AI Built Websites: The Truth, The Bad and the Data

AI can now draft a full website in an afternoon, and for a small business owner that is good news. The next question matters more: does it bring in more customers?

We went looking for the data on traffic, leads, bookings, maintenance time, cost and time saved. This post sets out what we found, what we could not find, and what it means for getting your site found on Google and in AI answers.

If you want the how-to side first, start with our guides on whether AI-built websites can perform in Google and how an AI-built site gets found on Google and ChatGPT.

The short answer

Not by itself. An AI-built website cuts what you pay to get online and speeds up the first draft. But no independent study shows that an AI-built site brings in more customers on its own, and the conversion claims that say it does come from companies that sell AI tools or SEO services.

Customers come from being found, and that depends on original content, a sound technical foundation and local signals. In our view, that is the layer where a specialist adds the most.

• More customers: unproven for AI-built sites on their own. The only conversion figures (18–35% higher) have no published method. What the data does support is that human-written pages hold the top spot in Google far more often than purely AI-generated ones.
• Time saved: real for writing (40% faster in a controlled study), small overall (1.4–5.4% of work hours), and unmeasured for maintaining a site.
• Money saved: clear on the build (about $360–$370 a year on a mid-tier builder plan). That covers the platform, not page copy or SEO setup.

What the data actually covers

We could not find a single independent study that tracked small businesses before and after rebuilding their website with AI. What exists falls into three groups, and only one of them is independent research.

Source	Who is behind it	What it measures	Independent?
Ritner Digital report	An SEO agency that sells optimization	Claims 18–35% higher conversion for AI-assisted sites, with no sample size or method given	No
NxCode case studies	An AI app-building company	Five case studies (dental practice, skincare store, SaaS and others), none naming the business	No
Noy & Zhang, Science 2023	MIT researchers	Time and quality of writing tasks, 453 professionals, randomized	Yes
METR, 2025	A non-profit research group	Task time for 16 experienced developers, randomized	Yes
Humlum & Vestergaard, Denmark	Academic economists	Time saved and earnings for 25,000 workers	Yes
St. Louis Fed	A Federal Reserve bank	Hours saved by US workers using generative AI	Yes
Orbit Media	A web agency analyzing real analytics	How visitors referred by AI assistants convert, 97 B2B sites	Partly: real data, but one agency’s sample

The first two rows are marketing. The case studies are unverifiable, and one of them claims a 1,247x return on a $204 dental website by assuming every inquiry becomes a paying patient.

The independent studies measures time saved doing tasks like writing and coding. They say nothing directly about whether a website wins more customers, so we keep those two kinds of evidence apart below.

One note on our own interest: TopSEOPh sells SEO-ready web building.

Weigh our conclusions the way we ask you to weigh the vendors above, and check the numbers in the linked sources yourself.

Traffic, leads, conversions and bookings

No independent data shows that an AI-built website brings more traffic, leads or bookings by itself. The numbers that do exist are vendor claims, and the independent numbers describe how search itself is changing.

What the vendors claim (none verifiable)

• Ritner Digital reports 18–35% higher conversion for AI-assisted sites, saying “platforms report” it and crediting a source called Dataintelo, with no sample size or method.
• NxCode’s dental-practice case says monthly inquiries rose from 8–12 to 35–45 and online booking reached 68% of appointments. The practice is not named.
• A SaaS case shows conversion rising from 8.2% to 18.7%. The page itself credits the lift to testing 12 page variations instead of 2, not to AI design.
• The strongest controlled number in these pages is about page speed: Rakuten 24 saw a 33.13% higher conversion rate with good load times. It says nothing about AI.

What independent data says about traffic

• Pew Research tracked 68,879 Google searches by 900 US adults in April 2025. People clicked a result on about 8% of visits when an AI summary appeared, versus 15% when it did not.
• Orbit Media analyzed 97 B2B sites and 29 million visits (July 2025 to June 2026). Only about 0.5% of visits came from AI assistants, but those visitors were about 3x more likely to become leads. ChatGPT converted about 2.1% versus 0.5% for Google Search, though Google sent about 100x more traffic.
• Orbit flags that Google’s AI Mode and AI Overviews are counted as organic search, so AI’s true share is likely higher than 0.5%.

