This AI SEO editing checklist is for the person holding the red pen after an AI tool has written the draft. AI drafts read smoothly, and that is exactly what hides their faults: a title that promises one page and a body that delivers another, a main point repeated four times, numbers nobody checked and links that say “click here”.

The nine passes below catch those faults line by line, each pass with its own question and method. The last section runs all nine on one paragraph, with a change log showing which pass caught what.

Summary

Edit an AI SEO draft in nine passes. Pass 1 makes the title, opening and headings agree on what the page is for. Passes 2 to 5 fix intent gaps, unsupported claims, repeated ideas and weak links, and the line edit, capitalisation, metadata and one last read follow in passes 6 to 9. Polishing before pass 3 wastes time on sentences that get cut.

Why edit an AI draft in passes?

One read that tries to catch everything catches the typos and misses the structure. Passes fix one kind of problem at a time, in the order that removes the most text first, so no edit gets spent on a sentence that a later pass deletes.

PassThe questionTypical fix in an AI draft
1Do the title, opening and headings agree on what the page is for?Rewrite the opening or the H1 so both name the same job
2Which searcher questions are missing or answered too late?Add a missing answer, move a buried one up, cut an off-topic section
3Is every factual claim supported, softened or removed?Link a source, narrow a claim, delete what nobody can source
4Which ideas appear twice?Merge paragraphs, delete summary sentences
5Which links are weak?Rewrite anchors, point links at the exact page, add the next-step link
6Which sentences sound machine-made?Cut restated openings, reflex triplets, empty intensifiers
7Is capitalisation consistent?Pick sentence case or title case and apply it everywhere
8Do the title tag, meta description and alt text match the edited page?Rewrite metadata after the body has changed
9Does it read well from top to bottom?One uninterrupted read in preview; mark, then fix

The first five passes change what the page says, and the last four change how it says it. Keep that split, and the editing time goes where the risk is.

Do the title, opening and headings agree on what the page is for?

AI drafts assemble the three parts from different places: the title from the keyword, the opening from a generic version of the topic, the H2s from competitor outlines. Each part reads fine alone, and together they promise three different pages.

The test takes two minutes:

  1. Write the page’s job in one sentence: “This page helps [reader] do [task].”
  2. Compare it with the H1, the first two sentences and the list of H2s.
  3. Fix whichever part disagrees. If the H2s describe a different job than the H1, rewrite the H1, because the H2s hold the content.

Google’s self-assessment questions include whether the main heading or page title provides a descriptive, helpful summary of the content. Pass 1 makes the answer yes before any line gets polished.

Where are the intent gaps?

An intent gap is a question the searcher brought that the draft skips, answers late or answers for a different reader. List the sub-questions first, then mark the draft against them.

  • Collect the questions: the People Also Ask box and related searches for the target query, plus the H2s of the top five results.
  • Mark each one: answered, answered late (below the fold when it belongs at the top), or missing.
  • Fix in order: add the missing answers, move the late ones up, and cut any section that answers a different query, because that section belongs on another page.

Example: a draft for “AI SEO editing checklist” that spends 400 words on the history of AI writing and never shows an edited paragraph has two gaps and one section to cut.

Is every factual claim supported, softened or removed?

This pass makes sure every factual claim in the draft is supported, softened or removed before anyone polishes the sentences around it. Work line by line and highlight every sentence that contains:

  • a number, a price, a percentage or a date;
  • a named person, company, study, tool or feature;
  • a superlative (“the best”, “the first”, “the only”);
  • a borrowed authority (“studies show”, “experts agree”, “Google says”).

Each highlight gets a verdict. Supported means you found the source and linked it. Softened means the source supports a narrower statement, so the sentence shrinks to match. Removed means nobody could source it, so it goes, and the paragraph around it gets repaired. Softening is the verdict editors forget, and AI drafts need it most, because they round every finding up:

As draftedWhat the source saysSoftened version
“Surfer’s Content Score tells you how well your article will rank.”Surfer’s help center: a combined measure of optimisation against top-performing pages; aim 10 to 20 points above competitors instead of chasing 100“Surfer’s Content Score compares your draft with top-ranking pages for the query, on a scale of 0 to 100.”
“Google requires sentence case in headings.”Google’s developer documentation style guide uses sentence case for titles and headings in its own docs“Google’s developer documentation style guide uses sentence case for headings.”
“Clearscope grades every article from A++ to F.”Clearscope’s editor guide names A++ as the highest Content Grade“Clearscope’s Content Grade tops out at A++.”

