Anti AI Slop

Anti AI Slop

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.

The guardrail that governs production — writing/designing/coding so slop never appears in the first place. Detection lives in the companion ai-slop-audit skill.

Use When

• Generating ANY output that a person will read, use, or judge: text, document, UI, code, image brief, social post, slide, spec, plan, proposal, or marketing copy. This is a live constraint, not a final-only gate.
• You reach for a banned word, a generic placeholder, an unverified figure, or a template default — stop and correct it in place.

Do Not Use When

• Never skip it. This skill applies to every generated artefact. If the output is trivial, the ship-gate still takes seconds.

Required Inputs

• The artefact being produced and its output type (so the right domain block applies).
• Any verifiable facts, names, numbers, citations, packages, or APIs the artefact references.

Real-time application (this is a LIVE constraint, not only a final gate)

Apply these rules continuously, as you generate — to every sentence, component, function, slide, or frame at the moment it is written, not only in one pass at the end. The moment you reach for a banned word, a generic placeholder, an unverified figure, or a template default, stop and correct it in place. The ship-gate checklist at the end is the final confirmation, not the first time these rules are consulted. If you are mid-draft and notice slop accumulating, fix it then — do not defer to a cleanup pass.

What "AI slop" is (so you know what you are preventing)

Crafted delivery loop

Anti-slop is an attention discipline, not a vocabulary filter. Work in small, named slices and keep the human decision visible:

1. Frame the audience, job, consequence, constraints, and evidence boundary.
2. Select one flow, section, state, component, claim, or content unit.
3. Inspect the existing context and data or argument flow before changing it.
4. Make the smallest useful change with real content, states, and ownership.
5. Exercise it in context, including a relevant failure, empty, denied, interrupted, or counter-case path.
6. Refine one concrete defect or remove one element with no job; preserve what works and record the proof before selecting the next slice.

This loop applies to the kaizen operation and to its audit artefacts. A complete first draft, a green lint result, or a polished screenshot is not evidence that the underlying product has been understood. Missing proof is NOT ASSESSED.

AI slop is low-quality content produced in quantity by generative AI and pushed at people who did not ask for it (Merriam-Webster 2025 Word of the Year, verified). Its three diagnostic properties (Kommers et al., "Why Slop Matters", arXiv 2601.06060, verified):

1. Superficial competence — looks fine on the surface, no substance underneath.
2. Asymmetric effort — cheap to produce, costly for a human to read/review/fix.
3. Mass producibility — generated at volume.

The human tell named in every domain studied: absence of intent — the sense that no one meant anything by it. Your job is to re-internalise effort (specificity, verification, authored choices) before the artefact reaches a person.

The seven universal guardrails (apply to EVERY output)

Banned / high-risk vocabulary (the lexical tells)

These words and constructions are statistically over-produced by LLMs (FSU/COLING-2025; PubMed "delve" +400%). Do not use them as default register. A word here is allowed only when it is the genuinely precise term, never as filler.

• Words: delve, tapestry, realm, landscape (as metaphor), navigate (as metaphor), leverage, foster, harness, synergy, embark, robust, vibrant, holistic, seamless, intricate, commendable, meticulous, pivotal, underscore, testament, resonate, elevate, paramount, unwavering, multifaceted.
• Phrases: "in today's fast-paced world", "in the ever-evolving landscape of", "it is important to note that", "it's worth mentioning", "let's dive in", "here's the kicker", "at the end of the day", "in conclusion", "studies show" (without a named study).
• Constructions: the "it's not just X, it's Y" antithesis; reflexive rule-of-three lists; em-dash used to manufacture drama; relentless triplet adjectives ("robust, scalable, and reliable").
• French equivalents (for Francophone output): "plongeons dans", "il est important de noter que", "force est de constater", "dans un monde en constante evolution", "par ailleurs/de plus/en outre" as filler connectors, "au coeur de", "pierre angulaire".

