Scrape Off the Polish: How ASD-STE100 De-Robots AI Writing

Read an AI-generated email aloud sometime. It opens with context nobody asked for, balances every claim with a hedge, transitions with "furthermore," and closes by thanking you for something. Grammatically impeccable. Nobody home. That hollow-professional sheen is what people mean by "AI polish" — and the strange part is how easy it is to switch off.

The fix isn't better adjectives in your prompt ("be natural!" — the model doesn't know what you mean). The fix is a rulebook. Tell the model to follow a real, formal writing standard, and it stops guessing what "professional" sounds like and starts following rules that forbid the polish outright.

The rulebook that works: ASD-STE100, the international standard for Simplified Technical English. It was written for aerospace manuals — where ambiguity kills people and every extra word is a translation risk — and it turns out to be a near-perfect antidote to AI sheen. This guide shows why it works, how to use it in one line, where it's the wrong tool, and how it fits with the other de-slop techniques.

What AI Polish Actually Is

The robotic feel isn't a bug in the model's grammar — it's a default persona. Models are trained to be helpful and inoffensive, and the safest version of helpful sounds like a corporate communications handbook: three paragraphs of context before the point, every claim balanced with its hedge, "in today's fast-paced world" doing warm-up laps around the actual message.

The tells are consistent:

• Delayed points — the answer arrives in paragraph three, after the "context"
• Hedge stacking — "it's important to note that it may often be worth considering"
• Transition padding — furthermore, moreover, additionally, in conclusion
• Adjective inflation — "comprehensive," "seamless," "robust," applied to everything

Humans write like they're in a hurry; the default AI writes like it's being deposed. Polish is the gap between the two.

Why a 40-Year-Old Aerospace Standard Fixes It

ASD-STE100 exists because aerospace manuals are read by maintenance crews with different native languages, at 3 a.m., under time pressure. Ambiguity and decoration are hazards. So the standard enforces rules like: one instruction per sentence, approved-words-only vocabulary, no synonyms (one word, one meaning), no idioms, no clause-stacking.

Those rules attack AI polish at the root, because polish is made of exactly what STE forbids:

And the mechanism matters as much as the rules. When you say "sound natural," the model guesses at a vibe — and guesses toward its trained middle. When you say "follow ASD-STE100," you hand it a rulebook. Models follow explicit formal constraints far better than they interpret aesthetic adjectives. You replaced a matter of taste with a matter of compliance.

The One-Line Implementation

Add this to any prompt, any platform:

Write this using ASD-STE100 Simplified Technical English:
short declarative sentences, one instruction per sentence,
one word one meaning, no idioms, no filler context,
verb-first, active voice.

(The expansion after the standard's name helps models that know the name only vaguely — it turns the reference into operational rules.)

The before/after on a typical request:

BEFORE (default):
"It is important to note that, in order to achieve optimal
results with your automation workflow, one may want to
carefully consider the various configuration options
that are available to you."

AFTER (STE-primed):
"Set these three options before you start the workflow.
The default values work for most cases."

Same information. One of them you'd skim past; the other you could act on.

Where It Shines

• Email and updates — clarity beats warmth in status notes; STE strips the throat-clearing
• Technical docs and SOPs — this is the standard's home turf; output arrives nearly publishable
• Instructional guides — step lists stop narrating and start instructing
• Coding assistants — ask for STE-conformant explanations and the code commentary drops its "Let's dive in!" opening; you get what the function does, in order
• LinkedIn and short posts — the constraint forces a point-first draft you can then warm up yourself

That last bullet hints at the pro workflow: use STE to get the structure and substance clean, then add your own voice in the final pass. Drafting under the rulebook kills the polish; your edit adds the personality.

Making It Permanent

If your platform supports persistent instructions (Claude's Projects and skills, custom instructions elsewhere), bake the rulebook in once:

SYSTEM: Follow ASD-STE100 Simplified Technical English.
- One instruction per sentence. Short declarative sentences.
- One word, one meaning. No synonyms for "use."
- No idioms, no figurative phrases, no filler context.
- Active voice. Verb first. Name the actor.
- Start with the answer. Add background only if asked.

Now every session drafts de-polished by default — and this composes with the other tools in the de-slop kit: the voice-profile approach (samples in, style out) handles who it sounds like, the slop-audit script catches the mechanical leftovers, and the STE rulebook handles how directly it speaks. Different layers, stackable.

When NOT to Use It

Honest limits — STE optimizes for clarity, and clarity is sometimes the wrong mood:

• Creative writing — the bluntness strips atmosphere; poems want their clauses
• Warm sales copy — a personal, emotional letter needs the human texture STE deletes; use "conversational tone, short sentences" instead
• Brand voice with personality — if your voice is playful, STE-first drafting sandbags it; draft with the style profile instead

The rule of thumb: STE for information delivery, style profiles for voice delivery. Most business content is the first thing pretending to be the second.

Frequently Asked Questions

Does this only work on Claude?

No. Any modern model — ChatGPT, Gemini, Claude, the coding assistants — understands what ASD-STE100 means and can follow the rules, especially with the expanded instruction spelling them out. The technique is model-agnostic because it replaces vibes with compliance.

Can I use it for coding and technical tasks?

Yes, and it's where the technique shines most. STE-primed coding assistants give concise explanations without conversational filler — what the function does, in order, no "Let's explore!" Cold documentation is a feature when you're mid-debug at 11 p.m.

Why does default AI output sound robotic at all?

Models are trained to be helpful and harmless, and the safest interpretation of that is a polite professional middle ground: extra words for politeness, hedges for safety, balance for fairness. Grammatically correct, and missing the directness humans use when they actually mean something.

Is STE-conformant output too dry for content marketing?

For the final draft of emotional pieces, possibly — that's the stated limitation. The pro move is STE for the structural draft (substance clean, points ordered), then your voice pass for warmth. You get neither robotic nor fluffy — ordered and human.

Wrap-Up

AI polish is a default persona, and personas yield to rulebooks. Point the model at ASD-STE100 — one instruction per sentence, one word one meaning, no idioms, no filler — and the hollow professionalism falls away because the rules that built it are now violations. One line in the prompt, baked into your system instructions for keeps, skipped for the poems. Clarity over polish: your readers' 3 a.m. selves will thank you.