AI traffic is small today but converts well, which is why being recommended by AI assistants is worth planning for.

We explains how that works in why your business doesn’t appear in ChatGPT (and how to fix it), and we looked at the wider shift in our Semrush data analysis of AI and search.

What it says about ranking AI-written content

• Semrush (April 2026; 20,000 keywords, 42,000 blog pages) found human-written pages in the top spot about 80% of the time and AI-generated pages about 9–10%. The detector cannot tell edited AI drafts from raw ones, the result is correlation, and Semrush sells AI content tools.
• Graphite found 86% of articles ranking in Google were human-written and 14% AI-generated, and that AI had reached about 48% of new articles by May 2025.
• Google allows AI-assisted content, but says using AI to mass-produce pages that add no value can violate its spam policy. It also tells publishers to fact-check AI output, including titles, meta descriptions, structured data and alt text.

The pattern is consistent: AI is a good way to draft, and the pages that perform are the ones where a person added real specifics and checked the details. That is the thinking behind our data-first content shop.

Speed depends on how the site is built and tested

In June 2026 field data (HTTP Archive, mobile), 80.1% of Wix sites passed all three Core Web Vitals, versus 73.1% on WordPress.com and 70.1% on Squarespace. The median across all platforms measured was about 58%. That compares platforms, not AI tools, but it shows a fast layout is not automatic. Fast, mobile-friendly pages are item 4 on our 12-point checklist for that reason.

Hours spent maintaining an AI-built site

Nobody has measured this. We found no survey or study that reports how many hours owners spend updating and maintaining an AI-built website, so any specific number you see for it is a guess. The closest evidence points in three directions.

• Owners have little time to give. In a Fiverr survey of nearly 6,000 owners in 25 countries (July 2024 to March 2025), 70% spent less than five hours a week on marketing in total. Fiverr sells freelance services, so it has an interest in that finding.
• Maintenance is a recurring chore, not a one-off. WPX says sites should be updated at least monthly and checked weekly to quarterly, depending on activity. It gives prices, not hours: $50–$200 a month for freelancer maintenance and $39 to $999+ a month for managed plans. WPX sells managed hosting and cites no source for those ranges.
• AI builders move the work, they do not remove it. B12, which sells AI builders, says the platform handles security patches and that owners make edits themselves immediately. It gives no hours. Google’s guidance adds a review task: it says to manually fact-check AI-generated content, including titles, meta descriptions and alt text.

Be careful with self-reported time savings. In the METR trial covered below, developers believed AI had made them about 20% faster while it actually made them 19% slower. If you track your own time for 30 days, you will have better evidence than any survey here.

What owners spend, and whether they save

The build cost clearly drops. Whether total cost drops is unproven, because nobody has priced the owner’s own time. Published prices (US dollars) show the gap:

Route	Published cost	Source
Wix (AI builder)	$17, $29, $39 or $159 a month on annual billing, a free domain for year one, about $21 a year for a .com after that	Website Builder Expert, updated Sept. 15, 2026
Squarespace	$19, $29, $49 or $99 a month on annual billing, a free domain for year one, $10–$20 a year after	ExpertMarket, updated July 30, 2026
Freelance developer	$50–$150 an hour in the US, then $500–$3,000 a year for maintenance	B12 (sells AI builders)
Paid maintenance	$50–$300 a month on average for a small business site	WPX (sells managed hosting)
Unsourced build claims	$0–$500 for AI builders versus $15,000–$50,000+ for traditional development	NxCode (no source given)

A mid-tier builder plan costs about $348 a year (Wix Core or Squarespace Core on annual billing) plus roughly $10–$21 for the domain after year one. That is about $360–$370 a year, including hosting, and it is the cheapest way to get online. A developer-built site costs more before it launches and, per B12, $500–$3,000 a year to maintain.

Owner time is the missing line. At an illustrative $50 an hour, every hour a month you spend on the site costs $600 a year. Two hours a month is $1,200 a year, more than the $500 low end of B12’s developer maintenance range. A builder subscription also covers the platform only. Page copy, SEO setup and someone to review the result are extra, whether you pay for them in money or in your own time.

