makes a reader trust there's judgment behind the words.

The flat default register is replaced with a consistent one. Drafts strip voice out because averaging every voice together leaves none. Put a real one back, deliberately.
Hedging is removed where evidence supports a direct claim. "This can sometimes potentially be a factor" asserts nothing. State what's true plainly. Confident, self-contained claims are also what answer engines can actually quote.
There's a genuine point of view. Say which option is best, what's wrong, what matters most. Refusing to take a position is the most generic move there is.
Clichés and filler phrases are cut. "In today's fast-paced world," "it's important to note," "when it comes to": the verbal tics of voiceless prose. Delete them.
Rhythm varies. Drafts fall into uniform sentence length. Mix short and long so it reads like speech, not a metronome.

4. Specificity

The fastest credibility upgrade available. Specific claims are more useful, more trustworthy, and (not incidentally) harder for anyone else to have written, which is its own competitive moat.

Vague generalities become concrete examples. "Many companies struggle with this" becomes a specific situation a real reader recognizes.
Abstractions get numbers, names, and details. Where you have a real figure, a named framework, or a concrete instance, surface it. Specificity is what original content is made of.
Claims are checkable, not just impressive. A statement you could verify beats a superlative you can't. "The largest" needs evidence; a concrete description doesn't.
Examples are real, not invented. Don't fabricate a case study to add specificity; that just reintroduces an accuracy problem. Use real ones or none.

5. Cuts

Last in sequence among the substance passes, highest in leverage. As a rule of thumb, many drafts run something like a quarter to a third too long, and the excess is reliably the weakest material. Cutting raises the average quality of everything that survives.

Every sentence is tested: does removing it lose anything real? If not, it goes. This single question does most of the work.
Padding is gone. Restated topic sentences, summary paragraphs that recap what was just said, conclusions that add nothing.
Redundancy is collapsed. Two sentences making the same point become one. Three examples proving an obvious claim become one.
The piece is shorter than the draft. If your edit didn't remove a meaningful fraction of the words, you probably edited too gently.

6. SEO & GEO

Once the content is right, make sure it can be found and cited. These aren't tricks; they're the same disciplined structure that already makes a piece good, pointed at discoverability. (The full playbook is in generative engine optimization.)

Key claims are extractable. Self-contained, quotable sentences are what answer engines lift into responses. A point that needs three paragraphs of build-up can't be cited.
Definitions stand alone. "X is…" sentences are among the most-cited passages because they're unambiguous out of context.
Lists, tables, and an FAQ package claims for extraction. Structured formats are easy for both readers and models to parse and reuse.
The target question is genuinely answered, not keyword-stuffed. Write for the question a person actually asked. Engines and readers both detect and discount stuffing.
Internal links connect the topic cluster with descriptive anchors. Linking related pages signals depth and tells the reader (and the model) what's on the other side.

7. Polish

Last, only after content has settled. Proofreading first means perfecting sentences you may delete in pass five: wasted effort. Now that nothing else will change, clean the surface.

Grammar and typos are fixed. The basics, done once, at the end.
Formatting is consistent. Heading levels, list styles, capitalization, and terminology are uniform throughout.
All links work and point where they should. Broken or wrong links undercut everything the rest of the edit earned.
Final read-through, start to finish. One last pass at natural reading speed catches what isolated passes missed: the awkward seam, the dropped word, the section that still doesn't flow.

The pass order, at a glance

The sequence is the whole point. Run it the other way and you waste effort at every step.

Pass	Targets	Why this order
1. Accuracy	Hallucinated facts, fake citations	Don't polish what you may delete
2. Structure	Buried points, meandering flow	Fix the skeleton before the skin
3. Voice	Sameness, hedging	Make it sound human before refining lines
4. Specificity	Generic filler	Add substance once structure holds
5. Cuts	Padding, redundancy	Trim before perfecting wording
6. SEO & GEO	Discoverability, citability	Optimize finished content, not a draft
7. Polish	Typos, formatting, links	Surface last, after content settles

Where to go next

Quality & editing: the human-in-the-loop: why editing is now the highest-leverage skill, and what drafts get wrong.
How to fact-check AI-assisted content: the deep version of the accuracy pass, where the real risk lives.

Print it, pin it, run it every time. A checklist only works when it's a habit, and the habit is what separates work that publishes from work that reads like everyone else's.

An Editing Checklist for AI Drafts