Designing Beautiful Websites with AI

I want to build websites with real aesthetic identity — not just competent, but beautiful. AI can now produce a decent site fast. But "decent" is the problem: AI-generated design has the same failure mode as AI writing. Competent, polished, and generic. A recognizable house style that isn't your style.

I've tried to fight this on two recent projects — Developmental Spaces and the Second Renaissance ecosystem site — by hand-picking a few input images rather than writing a pure text brief. For Developmental Spaces that meant the tower/spiral logo as a visual anchor; for the ecosystem site, some report-design references pulled from an Are.na board (not fully certain which ones made it into the actual prompt). This site (rufuspollock.com) has had an even looser, more ad-hoc process.

Both shipped. Both are fine. Neither has real personality — still reads as AI-average.

The theory of why is already written up in my ux-patterns project (howto-moodboards.md): vague briefs make a model regress to the statistical centre of "good design." A moodboard is supposed to fix this — it's a constraint device, not just inspiration. A few hand-picked images is a thin version of that, and it wasn't enough. What's missing is a real curated set, and — harder — the eye to curate it well, since I don't have strong design taste to begin with.

The idea I keep circling back to (see appendix, and planning/ideas/inspiration-by-ai.md) is offloading search to AI while keeping selection human: AI pulls real candidate sites from curated galleries, I pick between pairs, fast, repeatedly, until a felt direction emerges — closer to preference-learning than to browsing or prompting.

Screenshots of where things stand:

Next: work out whether a pairwise-picker tool actually closes this gap, or whether it's a reps/skill problem no tool shortcuts.

---

Appendix: SCQ

Level 1 — general

• S: Want to build beautiful, aesthetically distinctive websites. AI can now produce output fast.
• C: AI design output is competent but generic — same "middle of the road" failure as AI writing.
• Q: Can AI actually help produce great-quality websites, and how?

Level 2 — this attempt

• S: Tried a thin moodboard (a few hand-picked images) on Developmental Spaces and the ecosystem site, rather than a text-only brief.
• C: Still shipped generic-AI-looking results. Full moodboarding process (per ux-patterns/howto-moodboards.md) was never really run — too slow/high-friction to do properly, and curation itself needs an eye I'm not confident I have.
• Q: Does a faster AI-source / human-pairwise-select mechanism (see planning/ideas/inspiration-by-ai.md) close the gap — or is this a design-reps/skill problem no tool shortcuts?

Appendix: Related Projects & Ideas

• Design Reps with AI (planning/projects/2026-design-reps-with-ai.md) — project to build real design skill through daily small constrained exercises with AI. Prescribes moodboarding before every rep; source of the "moodboard first" rule that kept getting skipped on both live attempts.
• Inspiration by AI (planning/ideas/inspiration-by-ai.md) — idea for a chat/pairwise tool that sources real reference sites from design-award galleries and lets you converge on a direction via fast A/B picks instead of open browsing. Full text copied into Appendix 2 below.
• How to Get Good, Average… — working notes on why vague briefs produce averaged-out AI output, and how moodboards function as constraint devices rather than just inspiration.
• ux-patterns archive and report-inspirations — existing curated reference collections (50+ sites with screenshots) — raw material for any future moodboard, not yet drawn on for either live attempt.
• Developmental Spaces — live site, first design-rep attempt; input was the tower/spiral logo image.
• Second Renaissance ecosystem site — live site, second design-rep attempt; input was some Are.na report-design references.

Appendix 2: "Inspiration by AI" (idea, verbatim)

Interactively describe what you want and work with AI to rapidly find and refine visual examples.

Key framing note (for us)

Core insight to foreground: chat is becoming the primary interface for sense-making and creation.

This product applies that shift to inspiration and early-stage design:

• Instead of searching, scrolling, and bookmarking,
• you talk your way into a mood board.

Conversation replaces search queries. Selection replaces endless browsing.

Product pitch (working draft)

User story

When I start a new digital project—especially a website—the first thing I want is a mood board.

I want to quickly gather examples that feel right:

• websites,
• layouts,
• maps,
• icon styles,
• tone and structure.

I don't want to search, filter, and scroll for hours. I want to talk my way into a useful reference set.

The problem

Creating mood boards for digital products is slow and fragmented:

• Inspiration is scattered across search engines, Pinterest, bookmarks, articles, screenshots.
• Each project starts from scratch, even though relevant material already exists.
• Visual tools are good at display, but bad at intent.
• Notes about why something matters are lost or separated from the reference itself.

What's missing is a fast, conversational way to assemble a project-specific set of references.

The product

An AI-first tool for creating mood boards for digital projects, where chat is the primary interface.

You start by describing what you're working on:

"I'm creating a website for X."

From there, you converse your way into a board:

• Ask for examples.
• React to what you see.
• Refine by choosing options.

The system gathers real references—not generated images—and assembles them into a simple board or list.

How it works (conceptually)

• Chat-led discovery — You describe, react, refine.
• Visual options + selection — You're shown concrete examples and choose what resonates.
• Fast accumulation — The goal is speed, not polish.
• Light annotation — Short notes attach directly to references: "use this layout logic", "tone feels right".

A board can be as simple as a list of links, images, and notes.

What AI is doing

AI replaces friction, not judgment:

• It finds and groups relevant examples quickly.
• It offers visual refinements as choices, not commands.
• It lets you move forward without formulating perfect search queries.

No image generation is required.

What this is (and isn't)

This is:

• A fast way to create mood boards for digital products.
• A conversational interface to visual inspiration.
• A lightweight, project-specific tool.

This is not:

• A design or layout tool.
• A generative art system.
• A heavy asset-management platform.

Core value proposition

Create a useful mood board for a digital project in minutes by talking your way to real examples instead of searching, scrolling, and bookmarking.

Mental model

ChatGPT for inspiration gathering. Cursor-style workflows, but for visual references instead of code.

Refinement: pairwise comparison mechanism (2026-07-18)

Distilled summary. Sharpens the "how it works" mechanism above into something concrete:

1. Brief first. I give a short brief on the site/project I'm building.
2. Sourced pool. AI pulls candidate sites from curated design-award/showcase lists (Awwwards, One Page Love, CSS Design Awards, Siteinspire, etc.) and pre-filters loosely against the brief.
3. Pairwise choice. Shown two sites side by side, I pick the one closer to what I want — optionally with a short comment on why ("like this palette, not this layout"). Repeat with two new sites each round.
4. Fast convergence. A handful of rounds converges on a felt sense of direction — effectively a Tinder-style swipe/A-B refinement toward a mood board or lightweight design system, faster than manual searching/bookmarking.

This is the same "chat your way into a mood board" premise as above, but pairwise comparison (not open-ended chat) is the core interaction — closer to preference-learning than search.

Motivating application: the Developmental Spaces website — currently functional but missing the theme/vibe I want, and I lack an easy way to hand AI a design system or mood board reference to get better output. This idea is the missing step before another design pass there.

Relation to other work: overlaps with Design Reps with AI's point that moodboarding-before-designing is a skipped step worth habitualizing; this idea is a tool to make that step fast instead of manual. Existing curated references (ux-patterns project, ux-patterns/report-inspirations/) are raw material this tool could also draw on, not just award-site scraping.

Projects

• check for domains: inspirationby.ai is available. (inspirationbyai.com is not)