Claude outperforms Gemini and ChatGPT in free‑tier website redesign test

At a glance:

• Claude (Anthropic) produced a near‑final landing‑page prototype with the fewest edits needed.
• Gemini kept most interactive elements but suffered legibility and generic‑design issues.
• ChatGPT delivered a serviceable layout but altered menus, omitted images and required the most back‑and‑forth.

Why the test mattered

After a seven‑year corporate career, Tanveer – an MBA‑trained PC‑building entrepreneur and regular contributor on XDA – decided to revive his custom‑PC business website. The original landing page, built in 2023, still had solid brand colours, navigation and product images, but its typography and a few design elements felt dated. Rather than hiring a designer, Tanveer wanted to see how far free‑tier large language models (LLMs) had come in three years. He chose three of the most talked‑about cloud LLMs: Anthropic’s Claude (Sonnet 4.6 Adaptive), OpenAI’s ChatGPT‑4o (with Canvas), and Google’s Gemini 2.5 Pro (also with Canvas). By feeding each model the exact same prompt, he could compare not only the visual quality of the output but also how much iterative prompting each required.

The prompt read: “Redesign my website's landing page. It is for a custom PC building business, and has been on pause for the last year. Hence, the PCs and components are outdated. I need you to revise the hardware for each of the builds to reflect the latest components. I also need you to redesign the look and feel of the website, including the hero section, broader layout, typography, brand colors, accent colors, and CTAs. Remember that the site should still retain the interactive elements, and shouldn't remove any of the existing elements. You can add additional elements as you see fit. The overall design should reflect a premium retail business.”

Claude’s performance

Claude’s first output was strikingly close to a premium design. It kept the original dark theme, introduced clean accent colours that popped, and preserved every menu and product name. Within minutes Tanveer had a near‑finished prototype that required only minor tweaks – a redundant text block in one section and a blank rectangular placeholder. Those were removed with a single follow‑up prompt, although the free tier forced a five‑hour wait between messages. The free tier also hit its message limit quickly; Tanveer ran out of free messages before he could explore further refinements. Nevertheless, the overall experience was that Claude needed the least back‑and‑forth, delivering a polished layout on the first try.

Claude’s pricing structure reflects its tiered approach:

• Free plan – unlimited access to basic features but strict message caps.
• Pro plan – $17 / month per user.
• Max plan – $100 / month per person, aimed at teams needing higher limits.

Anthropic positions Claude as an AI assistant that “gets it” right out of the gate, and Tanveer’s experiment supports that claim for quick prototyping.

Gemini’s mixed results

Gemini retained the original menu and product names, which was a point in its favour. Its redesign, however, started out looking generic, relying heavily on stock images and an outdated aesthetic. After several prompts Gemini shifted to a gold‑and‑white colour scheme. The final gold‑and‑white layout was visually acceptable but suffered severe legibility problems: text and background colours often matched, rendering large portions of copy invisible. On the upside, Gemini kept almost every interactive element from the source page, something Claude’s free tier could not fully demonstrate due to its message limits.

Gemini’s free tier is notably generous in terms of message count, allowing users to iterate longer without hitting a hard cap. The trade‑off is that achieving a premium look required more manual direction and multiple redesign cycles.

ChatGPT’s shortcomings

ChatGPT‑4o produced a serviceable landing page on its first pass, featuring a dark background with blue accents. Unfortunately, the model automatically renamed the PC builds and altered the header menus, breaking the brand consistency Tanveer wanted to keep. Repeated prompting failed to restore the interactive components that were part of the original site, and the model omitted all images—a shortcoming shared by Claude and Gemini on their free tiers.

While ChatGPT is freely accessible after creating an account, its output demanded the most iterative work. Tanveer concluded that, for a simple webpage redesign, ChatGPT made him work harder than expected.

What this means for designers

The experiment highlights a clear trade‑off between speed and flexibility. Claude excels at rapid, high‑quality prototyping but its free tier’s strict limits make extended projects impractical without upgrading to Pro or Max. Gemini offers virtually unlimited free interactions, but designers must invest more time to steer the model away from generic templates and fix legibility issues. ChatGPT sits somewhere in the middle, providing a free entry point but often requiring more hands‑on editing to meet brand standards.

For professionals who need quick, polished mock‑ups, paying for Claude Pro or Max is advisable. For hobbyists or teams willing to spend extra iteration time, Gemini’s generous free tier can still yield acceptable results, provided they are prepared to correct design flaws manually. Tanveer’s takeaway: test the free tiers thoroughly before committing credit‑card resources, and match the tool to the project’s complexity.

Technical details and environment

• Operating systems used: Windows, macOS.
• LLM versions: Claude Sonnet 4.6 (Adaptive), OpenAI GPT‑4o (Canvas), Google Gemini 2.5 Pro (Canvas).
• Prompt consistency: Identical prompt supplied to each model.
• Outcome metrics: Number of edit cycles, preservation of original navigation/menu, inclusion of interactive elements, visual polish, and free‑tier message limits.

Final thoughts

Tanveer’s hands‑on comparison underscores how far cloud LLMs have progressed in web‑design assistance, yet also reveals the practical limits imposed by free access tiers. Claude’s superior initial output suggests that Anthropic’s focus on instruction‑following and design‑aware generation is paying off. Gemini’s flexibility and ChatGPT’s ubiquity keep them relevant, but both require more user guidance to reach a premium look. As AI‑driven design tools continue to evolve, the choice will increasingly hinge on budget, time constraints, and the desired level of creative control.