How to write SEO articles with AI that rank
Updated 3rd November 2026. First published in 2024.
Two years ago, I ran an experiment.
I had this article written entirely by AI, scored and packed with keywords in SurferSEO, and published it with barely any human input. It ranked well at first. Then it sank, and it has sat on page three ever since.
Reading it back, it's easy to see why.
Almost every section ended with the same filler line about "the SEO writing process".
It got basic facts wrong, called natural language processing "natural language programming" and misspelt Flesch. It said nothing that hundreds of other articles didn't already say, and it had no experience behind it at all.
So I've rewritten it the way we actually write SEO articles with AI now. AI does a lot of the work. A person still decides what's worth saying.
What Google actually says about AI content
Google doesn't penalise content because AI helped write it. Its guidance, published in February 2023, says that "appropriate use of AI or automation is not against our guidelines".
What it does penalise is using automation to generate content "with the primary purpose of manipulating ranking in search results".
Since March 2024, Google's scaled content abuse policy has targeted sites that publish large volumes of pages mainly to rank, however they're made. I cover how that plays out in how to avoid a Google penalty. The question Google asks isn't whether AI wrote it. It's whether the page helps anyone, and whether it shows the experience Google looks for under E-E-A-T.
What the research shows
The data backs this up from both sides. Ahrefs analysed 600,000 pages ranking in Google's top 20 and found that 86.5% contain at least some AI-generated content, with almost no link between how much AI a page used and where it ranked. Only 4.6% were entirely AI-written.
Pure AI content struggles, though. Graphite's analysis of 65,000 articles found that AI now writes roughly half of all new articles published online, yet only 14% of articles ranking in Google Search and 18% of those cited by ChatGPT and Perplexity are AI-generated.
AI writes half the web, yet rarely ranks
Share of articles that are AI-generated
Source: Graphite analysis of 65,000 articles, reported by Axios, October 2025. New articles figure as of May 2025.
In short, AI helps most top-ranking content get written, while content that AI writes alone rarely ranks. That matches what happened to my experiment, and what we've seen with clients and competitors who published AI-only content with no experience or human input behind it.
How to write an SEO article with AI, step by step
This is the process we follow for a single article. Our article on how we use AI in content production covers how it fits a wider content team.
1. Start with the search, not the keyword
Look at what people actually search for and what Google already shows, including any AI Overview at the top. We use Ahrefs to find the main term and the related questions, then target several closely related searches with one strong page, not one page per keyword. Ask what the person searching really wants to know, and what nobody on page one has told them yet.
2. Bring something AI can't
Before any drafting, decide what makes your article different. That's usually first-hand experience, such as a client result, a mistake you learned from, your own data or an expert's view. AI can't invent any of it, and it's exactly what Google and AI search tools look for.
For some clients, we build articles around interviews with their specialists. It varies by client, and it's especially useful for clients who publish in several languages. A typical run looks like this.
Claude drafts interview questions on the topic in seconds.
A specialist on the client's side reviews them in about five minutes.
We interview the expert, usually for about 30 minutes.
A freelancer transcribes it with AI support, for around $30 to $45 and within about 48 hours.
Claude turns the transcript into a first draft in 15 to 30 minutes.
The same draft produces around ten pieces for LinkedIn, newsletters and other channels.
For multilingual clients, translation takes about 48 hours and costs around $170 per article.
The interview is the part that matters. Everything else speeds up turning real expertise into publishable content.
3. Write a brief and a proper prompt
Most poor AI content comes from vague prompts. Ours include the target reader and search, the angle, our house style rules, approved sources, the author's voice and any interview notes. A shortened version looks like this.
"Write the section 'How to choose a platform' for a blog post aimed at first-time founders. Use British English, active voice, short sentences and no jargon. Base it on the interview notes below and the sources provided, and don't add facts that aren't in them. Write in the first person as David, who has 12 years of SEO experience. End with one practical recommendation."
One habit from my experiment still works. Prompt section by section, not for the whole article at once. AI holds the brief far better over 300 words than over 3,000.
