P-01
Chatbot theatre
Cheerful assistant voice, fake empathy, and a help-center tone on a page that is clearly static marketing.
Signals: "I am here to help", "let us explore", rhetorical questions with no user input.
Critical AI / tech editorial lab
Spot AI slop before it wastes your time
Very Artificial is a sharp, high-signal lab for operators, writers, and small teams who are tired of fake AI productivity, chatbot theatre, and automation claims with no proof.
Manifesto
AI realism over AI theatre
The domain only earns work if it helps people judge claims quickly. Not another generic AI blog. Not SEO slop about slop.
We document how artificial content reads: generic language, fake authority, missing proof, unverifiable agent workflows, and privacy hand-waving. The goal is practical judgment, not smug hot takes.
Useful wedge: a checklist you can run on a landing page, a deck, or a LinkedIn post; pattern strips with bad-to-better rewrites; and notes on what verification would actually look like.
Current themes
What we are tracking now
Editorial focus for the sprint MVP. Themes rotate as examples harden.
- AI realism - state limits, model boundaries, and what humans still must own
- AI slop - interchangeable copy that could describe any product
- Verification debt - claims shipped before anyone can falsify them
- Automation theatre - diagrams and agents without a reproducible run
Pattern gallery
Named failures you will recognize
Each pattern ships with an artifact panel and a rewrite pair. Examples are generic and fictional.
P-02
LinkedIn automaton voice
First-person hustle cadence, numbered lessons, and a conclusion that congratulates the reader for scrolling.
Signals: "Here is what nobody tells you", "game-changer", engagement bait without a verifiable claim.
P-03
SEO slop
Keyword-stuffed headings, thin definitional paragraphs, and internal links that exist for crawlers not humans.
Signals: "In today's fast-paced world", FAQ blocks that never answer the real objection.
P-04
Fake AI agency copy
We leverage cutting-edge models to unlock transformative outcomes - with no stack, no scope, no failure modes.
Signals: infinite capability, zero tradeoffs, no mention of data handling.
P-05
Unverifiable agent workflows
Multi-agent diagrams, tool logos, and orchestration maps with no logs, prompts, or reproducible demo.
Signals: autonomy implied, accountability absent.
Bad to better rewrites
Generic + fake authority
Before
Our revolutionary AI platform empowers teams to unlock unprecedented growth through seamless, intelligent automation.
After
We route support tickets to a draft model. Humans approve sends. A pilot report should name sample size, baseline, exceptions, and rollback path.
Swap superlatives for scope, owner, and evidence a reader could audit.
Workflow theatre
Before
Orchestrator agents collaborate across your stack to autonomously deliver outcomes.
After
Nightly job: export open tasks, classify, write draft comments. Runs on a schedule you can inspect in cron logs.
Show the schedule, inputs, and human gate - not the cast list.
Privacy hand-wave
Before
Simply upload your documents and we will handle the rest with enterprise-grade security.
After
Files stay in your bucket. We process in-region, retain 7 days for debugging, then delete. DPA on request.
Name retention, region, and deletion - or do not ask for uploads.
Checklist
Is this very artificial?
Check what you can observe in the copy or demo. This runs locally in your browser. Nothing is sent anywhere.
Verdict
Select signals above
Score: 0 / 6
The tally updates from your selections only. No server, no tracking.
Method: count checked rows. Three bands - few signals, caution, likely very artificial. Tune thresholds in src/_data/checklist.json and src/assets/js/main.js.
Essays and notes
Drafts and field notes
Longer pieces land here as the lab grows. Status is honest.
Draft
The verification gap in AI demos
Most demos show happy paths. Few show failure, rollback, or what happens when the model drifts.
Outline
LinkedIn automaton voice as a genre
Recognizable cadence, fake intimacy, and conclusions that flatter the reader - a field guide.
Queued
When irony becomes SEO slop
Criticising slop with slop is still slop. How to write about AI without becoming the template.
Experiments
Small lab runs
Low-scope tests that produce artifacts, not slide decks.
Experiment
Rewrite lint for landing pages
Strip superlatives, require one bounded metric or delete the claim. Output is a diff, not a grade.
Experiment
Checklist on real vendor copy
Blind scoring sessions: operators mark signals, then compare notes. No scores published without consent.
Experiment
Pattern strip zine export
Printable pattern cards for workshops - monospace IDs, no stock mascots.
Operating principles
How this lab stays useful
Rules we apply to our own pages before criticising anyone else.
01
Proof over posture
If we cannot show artifacts, we say so. No fake dashboards, no invented metrics.
02
Monochrome first
Hierarchy from type and spacing. Accent color is a signal, not wallpaper.
03
Verification is a verb
Claims must be testable: logs, prompts, retention policy, or a bounded pilot - not vibes.
04
Privacy named or omitted
Do not ask for uploads without retention, region, and deletion in plain language.
05
No slop cosplay
We do not ship rainbow gradients, robot mascots, or testimonial walls to criticise slop.
06
Generic examples
Critical examples stay fictional unless sourced and safe to name.