In short: Review an AI draft in a fixed order: first facts and sources, then substance and your own experience, then structure, then the SEO elements (title, meta description, headings, internal links, alt texts, structured data), then brand voice, and finally a read-through in the preview. Google asks for exactly this kind of manual check, metadata included. The 15 points below tell you what to look for and how to fix it.
What a review of an AI draft is for
A language model writes fluent text quickly. It does not know your customers, your prices or what changed last month, and it will sometimes state things with confidence that are simply wrong. The review is where a person with real knowledge turns the draft into something worth publishing.
Google says this plainly in its guidance on generative AI: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing." The same page adds that the review also applies to "<title> elements, meta description elements, structured data, and alternate texts for images" (Google: using generative AI content (external link)).
This guide is the hands-on editing checklist. If you want the background on Google's policy first (spam, scaled content, E-E-A-T, authorship), read our guide Does Google penalize AI content?.
The checklist at a glance
Work through the points in this order. Facts come first because a wrong number makes every later polish pointless.
| # | Check | What to look for | Quick fix |
|---|---|---|---|
| 1 | Numbers and claims | Figures, prices, dates, names, quotes without a source | Verify each one or delete it |
| 2 | Sources | Studies or links that do not exist or say something else | Open every source; link the primary one |
| 3 | Up to date | Old versions, past years, expired rules | Check against the current official page |
| 4 | Complete answer | Does the post answer the question in the title? | Add the missing sub-question as a section |
| 5 | Own experience | No example only you could give | Add one real case, photo or number of your own |
| 6 | Repetition and padding | Same point in intro, body and conclusion | Say it once, cut the rest |
| 7 | Answer first | Long warm-up before the actual answer | Move the answer into the first 2-4 sentences |
| 8 | Headings | Vague or decorative headings, one H1 | Turn each H2 into a real question or step |
| 9 | Title | Generic, too long, keyword repeated | Descriptive, specific, unique |
| 10 | Meta description | Keyword list or copy of the intro | One or two sentences that sum up the page |
| 11 | Internal links | No links, or "click here" anchors | 2-4 links to related pages with descriptive anchors |
| 12 | Image alt texts | Empty, stuffed or not matching the image | Describe what is actually in the picture |
| 13 | Structured data | Markup that does not match the visible post | Match author, dates and headline to the page |
| 14 | Brand voice | Wrong form of address, filler phrases, words you never use | Rewrite in your tone; delete stock phrases |
| 15 | Final read in the preview | Broken links, formatting, missing author | Read once as a visitor would, then approve |
Facts and sources
1. Check every number and claim
What to look for: percentages, prices, dates, product names, legal or technical statements and quotes. AI models can produce numbers that look precise but have no source behind them. Google's own quality questions include: "Does the content have any easily-verified factual errors?" (Google: creating helpful content (external link)).
Quick fix: mark every figure in the draft. Verify it against a reliable source, or delete the sentence. If a claim is important but you cannot verify it, replace it with what you do know from your own work.
2. Open every source
What to look for: links that lead nowhere, studies with a plausible title that you cannot find, or a real source that says something different from the draft.
Quick fix: click every link. Prefer the primary source (the official documentation, the law text, the manufacturer's page) over a blog that quotes it. Google asks whether content that draws on other sources "avoid[s] simply copying or rewriting those sources" (Google: creating helpful content (external link)), so add what the source means for your reader instead of restating it.
3. Check that it is up to date
What to look for: a model's knowledge stops at a training date. Watch for "the latest version", "this year", old prices, funding rules or software menus that may have changed.
Quick fix: check time-sensitive statements against the current official page and avoid vague words like "currently" unless you can name the date of the source.
Substance and own experience
4. Does the post answer the question completely?
What to look for: compare the draft with the title. Google asks: "Does the content provide a substantial, complete, or comprehensive description of the topic?" and "Does the content provide insightful analysis or interesting information that is beyond the obvious?" (Google: creating helpful content (external link)).