4. Rewrite it in your own voice
An AI draft is a starting point. I rewrite every one in my own tone, cut the phrases that give AI away, and add the opinions, examples and asides that make it sound like a person. Content scoring tools such as Surfer can flag missing subtopics, and we still use them for that. Don't chase a perfect score, as that's how my original article ended up stuffed with repeated phrases.
5. Check every fact and source
AI still invents statistics, misquotes studies and cites sources that don't say what it claims. Check every figure against the original research, link to it, and use recent sources. If you can't find where a statistic came from, leave it out.
6. Optimise the page
Write the title for what people search, not for a keyword formula, and give it a meta description that earns the click. Use one clear H1, logical subheadings and short answers under question-style headings that AI tools can quote. Add tables where you compare things, internal links to related pages and a named author. Our guide to the anatomy of a webpage covers each element, and our guide to Flesch reading ease scores covers readability.
7. Publish under a real name and keep it updated
Publish under the name of the person whose experience the article draws on, with a short bio. Then update it as things change. My experiment had neither, and it shows in the results.
What AI does well, and what needs a person
| Task | AI does well | A person must |
|---|---|---|
| Research | Summarise sources and suggest related questions | Check sources are real, recent and say what AI claims |
| Angle | List common angles competitors use | Choose the angle only you can write |
| Interviews | Draft questions and turn transcripts into drafts | Run the interview and supply the expertise |
| Drafting | Produce a structured first draft, section by section | Rewrite it in a real voice with real examples |
| Optimisation | Suggest titles, meta descriptions and missing subtopics | Decide what matches search intent and the brand |
| Publishing | Create social posts and other spin-offs | Put their name to it and take responsibility |
How much time does AI really save?
Less than most people expect. In our experience, AI saves around 25% of the time it takes to produce a good article. Writing the prompts, rewriting the draft, checking facts and adding experience still take most of the effort. If AI is saving you 90%, you're probably publishing the kind of content that sank my experiment.
Where AI saves far more time is everything around the article, from interview questions and transcripts to social posts and newsletters. Our article on content marketing for field service SaaS shows what that looks like across a long content programme.
Should you tell readers you used AI?
I don't think it matters, as long as the content is genuinely helpful and a real person stands behind it. Readers care whether an article answers their question, not which tools helped write it.
The law points the same way for most business content. From 2nd August 2026, the EU AI Act requires anyone publishing AI-generated text to inform the public on matters of public interest to disclose it, unless it has been through human review and a person or organisation holds editorial responsibility. Proper editing keeps you on the right side of it, and that's good practice anyway.
Get help writing SEO content
Our content writers combine AI with interviews, real expertise and proper editing, and our SEO team makes sure every article targets searches that matter.
We also handle web design, PPC and social media, so your content reaches the right people once it goes live. For clients in several markets, our guide to multilingual SEO explains how we approach other languages.
-
Yes, ChatGPT and Claude can write a complete SEO article in minutes, with headings, keywords and a meta description. Without human input, that article is unlikely to rank for long. AI can only repeat what's already online, so it can't add the first-hand experience, original examples or expert views that Google and AI search tools reward. A person needs to shape the angle, rewrite the draft in a real voice, proofread it and check every fact before it's ready to publish.
-
No. Google says appropriate use of AI is not against its guidelines, and Ahrefs found that 86.5% of top-ranking pages contain some AI-generated content. What Google penalises is content created mainly to manipulate rankings, such as sites that churn out large volumes of thin pages with little or no human review, whatever tools they use. If your article helps the reader and shows real experience, using AI to write it won't count against you.
-
In our experience, AI saves around 25% of the time it takes to produce a good SEO article. That's less than most people expect, because the slowest parts still need a person. Writing detailed prompts, rewriting the draft in your own voice, adding experience, proofreading and checking every statistic take most of the effort. AI saves far more time on the work around the article, such as interview questions, transcripts, social posts and newsletters.
-
For most business content, no. The EU AI Act, which applies from 2nd August 2026, requires disclosure for AI-generated text that informs the public on matters of public interest, unless a person has reviewed it and holds editorial responsibility for it. Proper editing covers you. Beyond the law, readers care whether an article answers their question, not which tools helped write it, as long as a real person stands behind it.
