Does AI website generation make growth-driven design obsolete?

The current tools do more than produce a first draft. Wix's AI website builder asks you to describe your business, goals and style, then generates a site with pages, layout and copy. After that you can keep changing it. You can prompt its built-in agent, Aria, for a new section or a different layout, or adjust any element by hand. Wix says the first version typically takes a few minutes.

So it would be wrong to say AI can only launch a site once and leave it there. The same tool that builds the first version will rebuild a page as often as you ask. If growth-driven design were only a way of getting pages produced, that would be enough to retire it.

Production was never the whole method. Growth-driven design starts with a smaller first release, usually called a launchpad website, and then improves it in repeated cycles. The method's description of that cycle has four steps: plan, build, learn and transfer. The team picks a focus metric and ranks the ideas most likely to move it. After building those, it reviews experiments and audience data to see what worked, and it shares what it learned with marketing, sales and service.

Only one of those steps is production, and that is the step a generator works on. The other steps run on evidence about your audience. Some of that evidence can be handed to a generator. You can give it last quarter's analytics or a set of customer interviews, and it can write from them. What nobody can supply in advance is how visitors will respond to the new version, because that response hasn't happened yet.

When a model writes a headline, it is making an assumption about what a buyer will respond to. The assumption can be a good one, especially when it draws on real customer research. Putting the page in front of people doesn't turn it into a fact. What visitors do in response is the evidence you use to judge whether the assumption was right.

Consider how a page test works. HubSpot's page A/B test puts two versions at the same URL and shows each to half the visitors. You compare how the two performed and choose one. Nothing in that process depends on who or what wrote the versions. The result comes from the visitors.

That is why being able to generate more versions doesn't produce answers at the same rate. Suppose your pricing page gets 600 visits a month, traffic stays level, and a test splits those visits evenly between two versions. Each version gets about 300 visits a month, and some of those are the same person returning. A generator can give you five new headlines whenever you ask. The page still receives 600 visits a month whether you have two candidates or twenty. How long a test has to run depends on how often visitors convert and how big a difference you're looking for, so a month may not settle it. Someone has to decide which candidate deserves the page's traffic, and that decision rests on what you already know about your buyers.

It's tempting to skip the visitors and ask a model how buyers would react. Nielsen Norman Group tried a version of this in a 2024 evaluation of synthetic users. In a study of online courses, the AI-simulated respondents said they had completed their courses and taken part in the discussion forums. The real participants often hadn't finished, and most said they didn't use the forums. The authors advised treating synthetic answers as hypotheses to test with real people. Their evaluation covered particular tools at one point in time, so it isn't a verdict on every model. The distinction it draws still holds. A generated answer about customers is a prediction, and what a customer did is a record.

None of this puts AI outside the method. A website generator is one use of AI, and the method has room for others. Lean Labs sells growth-driven design and uses AI in several parts of the work. In its post on growth-driven design and AI, the agency says its strategy starts from what customers, sales calls and support tickets show, along with research on competitors. Research agents collect competitor and search data in the background. Once the positioning and buyer inputs are written down, AI drafts copy from them. It also produces the headline and call-to-action alternatives that go into split tests.

Look at what the AI is working from in each case. The drafts come from positioning the team wrote after hearing from customers and salespeople. The test variants go in front of real visitors. The post makes the same distinction. In the agency's account, AI can build a site fast, and whether the site wins customers is settled by what visitors do on it.

AI can help with research material as well. It can organize the interview transcripts and notes a team has already collected, and what it organizes still came from people. We look at more of these uses in how AI improves growth-driven design research, testing and optimization.

Generation changes the build step. A team with one of these tools can produce a page, or a new version of a page, by describing it. If building pages was the main obstacle between your team and a better site, that obstacle is smaller now.

The reason for the rest of the cycle is unchanged. You don't know which pages and messages work until buyers respond to them. That is why the method puts a first release in front of buyers and plans to revise it, however the release was built.

Generation also leaves some setup in place. A generator can include forms and other business tools. Someone still has to check that they record what visitors do, and connect the site to any CRM or analytics the company already runs. Those records are one of the main ways a team sees the response to a change, alongside interviews and usability sessions.

A team with steady traffic and someone who reads the results can run this work with a generator and its own tests, without a formal program. Whether to run growth-driven design in-house or hire an agency is a separate question from whether the method still applies.

Checking a change against what visitors did isn't unique to growth-driven design. Other methods do it too. What the growth-driven design cycle adds is a routine for it: a focus metric, a ranked list of changes, a review of the evidence, and a way to pass the lesson on. A generator gives that routine more versions to work with, and the team still has to run it.