For articles that need a written record of every claim, use the fact-check log in human review for AI SEO articles.

Which ideas does the draft repeat?

AI drafts restate their main point in the opening, at the end of each section, in the conclusion and again in the FAQ. Each restatement reads as reasonable, and the page grows longer with nothing new in the extra length.

The margin-label method finds them fast:

  1. Write a five-word label next to every paragraph (“Clearscope grade is term coverage”).
  2. Any label that appears twice marks a repeat. Keep the stronger paragraph and delete the other.
  3. Delete every sentence that summarises the section it ends.
  4. Check the FAQ against the body: an answer that repeats a section gets replaced by a question the body does not answer.

Links tell readers and search engines where the page sits on your site. Google’s link guidance gives the quickest test: read only the anchor text, out of context, and check that it makes sense by itself. The same page warns against cramming keywords into anchors and against chaining links next to each other.

Weak linkHow to spot itFix
Generic anchor“Click here”, “read more”, “this article”Name the destination: “the fact-check log template”
Stuffed anchorFive keywords in one linked phraseKeep the words a reader needs to know where they are going
Homepage linkThe anchor names a topic, the URL is a domainLink the exact page that covers the topic
Thin destinationThe linked page is shorter or weaker than the sentence promising itLink a better page or remove the link
Chained linksSeveral links side by side with no words between themSpread them out, each in its own sentence
Missing next stepThe reader finishes a section with nowhere to goAdd one link to the page that answers their next question

AI link suggestions bring their own problems: the wrong page, the same anchor on ten articles, links to pages that should not rank. Checking them is covered in internal linking for AI content.

What do you fix in the line edit?

Only now do sentences get polished, because only now do you know which ones survive. AI drafts have line-level habits that readers notice even when they cannot name them:

  • The restated opening: the first sentence repeats the heading in different words.
  • Reflex triplets: three adjectives or three examples, every time, whether or not three exist.
  • Empty intensifiers: words that make a sentence sound bigger without adding a fact.
  • Hidden actors: passive sentences where nobody does anything (“mistakes were made in the review”).
  • Same rhythm: sentence after sentence of the same length and shape.
  • Hedges on checked facts: words of possibility attached to a number you verified in pass 3.

Keep a list of the words and phrases your site bans and search the draft for each one; it is faster than reading for them. A grammar tool helps with the rest: Grammarly’s free plan covers spelling and grammar mistakes. Neither tool tells you that paragraph four says what paragraph two said.

When do you check capitalisation and title case?

Capitalisation comes after the title, opening and headings agree on what the page is for, because passes 1 to 5 rewrite headings. Casing them earlier wastes the work. Run it as one step, across every element at once.

First pick a style, then apply it everywhere:

  • Sentence case: capitalise the first word and proper nouns only. Google’s developer documentation style guide uses it for document titles and headings, and the Microsoft Writing Style Guide says titles of blog posts, documentation articles and press releases use sentence-style capitalisation.
  • Title case: capitalise most words. Microsoft’s rules for it: always capitalise the first and last words, lowercase a, an and the unless they come first, and lowercase prepositions of four or fewer letters unless they come first or last.
ElementCheck
H1 and title tagSame style as each other, and as every other article on the site
H2s and H3sOne style for all; AI drafts mix a Title Case H1 with sentence case H2s
FAQ questionsSame style as the headings, since they render as headings
Table headers, captions, buttonsSentence case reads cleaner in small text
Brand and product namesKeep their own capitals: WordPress, Search Console, Content Score
AcronymsSEO, FAQ, AI stay uppercase in either style

Google’s title link documentation asks for descriptive, concise titles and says nothing about capitalisation, so pick the style your site already uses. A title case converter does the mechanical part; the editor handles the exceptions, such as brand names and hyphenated words.

What is left for the metadata and the last read?