Quick fix: list the follow-up questions a customer would ask (costs, duration, risks, alternatives). Add a short section for each one the draft skips.
5. Add your own experience
What to look for: a draft that could appear on any website. Google asks whether content "clearly demonstrate[s] first-hand expertise and a depth of knowledge" (Google: creating helpful content (external link)). This is the one thing a model cannot supply.
Quick fix: add at least one thing only you can say: a typical customer question, a mistake you see often, a short case from a real job, your own photo or a price range you actually charge.
6. Cut repetition and padding
What to look for: the same point in the introduction, in the body and again in the conclusion; sentences that announce what comes next; a summary of a section that was three lines long.
Quick fix: keep the strongest version of each point and delete the others. Google asks whether you are "writing to a particular word count" and answers in the same place that it has no preferred length (Google: creating helpful content (external link)).
Structure and readability
7. Put the answer first
What to look for: AI drafts often open with general statements ("In today's fast-paced world ...") before they get to the point.
Quick fix: answer the question in the first two to four sentences. Then explain who the answer applies to and go into detail. Short paragraphs, lists for steps and a table where readers compare options.
8. Make the headings do work
What to look for: exactly one H1, and H2 and H3 headings that match real sub-questions. Decorative headings like "Unlocking the potential" tell the reader nothing.
Quick fix: rewrite each heading so that a reader who only skims the headings gets the outline of the answer. Google asks whether "the main heading or page title provide[s] a descriptive, helpful summary of the content" (Google: creating helpful content (external link)).
SEO elements
9. Title
What to look for: generic titles, the keyword repeated, or a title that promises more than the post delivers. Google recommends "descriptive and concise" title elements and warns against "repeated or boilerplate text" (Google: title links (external link)).
Quick fix: name the topic and the benefit in plain words, once. Check length and wording with our free title checker.
10. Meta description
What to look for: a list of keywords, a copy of the first paragraph, or the same description as on other posts. Google notes that "Identical or similar descriptions on every page of a site aren't helpful" and that long strings of keywords "don't give users a clear idea of the page's content" (Google: snippets and meta descriptions (external link)).
Quick fix: one or two sentences that sum up what the reader gets from this page. Write and preview it with the meta description generator.
11. Internal links
What to look for: a post with no links to your other pages, or anchors like "click here". Google writes: "Good anchor text is descriptive, reasonably concise, and relevant to the page that it's on and to the page it links to" (Google: link best practices (external link)).
Quick fix: link to two to four related pages, for example the matching service page and a related guide, with anchor text that says what the reader will find there.
12. Image alt texts
What to look for: empty alt texts, alt texts stuffed with keywords, or a description of an image that was later replaced. Google advises: "Avoid filling alt attributes with keywords" (Google: image SEO best practices (external link)).
Quick fix: describe in a short sentence what is actually in the picture, in the language of the page.
13. Structured data
What to look for: if your site outputs structured data (for example Article markup from an SEO plugin), check that headline, author and dates match the post. Google's rules: "Your structured data must be a true representation of the page content" and "Don't mark up content that is not visible to readers of the page" (Google: structured data guidelines (external link)).
Quick fix: after the final edit, compare the markup with the visible page. Remove FAQ or how-to markup for content that is no longer on the page.
Brand voice
14. Make it sound like you
What to look for: the wrong form of address, a tone that is too formal or too salesy, words you would never use, and stock phrases that appear in many AI texts ("it's not just X, it's Y", "in today's digital landscape", "let's dive in"). Google also asks: "Does the content have any spelling or stylistic issues?" (Google: creating helpful content (external link)).
Quick fix: keep a short list of words to avoid and delete them on sight. Read two paragraphs out loud: if you would not say it to a customer, rewrite it.
