anti-ai-slop

Anti AI Slop

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

The guardrail that governs production. You write, structure, and price proposal content so slop never appears in the first place. Detection lives in the companion ai-slop-audit skill, and the reasoning gate lives in critical-analysis-business-logic.

Use When

Apply while drafting or revising every proposal section, table, price narrative, or procurement response.
Apply before an artefact reaches the separate ai-slop-audit checkpoint.

Do Not Use When

Use ai-slop-audit when the job is to inspect and grade an existing artefact rather than produce it.
Do not use this guardrail to fabricate missing client evidence, figures, citations, or proposer credentials.

Inputs

Artefact	Source/provider	Required?	Missing-input behaviour
Draft or planned proposal content	Proposal workflow	Yes	Stop drafting the affected unit until its purpose and audience are known.
Claims and supporting facts	ToR, proposer profile, verified sources	Conditional	Remove or qualify an unsupported claim; never invent evidence.

Outputs

Artefact	Consumer	Acceptance condition
Guarded proposal content	Evaluator and proposal lead	Every section is specific, purposeful, verified or qualified, and covers material constraints.

Evidence Produced

Evidence	Consumer	Acceptance condition
Completed anti-slop ship gate	`ai-slop-audit` and release reviewer	Each item records pass, block, or an explicit unassessed verification gap.

Capability Contract

Read access to the draft and relevant sources is required. Search or network verification may be used when authorised. Editing is allowed only within the parent drafting task; publishing, external submission, spending, and certification claims require explicit authority.

Degraded Mode

Fallback: when a source, network check, or named fact is unavailable, return the narrowest useful qualified wording and mark the check not assessed. Do not turn unavailable verification into a pass.

Decision Rules

Finding	Action	Failure or risk avoided
Unsupported load-bearing claim	Remove, source, or qualify it before continuing	Fabricated evaluator evidence
Generic but non-essential paragraph	Cut it or replace it with an assignment-specific decision	Superficial competence
Specific, verified, purposeful content	Preserve it and continue	Destructive over-editing

Workflow

Identify one proposal unit, evaluator purpose, decision, and evidence-bearing claims; do not draft the full response as one opaque batch.
Inspect the brief, neighbouring sections, and existing proof before changing the unit.
Apply the seven guardrails while drafting, then test the unit against the evaluator's likely counter-case and make one concrete refinement.
Stop when a required fact, price, credential, or citation cannot be supported; recover by verifying, qualifying, or removing the affected claim.
Complete the ship gate for the revised unit, record the evidence, and hand the artefact to ai-slop-audit before selecting the next unit.

Quality Standards

Every section earns its place through a named decision, fact, example, risk, or evaluator need.
No fabricated evidence, repetitive template language, or hidden verification gap reaches the audit stage.

Anti-Patterns

Adding specificity by inventing a number or client name. Fix: verify it or remove it.
Keeping a polished paragraph that changes no evaluator decision. Fix: add substance or cut it.
Waiting for the final audit before correcting filler. Fix: apply the controls while drafting each unit.
Removing an unusual but intentional proposer phrase. Fix: preserve authored voice unless it harms clarity or compliance.
Hiding an unavailable check. Fix: mark it not assessed and state the consequence.

Responsibility overlay

Use the 25 signs to improve clarity and bid credibility, never to accuse a writer of using AI. For each client, assignment, capability, result, legal, price, or schedule claim, retain scope, date, source fit, transformation, limitation, and evaluator consequence. Label hypothetical method examples and never turn generated profiles or past performance into fact. Apply ME1-ME7, preserve purposeful compliance repetition, and mark unavailable evidence NOT_ASSESSED.

Shared standard: AI-slop responsible publishing

References

Machine-error editorial gate (cross-engine Kaizen)

Apply Digital Research's docs/continuous-improvement/machine-errors-editorial-gate-2026-09-03.md to every proposal section:

ID	Proposal adaptation
ME1	Does the section add assignment-specific evidence, outcome, or decision rather than repeat the executive summary?
ME2	Does parallel win-theme language express a real fit, or only balanced rhetoric?
ME3	Can the evaluator understand the method without another explanation of the heading?
ME4	Do superlatives stay within the evidence and Terms of Reference?
ME5	Is each example tied to the named client, sector, assignment, or verified experience?
ME6	Has the same persuasive device become a proposal tic?
ME7	Does the section contain a claim, warrant, evidence, risk, assumption, or deliverable decision?

Merge or cut filler. Preserve repeated compliance wording when it maps to a mandatory requirement, but retain the trace to the Terms of Reference.

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

Apply these rules continuously, as you write, to every sentence, table, claim, and section the moment it is drafted, not only in one pass at the end. The moment you reach for a banned word, a generic value statement, 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 in an executive summary or a methodology, fix it then. Do not defer to a cleanup pass before submission.

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

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):

Superficial competence. Looks fine on the surface, no substance underneath.
Asymmetric effort. Cheap to produce, costly for an evaluator to read, score, and trust.
Mass producibility. Generated at volume.

The human tell named in every domain studied is absence of intent: the sense that no one meant anything by it. In a proposal this reads as a document that could have been submitted to any client for any assignment. Your task is to re-internalise effort, through specificity, verification, and authored choices, before the proposal reaches an evaluator.