Pass 8, metadata. The title tag, meta description and image alt text were written for the draft, and the draft has changed. Google’s guidance on generative AI content extends the manual review to title elements, meta descriptions, structured data and alt text, because all of them can appear in search results. Rewrite each one against the edited page.

Pass 9, the last read. Read the whole page once in the CMS preview without stopping. Mark problems in a list as you go, then fix them all at the end. Stopping to fix mid-read breaks the flow you are trying to judge.

What does a before/after edit look like?

Here is one paragraph of an example AI draft, the same paragraph after the nine passes, and a log of every change. Prices and features in the edited version come from the vendors’ pages, checked on 5 November 2026.

Before (example AI draft):

SEO Content Tools: Why They Matter. SEO content tools help writers create better content. Studies show that pages graded by Clearscope rank three times higher than pages without it. Clearscope plans start at $99 a month, and every article receives a grade from A++ to F. Surfer works in a similar way, and its Content Score tells you exactly how well your page will rank. In short, SEO content tools help writers create better content that ranks. Click here to see the full list.

After the nine passes:

What does an SEO content tool check? It compares your draft with the pages that already rank for your query. Clearscope’s Content Grade rises as the draft includes more of the relevant terms found in top-ranking content, with A++ as the top grade, and its Essentials plan costs $129 a month (pricing page, checked 5 Nov 2026). Surfer’s Content Score runs from 0 to 100 against top-ranking pages, and Surfer advises aiming 10 to 20 points above competitors instead of chasing 100. Neither score checks whether your facts are true, so the claim check stays with the editor. The AI content quality score guide compares what each tool measures.

#Problem in the draftPassEdit
1Heading in Title Case that says nothing on its own1, 7A question, in sentence case
2Opening restates the topic4, 6Replaced with what the tool does
3“Three times higher” with no study behind it3Removed
4“$99 a month”3Corrected to $129 from the pricing page
5“A++ to F”3Softened to what the docs confirm
6“Tells you exactly how well your page will rank”3Softened to what Surfer’s docs say
7“In short…” repeats sentence two4Deleted
8“Click here”5Descriptive anchor to the next page
9Nothing on what the scores miss2One sentence added: tools do not check facts

The edited version runs about 30 words longer, and every claim left in it has a source. Four of the nine changes came from pass 3, as you would expect in a paragraph about tools, where every price and feature claim needs a source.

Editing sits between the draft and the final review.

Related

Human review for AI SEO articles: reviewer roles and a claim-by-claim fact-check log.

Related

AI content quality score: score the edited article out of 100 before publishing.

Related

AI content outline: fix the structure before the draft, so pass 1 has less to do.

Related

Internal linking for AI content: checking the links AI tools suggest.

Related

AI SEO article quality checklist: the short pre-publishing list.

FAQ

Should SEO headings use sentence case or title case?

Either works for search, as long as the whole site uses one. Google’s title link documentation asks for descriptive, concise titles and does not mention capitalisation. Sentence case is easier to apply consistently, which is why Google’s and Microsoft’s own style guides use it for headings.

When should an editor send an AI draft back instead of editing it?

When pass 1 shows the draft answers the wrong question, or pass 3 removes so many claims that whole sections fall apart. At that point a better brief and a new draft cost less than rebuilding the old one line by line.

What does softening a claim look like?

Shrinking the sentence to the scope of its source. “Clearscope grades every article from A++ to F” becomes “Clearscope’s Content Grade tops out at A++”, because the docs confirm the top grade and nothing more. The sentence stays useful and stops claiming more than anyone checked.

Can an AI tool run this editing checklist on its own draft?

It can help with passes 4 and 6: flag paragraphs that repeat each other, find banned words, vary sentence length. Pass 3 stays with a person, because confirming a claim means opening the source, and a model checking its own claims checks them against the same memory that produced them.

How is the editing checklist different from a quality score?

The checklist changes the draft; a score grades the result. Run the nine passes first, then score the edited article with the AI content quality score rubric to decide whether it is ready to publish.

What should you do next?

Take your newest AI draft and run pass 1 and pass 3 only. If the title and headings disagree, or a third of the claims cannot be sourced, you have found where your editing time should go.