Final checks
15. Read it once in the preview, then approve
What to look for: broken links, formatting errors, a missing featured image, and a missing author. Google asks: "Is it self-evident to your visitors who authored your content?" (Google: creating helpful content (external link)).
Quick fix: read the post in the preview as a visitor would, click each link once, and make sure a real, named person stands behind it. Where readers might wonder how the post was made, Google says that "AI or automation disclosures are useful" (Google: creating helpful content (external link)).
Typical AI draft problems: before and after
The examples below are our own, written for a fictional heating installer. They show the pattern, not real data.
Invented number or source
- Before: "According to a recent study, 73% of homeowners replace their boiler too late."
- After: "Many of the boilers we replace have already broken down in the middle of winter. Have an older boiler checked in summer, when there is time to plan." (A statement from your own work instead of a study nobody can find.)
Outdated fact
- Before: "This year, the state covers up to half of the cost of a new heat pump."
- After: "Funding rules change often. Check the current conditions on the funding body's official website before you sign a contract; we help you with the application."
Generic phrase
- Before: "In today's fast-paced world, a reliable heating system is more important than ever."
- After: "If your heating fails on a cold weekend, you want to know who to call. Here is what to check first."
Repetition
- Before: introduction, body and conclusion each say "regular maintenance saves money".
- After: say it once, in the section on maintenance, with the reason: "A yearly service lets us replace worn parts before they fail."
Missing own example
- Before: "Many customers have questions about maintenance."
- After: "The question we hear most often: 'Does the service also cover the hot water tank?' The short answer: ..."
How long does a review take?
There is no fixed figure, and we have not measured one. As a rule of thumb, plan around 20 to 45 minutes for a standard post on a topic you know well, and one to two hours when the post contains many facts, prices or rules, or when it touches health, money or law. The first few reviews take longer; once you have your own list of words to avoid and your usual sources, it goes faster. If a draft needs more time than writing it yourself would, improve the brief or the brand voice settings rather than the single post.
A review note you can copy
Keep a short note with every post. It makes the review repeatable and shows who checked what.
Post: [title]
Reviewed by: [name], [date]
Facts checked: [yes / removed: ...]
Sources opened and linked: [list]
Own example added: [what]
Title and meta description: [checked]
Internal links: [pages]
Alt texts and structured data: [checked]
Open points: [none / ...]
How Blog Automation supports the review
In Blog Automation, AI posts go to review first by default (the setting "always send AI posts to review first" is marked as recommended). With this setting, nothing is scheduled or published before someone approves it, and a reviewer can send a post back to the author. The editor offers SEO suggestions for title, description and slug, an alt text field for the featured image, versions and a preview, and the media library can filter images without alt text. Your brand voice settings include tone, form of address, words to avoid and a sample text in your style. In the Pro plan, roles decide who may approve. Read more about the approval workflow and the AI blog writer, or see how review fits into an automated schedule in how to automate blog posts.
Frequently asked questions
How much editing does an AI draft really need?
It depends less on the tool than on the topic. A draft about something you explain to customers every day may only need facts checked and one own example added. A draft full of numbers, rules or prices needs every one of them verified. Use the 15 points above as the minimum and stop when each one is done.
Should I run AI text through a "humanizer" tool?
That does not fix what matters. A rewording tool does not check facts, add your experience or correct a wrong title. Google judges the quality of content, not how it was produced; see Does Google penalize AI content? for the details. Spend the time on points 1 to 5 instead.
Can AI review its own draft?
AI can help: ask it to list every factual claim in the draft or to find repeated sentences. But the actual check has to be done by a person, because the model cannot confirm its own facts. Google's guidance asks you to "manually factcheck and review" AI content (Google: using generative AI content (external link)).
Who should approve a post before it goes live?
Someone who knows the subject and is allowed to speak for the business, ideally a different person from the one who prepared the draft. For a small business that is often the owner; in a team, define who has the right to approve so that no post goes live unchecked.