The only spending survey we found is old. Clutch’s 2018 survey of 351 owners found 28% of those with a website spent $500 or less and 48% spent $500–$10,000. It predates AI builders, so it shows the old baseline, not the savings.

Time saved: what the studies show

The best evidence says AI cuts writing time sharply, but across a whole working week the saving is small, and for skilled specialists it can be negative. No study tested website building itself.

Study	Sample and task	Result
Noy & Zhang, Science, July 2023	453 college-educated professionals, randomized, mid-level writing tasks	Time down 40%, quality up 18%
METR, 2025	16 experienced developers, 246 real tasks, randomized	Tasks took 19% longer with AI. They had expected 24% faster and afterward believed 20% faster
St. Louis Fed (Aug. and Nov. 2024 surveys)	Representative US workers; 28% used generative AI at work	Users saved 5.4% of work hours (about 2.2 hours on a 40-hour week); all workers 1.4%; aggregate productivity gain 1.1%
Humlum & Vestergaard, Denmark	25,000 workers, 7,000 workplaces, 11 exposed occupations	Average time savings 2.8%; no significant effect on earnings or recorded hours
Thryv survey (May 2025)	540 small business respondents, self-reported	Most AI users say they save 11–20 hours a month. Thryv sells small business software

Three readings matter for a website owner. First, writing is where the proof is strongest: first drafts of page copy are the part of a build most likely to get faster. Second, the saving shrinks when work is judged across a full week (1.4–5.4% of hours), and it is gone for experts doing complex work on their own project. Third, the only large figure, 11–20 hours a month, is the self-reported one from a vendor, and METR shows self-reports can point the wrong way.

The studies measure general AI tools, not AI website builders, and METR studied software developers, not site owners. Treat them as a guide to what to expect, not a forecast. They do point to a sensible split of labor: let AI produce the fast first draft, and spend human time on review, where it pays off.

Verdict: does it help owners?

AI-built websites save money and some time on the build. The data does not show they bring in customers by themselves, and it cannot yet say what they cost in ongoing effort. Customers come from being found, and that is where the work moves.

• Money: yes, on the build. Published prices put a builder site at roughly $360–$370 a year, versus thousands up front and $500–$3,000 a year in maintenance for a developer-built one. Copy, SEO setup and your own time are the costs the subscription does not show.
• Time: probably yes for first drafts, uncertain overall. Controlled studies show 40% less time on writing tasks, but only 1.4–5.4% of total work hours saved across workers. The one large claim (11–20 hours a month) is self-reported and from a vendor.
• Effort: it shifts, it does not vanish. Someone still has to fact-check copy, titles, meta descriptions and alt text, update the site, and check that Google can find it.
• Customers: unproven on their own. No independent before-and-after study exists, and the vendor claims of 18–35% higher conversion have no method behind them. What drives leads is the same as before: a fast, clear site, original content, and being findable in Google and in AI answers. That is why SEO and AEO belong in the web design process, and why a website that is not found can quietly cost you clients.

If you are deciding, the practical test is cheap. Before launch, set up Search Console and analytics (our guide to reading Google Search Console shows how to use it) and write down your current monthly leads and inquiries. Log the hours you spend for 30 days. Check where you rank with our free Google rank checker, and compare leads after 90 days. That gives you the before-and-after data nobody has published, for your own business.

Sources and data quality

Every page below was opened and read for the figures used. Prices and statistics are as of October 17, 2026.

• Vendor and agency pages (commercial interest): Ritner Digital, NxCode, B12, WPX, Thryv, Fiverr, Semrush.
• Research and data: Noy & Zhang, METR (summary), Humlum & Vestergaard (summary), St. Louis Fed, Orbit Media, Pew (summary), Graphite (via Axios), Core Web Vitals by platform, Clutch 2018, Google on AI content.
• Pricing: Wix and Squarespace prices come from review sites, not the companies’ own pricing pages. Check the vendors before quoting them.

Known limits: several studies are second-hand summaries rather than the original papers. The Semrush AI detector is unvalidated in the article, and its 80% and 9% figures differ slightly from the 80.5% and 10% in the same piece. The Clutch survey is from 2018. No source measured maintenance hours for AI-built sites, so that section has none.