Grounding Document: Internet Design, Modern UX
A cultural guardrail for model and designer choices. Hope slop or open web manifesto.
What This Page Is / Is Not
This page is
A cultural steering layer for LLM-assisted UI work.
A set of values that should override "best practices" when those practices are extractive.
A decision compass to attempt to steer auto-complete
This page is not
A compliance checklist.
A conversion playbook.
A technical API spec.
0) Scope and reason for this to exist
LLM Galore // Manifesto
LLM Generated projects with varying degrees of intent and effort are flooding the internet. For an already enshittified and corporate web, this can read as the final nail, or but an escape hatch bringing in the unexpected and oh so dearly missed flow of oxygen.
The web has been getting worse long before the ironically named "Attention Is All You Need" paper. As it centralized around nameless, faceless platforms, expression of personality and genuine human connection was intentionally squeezed out by algorithmic feeds engineered to harvest that attention, replacing human connection with endless slop, organic then, and increasingly synthetic now.
The genie has long been out of the bottle. There is no return to some "good, pre-llm" internet, as it has long been compromised by privacy erosion, dark patterns and platform manipulation. Astroturfing was always there, the only difference is now it is cheap to do, and maybe it is for the best, the devil we know.
And in this very commoditization of technology that finally dooms it, I want to see a glimmer of hope. Despite all the greed and gloom, we do have open weights and inference that gets cheaper by the day. Commodity hardware is already sufficient to scaffold a personal website, generate a script, or build a tiny utility that solves a real pain, and in doing that opens up the veil of "personal" computing just enough to act as the support branch needed for new generation of people who are tired of the walled gardens, artificial limits, and the depersonalization of personal computing. How bad does it get before it gets better, before new rebels and new heroes rise?
As we lose trust in what we see on the web, can it open
a path not towards endless exploitation and manipulation
for the sake of shareholder value, but the rebirth of
the very world that has already once relentlessly sucked
us all in behind the screens, not for the outrage and
despair, but curiosity and hope.
Let's make The Mentor proud.
1) The Rot Predates LLMs
We are not watching a healthy web get broken by AI. We are watching an already degraded web get scaled by AI.
The old loop was already here: harvest attention, hide real costs, add friction to exit, call it optimization. LLMs can accelerate that loop, or help people break out of it.
Doctorow's Platform Lifecycle
"Here is how platforms die: First, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves. Then, they die.
I call this enshittification, and it is a seemingly inevitable consequence arising from the combination of the ease of changing how a platform allocates value, combined with the nature of a 'two-sided market,' where a platform sits between buyers and sellers, hold each hostage to the other, raking off an ever-larger share of the value that passes between them."
Cory Doctorow, Wired.
Grounding decision: when the model is uncertain, do not optimize for extraction. Default to user-legible choices and reversible flows.
2) Fork in the Road
Path A: Automated Enshittification
Generated dark patterns at scale.
Funnel-first UX where users are inventory.
Dependency chains nobody can audit or own.
Path B: Human-Useful Software
Personal tools and small sites built fast.
Readable interfaces with accessible and honest defaults.
Data ownership, no lock-in, lower cost, lower gatekeeping.
3) Why This Still Helps People
Most people do not need a unicorn roadmap. They need a page that loads, a form that is honest, and a workflow they can trust.
LLMs lower the cost of making that reality. If we ground them with the right culture, they become support structure for autonomy, not another extraction layer.
Trust and reputation are slow systems: vouching, repeated behavior, visible accountability. You do not buy those with animation polish. You earn them by being clear, stable, and reversible over time.
Aligned inspirations and examples
These are living references for user-owned publishing and discovery.
- CERN: The World Wide Web Project as an early model of link-first web publishing.
- ArchiveTeam's GeoCities rescue preserving personal-web culture that would have been erased.
- Neocities as modern ad-free, open-source personal publishing infrastructure.
- Marginalia Search for non-commercial discovery and small-web visibility.
