- AI website SEO — what it means
- AI website SEO refers to how search engines evaluate websites built with AI. Google officially states that it judges content on quality, not on whether it was AI-generated, so the production method itself is not penalized.
Hi, I’m Masato. I built my career as a Webflow specialist and am now expanding into AI website production.
✓ In this article
- Testing the conventional wisdom that “AI-built sites are weak at SEO” against measured data
- Index status the day after our rebuild, and our actual ranking for “webflow”
- What went into the AIO implementation across 63 articles — something only AI made possible
In July 2026, I fully rebuilt our company site with AI (Claude Code) over seven days: 26 pages built and 62 articles migrated, with me as the only human. What I feared most was precisely that conventional wisdom — “AI-built sites are weak at SEO.” Destroying a blog with 190,000 impressions a year would have been no joke.
Here’s the conclusion up front. It wasn’t weak. What makes a site weak isn’t AI — it’s how you build it. In this article I lay out three pieces of evidence, using measured Search Console data.
Why people say AI-built websites are weak at SEO
First, the premise. In its official guidance, Google states that it evaluates content on quality, not on whether it was AI-generated. In other words, the search engine does not penalize the fact that AI built a site.
The reason the belief persists, I think, is that genuinely weak AI sites are out there. The usual patterns:
- Mass-producing articles with AI and no first-hand experience, so the content is the same as everywhere else
- Neglecting technical groundwork like sitemaps and structured data
- Not designing how URLs and content carry over during a rebuild
Timing is another big factor. In my experience, AI alone became capable of building a site of practical quality only around May–June 2026 — the last month or two. I looked at services that were advertising “cheap because it’s AI-built” before that, and my honest impression is that many of them looked like assembled templates. “AI” was the sales pitch; the reality was closer to template distribution. Those spreading as “AI sites” first is, I think, what strengthened the belief.
So if you commissioned an AI site purely because it was cheap and feel it isn’t delivering, take a hard look at what you actually got. Frankly, rebuilding it yourself with ChatGPT or a similar tool could still produce something more valuable. That’s how far the AI available today has come.
None of this is about “AI” — it’s about how and when the site was built. So what happens when you build it right? Below is what we measured on our own site.
Test 1: 24 of 30 legacy articles kept their index the day after the rebuild
The scariest part of any rebuild is having years of accumulated article rankings reset. This time we swapped the entire build system, from Webflow to Astro.
The result: of the 30 legacy articles I checked in Search Console the day after the rebuild, 24 had kept their index. For the day after replacing the whole system, that was frankly a relief.
The reason is simple: AI handled the build only, while a human made the decisions on SEO carry-over. We kept the article URLs, and included GA4, Search Console, structured data, the sitemap, and llms.txt in the rebuild process itself. The full process is documented in the AI rebuild log.
Test 2: the main keyword “webflow” held at position 3.9
Our blog’s flagship keyword is “webflow.” Search Console’s 92-day data shows an average position of 3.9, 681 clicks, and 15,084 impressions. We held that position for a single keyword even through a full AI rebuild.
There’s another interesting data point: one article ranks 1.7 with 162 clicks for the keyword “address 最悪” (roughly “ADDress is terrible”) — an honest review of a housing subscription service I actually used. The article that hides none of the downsides owns the strongest keyword. Before AI even enters the picture, what wins in search is first-hand experience and honesty.
AI can’t create that first-hand experience. But it is extremely good at shaping first-hand experience into a form that reaches search. That’s the third test.
Test 3: AIO implemented across all 63 articles — only possible with AI
In July 2026, we implemented AIO (AI search optimization) across all 63 blog articles at once. Specifically, four things:
- A key-takeaways box at the top of each article (conclusion first)
- An FAQ section plus FAQPage structured data
- A one-sentence definition of the keyword (“X is …”)
- Speakable structured data
Scored against our in-house criteria before implementation, the 63 articles averaged 47.8 points. After: 97.3. These are our own scores, so they don’t guarantee rankings — but the change is that a structure AI search can easily cite went from zero articles to every article.
Along the way we corrected over 450 typos and merged three pairs of articles that had been competing for the same keywords, using 301 redirects. Doing that volume by hand is, frankly, unrealistic. Implementing it across 63 articles at once was possible precisely because the site was built with AI.
So what about results? Traffic via AI search currently runs about 12 sessions a month, almost all via ChatGPT. A small number, yes. But I think “still small” is exactly the point: rather than scrambling once that traffic takes off, getting the structure right now is what makes you an early mover. The full implementation details and steps are in AIO Field Report #1 (implementation), and I’ll follow the results as a series.
