AI website SEO · I rebuilt my own site with AI to find out

Do AI-Built Websites Actually Rank on Google?

The experiment: 25 pages, one rule set, three weeks

On 7 and 8 October 2026 I published 25 rebuilt pages on fusiongrowth.com.au, every one drafted with AI assistance and every one passed through the same automated gate before it went live. Service pages, city pages and one long guide. The old versions had been written the lazy way in 2025, with the keyword repeated dozens of times and suburb lists pasted at the bottom, and the site had been sliding since September.

Search Console, fusiongrowth.com.au, weeks ending 28 September to 25 October 2026

Week ending	Search impressions	Distinct queries
28 September, before the rebuild	1,032	158
5 October	1,698
12 October	3,208
19 October	3,093	454

Pulled through the Search Console API into my own reporting. Impressions tripled and the number of different searches the site appeared for nearly tripled inside three weeks.

Which pages moved, and by how much

The AI-assisted pages Google had not been showing at all started appearing almost immediately once they were indexed. Comparing the ten days before and after the indexing requests:

Rebuilt page	Impressions, 5 to 14 October	Impressions, 15 to 24 October
Facebook ads management	24	304
Digital marketing Toowoomba	2	303
Web design Ipswich	1	284
Website design	2	256
Digital marketing Bundaberg	0	161
PPC advertising Bundaberg	0	130

The honest half of the result

Impressions are not rankings. Most of those new appearances were deep, at average positions in the sixties and seventies, because new pages enter low and Google was still deciding which of several rebuilt pages should own each search. For my main term it tested eight different pages across three weeks before settling. Clicks were roughly one a week throughout. Indexed pages went from 38 to 42. That is the real shape of an AI-assisted rebuild: discovery first, re-evaluation second, rankings last, and the rankings take months.

One page Google crawled and refused

Of the 25, one city page was crawled on 27 October, fetched successfully, allowed to index, and still left out. It is not a duplicate; it shares under a third of its text with its siblings. Google simply has not valued it yet. Same tool, same rules, same week as the pages above. The method does not guarantee anything; it just removes the reasons for a no.

The pages Google refused, on the same site

The same site carried six blog posts from 2025 and early 2026 that were drafted with AI and published without a second pass. Between 724 and 1,075 words each, correct headings, FAQ schema, an author box, and not a single number from a real job. One of them used its target phrase 25 times in 1,082 words. Google crawled every one of them and declined to index a single one.

That is the clearest answer to the question in the title that I can give you. Google did not care that AI wrote them. It cared that they said nothing. The pages that ranked and the pages that were refused came from the same tool on the same domain in the same year. I have since set those six to noindex and I am rebuilding them one at a time with my own data, starting with a real comparison of SEO and Google Ads built on one client’s numbers.

The AI-built site that cannot rank at all

In October I looked at the website of an electrical and security installer who had built it himself with one of the AI app studios. It looked fine in a browser: dark theme, services, reviews widget, contact form, a map. Underneath, it was a web application rather than a website.

What the page actually contained, October 2026

Page body in the HTML	Empty. Everything is drawn by a script after the page loads.
Title tag	Present, the only piece of text in the HTML
Meta description, schema	None
robots.txt, sitemap	Both return 404
Analytics, conversion tracking	None
Images	About 25, hosted on a free image-sharing service
Contact form	Posts to a third-party form service
Styling	Loaded from a public CDN at runtime

Google can render JavaScript, but a page with an empty body and no sitemap is asking it to do all the work and giving it no reason to. The owner cannot see whether the site produces a lead because nothing is tracked, and the images would vanish if the sharing service changed its rules. None of that is an AI problem in the sense people mean. It is a platform problem: the tool he used cannot produce the basics. The fix is a rebuild on a platform that can, which is the same website design process I run for any trade.

The rules my pages have to pass

Every AI-assisted page I publish runs through an automated check before it goes live, and it fails the page on any one of these. These are the exact rules, because the gate is the difference between the 25 pages that moved and the six that were refused.

Live from my own rank tracking

Every site in this tracking was built or rebuilt with AI assistance under the rules below. These are the current numbers.

What AI-assisted builds have done for clients

Every client site I have built or rebuilt since mid 2026 was drafted with AI and shipped through the same gate. Two results from my own reporting, with the businesses unnamed:

• A Hunter Valley painter, 33 pages rebuilt, 45% more quote requests in four months.
• A Gold Coast electrician, site and profile work, 2.4 times the website leads he had before and the top 3 in Google Maps for his main term.

The painter’s story also carries the warning. In October his contact form, the only one on the site, was rejecting real enquiries as spam for eleven days before a customer sent a screenshot. A rebuilt site still needs someone watching it, which is most of what I describe in what you actually get when you pay for SEO.

Seven checks for your own AI-built site

1. Search site:yourdomain.com.au. If Google returns nothing, or only the home page, start here.
2. View the page source, not the page. Right-click, view source, search for a sentence from your services section. If it is not in the HTML, Google is being asked to render it, and may not.
3. Open Search Console and read the Pages report. “Crawled, currently not indexed” is Google telling you the page said nothing. “Discovered, not indexed” usually means a dead sitemap or no internal links.
4. Check for noindex and canonical tags left over from a staging setup. A builder can ship a site that asks Google to ignore it.
5. Fetch robots.txt and the sitemap. Both should return a page, not a 404.
6. Read one service page as a customer. Does it name the service, the area, a price or a timeframe, a licence, and a real job? If it could be any business, Google will treat it as none.
7. Count the internal links. Every service page should link to the related guide and the contact page in plain sentences, and be linked from somewhere other than the menu.

If the words “AI Overview” are why you are asking, the same rules carry over, and I wrote up three tests of my own in how to rank in Google AI Overviews. The checklist I use for the whole site is on AI Search Readiness.

Verdict

AI-built websites rank when the build produces real HTML and a person puts real substance in; they do not when the tool produces an empty shell or the content could be anyone’s. I have both outcomes on my own domain inside the same month. The tool was never the variable. The gate and the data were.

Frequently asked questions

Does Google penalise websites built with AI?

No. Google’s published guidance judges the page, not the tool, and my own site is the proof in both directions. Twenty-five AI-assisted pages built under strict rules were indexed and climbing within weeks; six AI-era posts that said nothing specific were crawled and refused. Same site, same tool, opposite outcomes. The difference was the content, not the method.

Can a website made with an AI site builder rank for local searches?

Only if the builder produces real HTML Google can read, and most of the pure AI app builders do not. The installer’s site above had an empty page body and a title tag; everything else was drawn by a script after load. Google indexes what is in the HTML. Check your own site by viewing the page source: if the words are not there, neither is your ranking.

How do I know if my AI-built site is in Google at all?

Search Google for site:yourdomain.com.au and count the results, then open Search Console and read the Pages report. The report lags the index by a few days, so confirm any page you care about with URL Inspection. Mine showed 42 pages indexed and 19 crawled but refused, and it named every one of the refused pages.

Should I rebuild my AI website or fix it?

Fix it if the HTML is sound and the content is thin; rebuild it if the platform cannot serve real HTML, has no way to set titles, descriptions, schema, a sitemap or analytics, or hosts your images somewhere you do not control. The painter’s site was a rebuild and produced 45% more quote requests in four months. The posts on my own site were fixes, one at a time, with real data.

Is AI-written content allowed on a business website?

Yes, when a person who did the work puts the specifics in and takes responsibility for the result. Every page on my site is drafted with AI assistance and every number on it comes from my own reporting. The pages Google refused were the ones where nobody had done that second step.