Why Every AI-Built B2B Website Looks the Same and What To Do Instead

Marketers and CEOs keep asking the same question. Why hire a website agency when AI can build a site in an afternoon?

Lovable, Cloud Code, ChatGPT, Base44, and a dozen others will all produce something usable from a single prompt. The problem is what “good” looks like tends to look like a lot of other websites out there.

AI tools are trained on existing patterns. Feed them a generic prompt and they return a generic website: the same layouts, the same hero structures, the same flat imagery, the same feel. 

The fight to get noticed through the noise has escalated dramatically. AI did not solve that fight; it intensified it.

In this Exit Five roundtable, Airfleet CEO Elad Hefetz and Art Director Dina Goldshtein make an important point: AI output is only as good as the expertise and the knowledge base feeding it. Take either of those away, and you get the website everyone else gets.

TL;DR

• AI tools generate from existing patterns; that is why most AI-built websites look the same.
• The marketer’s job is no longer to make content; it is to direct AI. That requires expertise and a robust knowledge base.
• A strong creative brief, brand guidelines, and a curated source of truth all do more for output quality than the choice of tool.
• Dina walks through four real client examples where AI tools sped up the work without flattening it: Plexivity, a conceptual lab visualization, a Claude-inspired interactive diagram, and Rig’s 3D mascot.
• Tool choice matters less than process and expertise.

Expertise Is the Variable That Matters

Dina makes the case directly. AI did not replace her job; it made her better at it. The reason is simple. A director’s job is not to make content look nice. It is to make content feel valuable. That requires judgment, and judgment comes from experience.

If I was a junior designer, I would probably use AI in a wrong way, because I didn’t have the experience to manage teams or to manage a project at scale.

Dina Goldshtein

Art Director

The same logic applies to marketers building websites with AI. Elad notes that most teams reach for a single tool and a single prompt. They expect Lovable or Claude Code to give them a finished site. Then they wonder why it looks like every other AI-built site.

What is missing is the work that happens before the first prompt. Strategy. Story. A clear sense of what the website needs to convey and to whom. AI cannot do that work for you. It can only execute on it once you have done it yourself.

The Knowledge Base Is the Other Variable

Expertise is not enough on its own. The AI also needs context, and the better the context, the better the output.

Elad describes his own workflow. He uses NotebookLM as a brand knowledge base, collecting everything about a client in one place: positioning, audience, product, voice, prior assets. He then distills that into a single concept or story for the website. Only then does he move to the build tool.

If your website doesn’t have a story before you start prompting, no tool will give you one. AI cannot invent strategy in a vacuum. It can only reflect what you feed it.

This is why every example Dina walks through starts the same way. The client materials, the brand guidelines, the existing brand book, the strategic context. All of it lives in a structured knowledge base before the first AI prompt gets written.

Four Examples of Direction in Practice

Dina walked through four recent projects to show what this looks like in real client work.

1. Plexivity: A Scrollable Video Hero

Plexivity needed a hero that combined a brand refresh with AI and motion. Dina’s solution was a scroll-controlled video that reveals the brand as the user moves down the page.

The process started with two static frames generated in Midjourney: a beginning and an end. From there, Dina assembled a short video in After Effects, correcting the imperfections that AI image generation always produces (the car in the wrong lane, the orb at the wrong angle, the perspective slightly off).

To make the video usable in development, the team exported it as 160 individual PNGs and played them back frame by frame on scroll.

The detail Dina lingers on is worth pulling out. Frame-by-frame animation is a technique invented more than a hundred years ago. AI generated the imagery; a classical animation method delivered it to the browser. New tools, old craft.

2. Connected Labs: A Conceptual Pitch

Dina pitched a hero that zooms through a lab and lands inside a monitor, where it displays the platform for an AI laboratory management company.

The client did not pick this concept. Dina included it as an example because it shows the design concept is repeatable. Once the workflow is clear, the same approach can carry very different messages.

3. A Claude-Inspired Interactive Diagram

Dina didn’t want a static list of use cases on a client’s website. She used a diagram on Claude’s website as a starting point for a more interactive way for users to explore what’s possible in the product.

Her workflow for recreating something similar was to start with Claude to write a prompt she could fill in with client information. Then she opened NotebookLM, where the client’s materials were already loaded, and asked it to fill in the prompt. She fed the populated prompt back to Claude, which generated an interactive HTML diagram on the first attempt.

Then she uploaded the brand guidelines and asked Claude to apply them. The final result was a branded, interactive, editable HTML element ready to hand to a developer or drop on the site.

Two or three prompts produced a usable artifact. The reason it worked was not the prompts. It was the knowledge base sitting behind them.

4. Set: An AI-Generated Illustrator Brief

Set, an AI automation company for playable ads, needed six pages with distinct narratives. The technology page needed to convey “super intelligence.” Dina had a Buddha-inspired idea and turned to ChatGPT to generate a visual reference for her illustrator.

The first pass was too flowery and cheerful for the client. They wanted something less colorful and more focused on the technology, which allowed Dina to use AI to help refine the brief.

AI did not replace the illustrator. It compressed the briefing step from days to hours and gave the illustrator something concrete to work from on day one.

5. Rig: A 3D Mascot Without Blender

Rig, a cybersecurity startup, came to Airfleet with an established brand book and a 2D mascot named Rigo.

Dina felt the 2D illustration was too childish for a very technical brand. Her prompt was simple but effective: make this illustration a 3D character.

Five years ago, that would have meant building Rigo from scratch in Blender. With one prompt, she could look at the same character from different perspectives. This opened up options for blog illustrations, iconography, and animated hero moments.

The new hero animation of Rigo took about five hours. Most of those hours were spent communicating with the client to confirm the details.

What Marketers Should Actually Do Differently

The roundtable closes on a practical question from one of the attendees. If a marketer is choosing between Webflow, WordPress, Lovable, Cloud Code, or building from scratch, how should they decide?

Elad’s answer is honest. It does not matter as much as people think. Any of those tools can produce a good website. None of them can produce a good website on their own.

What matters is the surrounding discipline. A website is not a one-time project. It is a living system. That means a few things have to be in place regardless of which tool builds the first version:

• A CMS that keeps content and development separate so the site stays maintainable.
• A design system the AI is forced to follow, ideally with CSS tokens, so that future edits do not produce inconsistencies across pages.
• Components rather than hard-coded content, so a single edit propagates correctly.
• A clear understanding of what the website is for. In B2B, Elad puts that in three buckets: sales enablement, brand awareness, and pipeline generation.

The pitfall Elad sees most often is the gap between what AI delivers and what’s ideal. If the marketer doesn’t know the ideal, the problem isn’t caught until too late. Bespoke AI web builders typically handle security defaults reasonably well. It is much weaker on SEO, analytics, integrations, and the dozens of small structural decisions that compound over the life of a site. If you don’t know to ask for something specific, it will get missed by AI.

Speed to Result Isn’t a Substitute for Experience

At least two ideas from this webinar are worth taking seriously.

The first is that the marketer’s role is shifting from making to directing. AI changes the speed of delivery, not the quality of the output the marketers are capable of delivering. The marketer still has to know what good looks like, what the site is for, and what the brand is supposed to convey. That work can’t be outsourced to AI.

The second is that a solid knowledge base has far more influence on your final result than the quality of the prompt. Every example Dina walked through only succeeded because of the context AI did not have on its own: brand guidelines, client materials, strategic positioning, and a clear narrative for the page. The prompts were short. The preparation was long.

If your AI-built website looks like every other AI-built website, the answer isn’t a better tool. It’s a clearer brief, a stronger knowledge base, and someone with the experience to direct the work.