Test 4: an SEO audit went from a full day’s work to 10 minutes
So far this has been about “it didn’t get weaker.” But having actually operated the site, my view is that building with AI is becoming an SEO advantage. The reason is speed.
The items in an SEO audit vary by company and site, but a thorough one can run to around 100. AI checks all of them in a single instruction and comes back with a score. It takes 10 or 15 minutes. When humans did it, that work took a day or two — a week for some content.
The same applies to troubleshooting after launch. In this rebuild, I can’t recall any particular Search Console warning or error. Not because none appeared, but because the moment one appeared I handed it straight to AI to fix, so none stuck. The old me would have burned time just tracking down the affected spot.
To be honest, I have so far found nothing that makes an AI-built site — with a human holding the design and the decisions — lose to a site built by humans alone. This is of course limited to what I’ve seen, and I can’t prove it applies to every site. But this gap in audit speed compounds over time.
A true story: the Korean pages I almost threw away, rescued by the numbers
This is the part of the article I most wanted to write.
The old site had a Korean version, built with Webflow’s automatic translation. I honestly expected nothing from it, so during the rebuild I decided it wasn’t needed and redirected every Korean URL to the Japanese version. I was going to wind it down.
Then, looking at the Search Console numbers afterwards, I noticed something. The Korean pages I had supposedly retired were still getting clicks. Not many, but not zero. Seeing that, I reconsidered — “throwing this away would be a waste” — and this time rebuilt the Korean version with AI and reopened the URLs.
Incidentally, of the pages that dropped out of the index while the redirects were in place, a few never came back. I’m not chasing them: those articles are still Webflow’s automatic translation, and I never expected them to be rated well in the first place. What matters more is that Korean can grow now that AI handles everything from translation to writing.
What I learned is that SEO decisions made on gut feeling are usually wrong. A page I assumed “nobody looks at anyway” was, by the numbers, being looked at. The decision to discard something, above all, should wait until you’ve checked the numbers.
When an AI site does get weak
Flip the three tests around and the conditions that make an AI site weak at SEO become clear.
- No carry-over design: change the URLs without setting up redirects or a sitemap, and your rankings simply vanish
- Mass production with zero first-hand experience: AI-written “plausible” articles can’t beat one honest review
- No final human check: AI-written text contains unexpected wording and figures. Fact-checking is a human job
Put the other way: if humans hold the decisions and the design while AI handles volume and structuring, AI isn’t SEO’s enemy — it’s its strongest ally.
Summary: what’s weak isn’t AI, it’s the build
My answer to “Are AI-built sites weak at SEO?”:
- 24 of 30 legacy articles kept their index the day after the rebuild
- The flagship keyword “webflow” held at position 3.9 (681 clicks in 92 days)
- AIO implementation across 63 articles was, if anything, only possible with AI
What’s weak isn’t AI — it’s the build. That’s the measured conclusion from a 63-article site. Technical health is verified separately in our Ahrefs free-tier audit log (score 6 → 98).
This test continues as the series “A public experiment in growing a blog with AI.” Detailed AIO reports and the latest measurements go out first through the newsletter (Japanese). If you want your own site built or rebuilt with AI, we offer AI Website Production built on the same design philosophy as this article.
FAQ
Do AI-built websites get penalized by Google?
Not for being AI-built. Google’s official guidance states that it evaluates content on quality, not on whether it was AI-generated. The author’s site held position 3.9 for its main keyword “webflow” after a full AI rebuild.
Will rebuilding a site with AI hurt search rankings?
It depends on how URLs and content are carried over. The author kept the URLs and set up the sitemap, structured data, and redirects; the day after the rebuild, 24 of 30 legacy articles kept their index and the main keyword’s ranking did not drop.
What does AIO (AI search optimization) actually involve?
Structuring content so AI search can cite it easily: a key-takeaways box at the top of each article, FAQs with FAQPage structured data, a one-sentence definition of the keyword, and Speakable structured data. The author implemented these across 63 articles, raising in-house scores from an average of 47.8 to 97.3.
What do AI sites that are weak at SEO have in common?
Mass-producing articles with AI and no first-hand experience, neglecting technical groundwork such as sitemaps and structured data, not designing URL carry-over during a rebuild, and skipping a final human check. All are problems of how the site is built, not of AI.
