SEO Content Writing with AI

SEO content writing with AI works when the AI does the wording and you supply what only you know: check what Google shows at the top for your term, give the AI your own material (experience, numbers, cases, sources), have it write to a fixed outline, check every claim against its source, get a second review and approve it yourself. Google has no problem with AI as a tool: quality counts, not how content is made, and AI gives a page no special gains. What Google lists as a spam pattern is using AI to churn out many pages that add nothing for users. Generic content happens where a text only contains what the AI already knew.

On kevinwelter.com, Sebastian, my AI employee for SEO and GEO, writes the articles, in separate roles and with checks; I proofread, answer his questions and approve. Nothing goes live without me. That's how the site grew from 10 to 224 posts per language between August 27 and September 21, 2026. The basics of writing with AI are in Writing with AI: A Beginner's Guide. All my SEO articles are collected on AI SEO for Google and AI Search.

What Google says about AI content

Three pages of Google's Search Central documentation cover it, as of November 2026:

Google's 2023 FAQ puts it plainly: using AI gives content no special gains. "It's just content." The real question: does the page offer something readers can't find elsewhere, and has someone checked that it's true?

SEO content writing with AI: the workflow in six steps

1. Check the search intent

Search your term and read the top ten: guides, comparisons, prices or providers? That's the intent to meet. An example: in early November 2026, the US top ten for "seo content writing" mixed a free writing tool, beginner guides, Q&A threads, writing services and a job board. Hand those results to the AI instead of letting it guess.

Autocomplete shows how people phrase the question. Turning customer questions into pages is covered in SEO Basics for Small Businesses. Google Search Console shows which terms your pages already appear for, but not search volume. For volume, Sebastian used SE Ranking* in my test and got a figure for 411 of 1,000 terms (AI Employee With and Without an SEO Tool). Such a number stays an estimate until Search Console or the live page confirms it. Upfront: I have a partnership with SE Ranking and earn a commission through this link.

* Affiliate link: if you subscribe or buy credits through this link, I earn a commission. Your price stays the same. I have a partnership with SE Ranking.

2. Gather material only you have

This decides whether you get generic content. Google asks for first-hand expertise, and an AI has none. Create a fact sheet, one file with everything allowed into the text: real customer questions, your own numbers, processes and cases, prices with source and date, a link for every outside claim, and what you don't want to claim.

An example: a tiler writes about bathroom renovation costs. Without a fact sheet, the AI delivers averages found everywhere. With one, the page shows his common bathroom sizes, what drives his prices and the question every first call brings.

3. Fix the outline first

Sebastian writes to a fixed guide: search phrase from real Google suggestions, the answer in the first paragraph, a question block and sources. Subheadings follow the questions from step 1; one need, one page. Check whether a page on the topic already exists, or two half pages will compete. How a pillar page and its cluster fit together is in Pillar Pages and Topic Clusters.

4. Have the draft written with rules

A brief to copy:

Write a blog post for the search term "<term>".
Search intent: the top results are <e.g. guides and providers>. Cover both.
Reader: <who, with what prior knowledge>.
Structure: the first paragraph fully answers the question. Then sections
on: <questions from your research>. End with four frequently asked questions.
Own experience: include these points visibly: <case, number, process>.
Rules: only facts from the fact sheet. If something is missing, write
[MISSING: …]. No intro about why the topic matters. Use the search term
only where it fits naturally.
Fact sheet: <paste>

The [MISSING: …] line matters most: without it, the AI fills gaps with something plausible. No word count, on purpose.

5. Check every claim against its source

Check numbers, names, dates, quotes and links one by one, and open every link, including in titles, descriptions, structured data and image text. Why language models make things up is explained in AI Hallucinations: Why AI Makes Things Up. How to validate structured data is in Schema Markup Checker: Which Tool Tests What.

6. Get a second review, then approve

An AI checking its own draft in the same chat is no second check; as Sebastian's page says, two checks from the same source are only one check. Use a new chat with the fact sheet and this brief:

You review, you don't rewrite. List every number, name, date and claim
not supported by the fact sheet, including in the title, description and
image text. Flag sentences that could appear in any other text on the topic.

I saw why on September 18, 2026: an independent cross-check found factual errors in a delivery that Sebastian's own editing had missed. Since then, large deliveries get one. The last role is still yours: I proofread all 76 posts of the second stage and answered 42 questions.

Then measure in Search Console. After the build-out of September 7, 2026, Search Console showed me twice as many clicks within a week. That's a measurement from my project.

How to spot generic content

More AI tells are in Can AI Write Emails in My Voice?.

What doesn't work: mass content without value

Google's spam policy names many AI pages without value and rewritten or stitched-together content from elsewhere as the main patterns; a new date without substantially new content is a warning sign too. Where the line runs for pages generated at scale is in Programmatic SEO with AI Agents.

Many pages alone aren't spam under Google's definition; what matters is whether they mainly aim to manipulate rankings and offer users little. kevinwelter.com also grew in stages with many posts at once, 76 per language in the second. So topics come from proven demand, and every post goes through review and approval. Whether that's enough is Google's call, not mine.

Getting cited: what AI answers add

For AI Overviews and AI Mode, Google says there are no additional requirements. The page must be indexed and eligible for a snippet; then the SEO fundamentals apply, such as content in text form, good internal linking and structured data that matches the visible text. No new AI text files needed, says Google (page as of December 10, 2025). What llms.txt does and doesn't do is in llms.txt: Hype vs. Evidence.

A working rule for readers, not a proven ranking factor: write each section's key point so it stands on its own, with number, source and date. Google says explicitly that you don't need to chop content into tiny pieces for AI. What ChatGPT and Perplexity need technically is in Optimizing Your Website for AI Search, the background in What Is Generative Engine Optimization?. Sebastian measures Google and AI search separately: in the three months through September 16, 2026, my most-cited page in Microsoft Copilot was not the one with the most Google clicks. How to measure it yourself is in AI Visibility Tracking: How I Measure It.

Write it yourself or hire a writer?

If you hire SEO content writing services, ask: where does the material come from? Is AI used, and who checks facts? Where are the sources? Does anyone measure whether the page gets found? When an agency is worth it is covered in Hire an SEO Agency or Do It Yourself?. On my team, Ralf, my AI employee for websites, adds new blog sections and SEO pages to existing websites; Sebastian writes the articles and measures.

Frequently asked questions

Does Google detect and penalize AI content?

Google says it focuses on quality, not on how content is produced. Content made mainly to manipulate rankings breaks the spam policies, whoever made it. On how unreliable AI detection is, see Can You Detect AI Text? What Detectors Do.

How long should SEO content be?

As long as the question needs. Google says it has no preferred word count.

Do I have to disclose that AI wrote a text?

Google recommends a disclosure where readers would reasonably ask how the content was created, and advises against listing AI as the author. The legal side is in AI Disclosure Rules Under the EU AI Act.

Can I update old content with AI?

Yes, if the update adds something: new facts, a better answer, current sources. Changing the date without substantially changing the content is a warning sign, according to Google.

How to continue

Pick one page that should bring in enquiries and do only steps 1 and 2 today: read the top ten, create the fact sheet. How to connect the new page with your other pages is in Internal Linking: Finding the Links You Miss. Sebastian's template with tool chain, writing guide and project folder structure is in my community AI Employee Framework. And if you want to know whether an AI employee like Sebastian fits your business, get in touch.