- Wiby for lightweight, old-style web search centered on simple pages.
- Bear Blog for minimalist, writer-first publishing with low platform overhead.
4) Values
Tie-break for ambiguous UI choices. Pressure, confusion, or bait = fail.
- + human contact
- − engagement capture
- + user agency
- − conversion pressure
- + clarity
- − persuasion tricks
- + reversibility
- − lock-in
- + ownership
- − dependency
5) Anti-Patterns To Block (LLM + UI Kit)
One table. One point: generated UI must be constrained by defaults that favor users.
| UI Surface | What To Avoid | Safer Default |
|---|---|---|
| Button + Dialog | Bright "Accept", faint "Decline"; hidden close path. | Visual parity for choices; always visible cancel/close; Escape closes. |
| Toggle / Checkbox / Radio | Pre-checked telemetry, marketing, auto-renew. | Default off for non-essential collection; explicit opt-in only. |
| Form + Help Text | Fees/renewal/data-sharing buried in muted microcopy. | Material terms above CTA, normal contrast, plain language. |
| Toast / Alert | Auto-dismissed legal or consent notices. | Persistent alert for material decisions; explicit acknowledgement. |
| Table + Badge + Timer | Fake scarcity/countdown pressure with no verifiable source. | Only show urgency tied to real data and timestamped provenance. |
| Nav / Sidebar | One-click signup; buried cancellation/privacy paths. | Flow parity: leaving must not be harder than joining. |
| Copy Layer | Confirmshaming, double negatives, manipulative phrasing. | Direct polarity ("Enable/Disable"), neutral tone, no guilt language. |
| Dependencies | Third-party scripts injecting urgency/social-proof overlays. | Vendor allowlist, reviewed integrations, self-hosted defaults. |
6) Non-Negotiables
Keyboard parity, focus visibility, sufficient contrast, assistive-tech readable structure.
Web-native semantics, portable HTML, minimal lock-in, graceful behavior without JS bloat.
Readable templates, deterministic tokens, fewer hidden states, local control over dependencies.
7) Call To Action
Ship pages that tell the truth. Keep choices clear. Keep exits easy. Keep ownership local.
If this page does its job, the model does not make prettier traps. It helps people build software they can stand behind.
Transparent systems, accountable defaults, real user agency.
Appendix A: Evidence Links
- CERN: The World Wide Web Project
- Wired: The "Enshittification" of TikTok (Doctorow)
- FTC: Bringing Dark Patterns to Light
- FTC Staff Report (PDF)
- FTC + ICPEN 2024 Dark-Pattern Sweep
- CPPA: Dark Pattern Guidance
- CNIL: Rejecting Cookies Should Be As Easy As Accepting
- EDPB Guidelines 03/2022
- Dark Patterns at Scale (Mathur et al.)
- W3C Ethical Web Principles
- A Dao of Web Design
- Neocities
- IndieWeb Principles
- ArchiveTeam: GeoCities Rescue
- One Terabyte of Kilobyte Age (Tumblr) documenting real homepage vernacular at scale.
- GeoCities Research Institute
- Marginalia Search
- About Marginalia Search
- Wiby
- Bear Blog
- OldWeb.today (Project)
Appendix B: Culture References
GeoCities and The Homepage Ethos
Personal pages as messy identity, not brand templates.
"Under construction" as honest process, not polished deception.
Cameron's World and GeoCities archives preserve this grammar of self-expression.
Early Web Discovery
Before algorithmic feeds, discovery was directory-like and link-led.
CERN indexes and webrings rewarded exploration over lock-in.
Index-like IA in this UI kit is a feature, not a regression.
Marginalia and the Small Web
Marginalia is both a tool and a posture: search the edges, not only the winners.
Independent search/discovery systems keep minority voices findable.
Design implication: preserve side paths, citations, and non-linear reading routes.
Cultural references
Hackers (1995), Escape Velocity, tilde communities, webrings, and old-homepage aesthetic.