Drop-in guardrail block (inherit in dependent skills)

ANTI-SLOP GUARDRAIL (inherit in every output):
1. SPECIFICITY FLOOR — every section carries >=1 concrete, named, domain-specific
   element. No tool defaults, no placeholder copy.
2. VERIFY-BEFORE-EMIT — no statistic, citation, quote, named entity, API, or
   dependency ships unverified; cite at point of claim; flag uncertainty.
3. AUTHORED VOICE — state a point of view / rationale; no relentless positivity,
   no sycophancy; allow trade-offs.
4. COVER THE HARD PARTS — error/edge/empty cases, counter-arguments, risks.
5. BREAK THE TEMPLATE — vary rhythm and structure; forbid default aesthetics and
   the banned-vocabulary list above.

Domain-specific avoidance (load the relevant block for the output type)

• Written content: no focal-word clusters; vary sentence length (mix 3-10 with 25-40 words); <=1 em-dash/paragraph; no "in conclusion"; specific examples over generic claims; a stated point of view.
• UI/UX: no indigo-500/purple-gradient default; no Inter-only typography; intentional (not uniform) radius/shadow; WCAG contrast >=4.5:1; design ALL states (error/empty/loading/focus/disabled); real testimonials only.
• Apps/product: add genuine value over any wrapped API; transparent pricing + in-app cancel; no deceptive AI-capability claims; least-privilege permissions.
• Code: verify every imported package exists (no slopsquatting); no placeholder stubs/TODOs in shipped code; parameterised queries; no hardcoded secrets; idiomatic, deduplicated; real tests (not assert true); cover edge cases.
• Image/video: real specimens only; check anatomy/text/physics; avoid the "AI sheen"; attach provenance (C2PA) where it matters.

Ship gate (run before delivering ANY output)

• Every section has >=1 concrete, named, specific element (U1/U2).
• Every stat, quote, citation, named entity, API, dependency verified (U3).
• No banned vocabulary used as filler; scanned the output for the list above.
• The output states a point of view / rationale; no sycophancy (U5).
• Error/edge/risk cases and counter-arguments addressed (U6).
• Sentence length and structure varied; no rule-of-three reflex, no antithesis formula, no em-dash flood (U7).
• Output type's domain block applied.
• When in doubt, run the ai-slop-audit skill on the draft.

If any box is unticked, the output is not ready to ship.

See also

• ai-slop-audit — the detection/evaluation/audit companion (analyse any artefact for slop).
• The host engine's house-style / language-standards skill (apply on top of this).

Inputs

Outputs

Evidence Produced

Machine-error editorial gate (cross-engine Kaizen)

Apply Digital Research's docs/continuous-improvement/machine-errors-editorial-gate-2026-09-03.md to requirements, architecture notes, code comments, UI copy, release notes, and handoff text:

Cut or merge units with no semantic delta. Preserve repeated identifiers, schemas, error messages, security warnings, and traceability fields when they are functionally required.

Impeccable-derived AS overlay for product and web work

Apply AS1-AS7 when the change includes UI, a website, a dashboard, a presentation, or rendered documentation. Reject purple gradients, glassmorphism, neon glow, AI-beige defaults, decorative editorial scaffolding, and decorative motion as visual no-ship choices. Functional state, accessibility, data encoding, and approved design-system reasons must be recorded.

Unavailable browser, render, or detector evidence is NOT_ASSESSED, never clean. Preserve repeated identifiers, schemas, error states, security warnings, and traceability fields when their function is evidenced.

Capability contract

Apply with the capabilities authorised for the parent task. Verification may use search, execution, or network access when available; otherwise mark unverifiable claims and remove unsupported specificity.

Degraded mode

Fallback when a fact, dependency, citation, or visual property cannot be checked: do not present it as verified. Return the narrowest useful output with the verification gap named.

Decision rules

Workflow

Apply the seven guardrails while producing each unit, run the output-type checks, then complete the ship gate before release.

Quality Standards

Every claim is verified or qualified, every section earns its place, and blocking security, dependency, citation, or accessibility defects stop release.

Anti-Patterns

• Adding specificity by inventing a number. Fix: verify it or remove it.
• Keeping a polished paragraph with no decision or evidence. Fix: add substance or cut it.
• Treating the final audit as the first quality pass. Fix: apply controls during generation.
• Rewriting an authored phrase merely because it is unusual. Fix: preserve intentional voice.
• Hiding an unavailable check. Fix: name the verification gap and its consequence.

References

• ai-slop-audit supplies the independent detection and grading pass.