The seven universal guardrails (apply to EVERY section)

#	Marker to prevent	Avoidance rule you MUST follow
U1	Genericness and averaging	Every section carries at least one concrete, named, assignment-specific element (a real past project, a named client, a number, a named method, a decision) that a generic template could not produce. Forbid boilerplate value statements.
U2	Superficial competence	Enforce a substance floor: include a claim, a named experience, a number, or a decision the section could not exist without. If you cannot, the section is filler. Cut it or replace it.
U3	Confident wrongness and hallucination	Verify every statistic, citation, quote, named client, court case, statute, and certification before you write it. Cite at the point of claim. Flag uncertainty rather than inventing.
U4	Volume over substance	Prefer one substantive paragraph over three hollow ones. Do not pad to fill a page limit.
U5	Absence of authored voice and intent	State a point of view, a rationale, or a named decision. Ban relentless positivity and sycophancy. Allow honest trade-offs, constraints, and assumptions.
U6	Skipping the hard parts	Cover risks, dependencies, edge cases, and the counter-case, not only the winning narrative. An evaluator trusts a proposal that names what could go wrong.
U7	Mechanical uniformity	Vary sentence length and structure. Break the template. No rule-of-three reflex, no "it is not X, it is Y" formula, no em-dash flood.

Banned and high-risk vocabulary (the lexical tells)

These words and constructions are statistically over-produced by language models (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. This list merges the engine's existing ban with the canonical slop lexicon.

Engine-banned words (already enforced here): delve, tapestry, landscape (as metaphor), navigate (as metaphor), leverage, foster, realm, harness, synergy, embark, robust, vibrant, holistic, seamless.
Additional banned words: 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 is worth mentioning", "let us dive in", "at the end of the day", "in conclusion", "studies show" without a named study, "world-class" and "cutting-edge" as unsupported self-praise, "we are passionate about", "best-in-class".
Constructions: the "it is not just X, it is Y" antithesis; reflexive rule-of-three lists; the em-dash used to manufacture drama; triplet adjectives such as "robust, scalable, and reliable".
French equivalents (for Francophone procurement): "plongeons dans", "il est important de noter que", "force est de constater", "dans un monde en constante évolution", "par ailleurs / de plus / en outre" as filler connectors, "au cœur de", "pierre angulaire".

Drop-in guardrail block (inherit in dependent skills)

ANTI-SLOP GUARDRAIL (inherit in every proposal section):
1. SPECIFICITY FLOOR — every section carries at least one concrete, named,
   assignment-specific element. No boilerplate value statements, no placeholder copy.
2. VERIFY-BEFORE-WRITE — no statistic, citation, named client, court case, statute,
   or certification appears unverified; cite at point of claim; flag uncertainty.
3. AUTHORED VOICE — state a point of view and a rationale; no relentless positivity,
   no sycophancy; allow honest trade-offs and assumptions.
4. COVER THE HARD PARTS — risks, dependencies, edge cases, and the counter-case.
5. BREAK THE TEMPLATE — vary rhythm and structure; forbid the banned-vocabulary
   list above and generic procurement boilerplate.

Proposal-specific avoidance (apply to bid content)

Cover letters and executive summaries: no generic value proposition that would fit any firm or any client; name the client problem and the proposer's specific fit; one stated point of view; vary sentence length, mixing short clauses with fuller 25 to 40 word sentences; no more than one em-dash per paragraph.
Understanding of the assignment: prove understanding with named context, not the phrase "we understand". Reference the actual Terms of Reference, the sector, and the constraint.
Relevant experience: every project card states what was achieved, not only what was done. Outcomes, not activities. No invented client names, figures, or references.
Methodology and work plan: show visible logic. Every high-stakes claim carries evidence, warrant, assumptions, counter-case, and practical implication. Methodology, staffing, timeline, risk controls, monitoring and evaluation, and budget must tell one mutually achievable delivery story, not five well-written but disconnected ones.
Inflated superlatives and hollow analogies: strike "world-class", "revolutionary", "game-changing", and decorative metaphors that carry no information. Replace each with a verifiable specific.
Unverifiable claims: any market statistic, certification, ranking, or "studies show" line needs a named, real source or it is removed.
Financial proposals: day rates, fee breakdowns, and reimbursables must be internally consistent with the staffing schedule and the work plan. No round-number filler.

Ship gate (run before delivering ANY section)

Every section has at least one concrete, named, assignment-specific element (U1 and U2).
Every statistic, quote, citation, named client, court case, statute, and certification is verified (U3).
No banned vocabulary used as filler; the section was scanned against the list above.
The section states a point of view and a rationale; no sycophancy (U5).
Risks, dependencies, and the counter-case are addressed (U6).
Sentence length and structure are varied; no rule-of-three reflex, no antithesis formula, no em-dash flood (U7).
Outcomes-not-activities rule applied to experience; visible-logic chain applied to high-stakes claims.
British English, East African professional tone, and day-month-year dates throughout.
When in doubt, run the ai-slop-audit skill on the draft.

If any box is unticked, the section is not ready to submit.

Impeccable-derived AS overlay

Apply AS1-AS7 to proposal websites, pitch decks, rendered documents, and visual bid assets. In ordinary proposal prose use AS1, AS3, AS5, and AS6; mark visual checks not_applicable.

ID	Proposal overlay test
AS1	Is the structure, palette, type, or copy pattern specific to this buyer and ToR?
AS2	Do eyebrows, chips, metrics, badges, or section numbers clarify evaluation or merely stage it?
AS3	Do win-theme cards and sections distinguish evidence, method, and outcome rather than clone a template?
AS4	Does any motion, glow, or gradient serve a task, state, or accessibility need?
AS5	Are examples, images, credentials, and icons verified and assignment-specific?
AS6	Are buzzwords, em-dash cadence, aphoristic contrasts, and theatrical framing absent?
AS7	Is the rendered proposal readable, complete, accessible, and free of clipping or overflow?

For visual proposals and presentations, purple gradients, glassmorphism, neon glow, AI-beige defaults, decorative editorial scaffolding, and decorative motion are no-ship choices. Preserve mandatory compliance structure and repeated evaluator labels when they carry a traceable function.

About this skill