ASD-STE100 is a technical writing standard developed for aircraft maintenance manuals. It exists for one reason: a tired technician working a double shift cannot misread a step. The same rules that make technical documentation impossible to misread also make AI output dramatically cleaner. Here are four formats to apply it to, right now.
Most AI output is readable but not clean. The model answers the question and buries the answer in hedge phrases, passive constructions, and sentences that could have been split in half. The model is not dumb. It is writing to a vague prompt. Give it a stricter writing constraint and the output follows.
ASD-STE100 is that constraint. It is a Simplified Technical English standard maintained by the ASD (Aerospace and Defence Industries Association of Europe). The rules: active voice, sentences under 20 words for descriptive text and under 25 for procedural, one idea per sentence, an approved vocabulary where every term has one meaning and one meaning only. The standard was built so that a maintenance manual could never be misread by a non-native English speaker working in a hangar at 2 a.m. The models already know it in full because it is well-documented in their training data. Andrej Karpathy shared four specific ways to put it to work, and all four are worth having in your prompt toolkit.
Step One: Writing
This is the simplest use. Add one phrase to any explanation prompt.
Copy: The ASD-STE100 Writing Prompt
Copy this.
Explain [TOPIC] following ASD-STE100.
That is the full addition. No special configuration. No preamble. The model applies the standard automatically.
What changes in the output: sentences end where they should end. Hedge phrases disappear because STE disallows vague constructions. The vocabulary becomes consistent. Where the default output might spend two sentences saying "it is important to note that this may potentially result in," ASD-STE100 output says "this causes." Shorter sentences, cleaner logic, faster to read.
The 80% Rule
If the output starts sounding like a maintenance manual, add one modifier:
Copy: The 80% Rule Variant
Copy this.
Explain [TOPIC] following ASD-STE100 about 80% of the way.
That modifier softens the strictest vocabulary constraints. The output keeps most of the clarity without losing natural tone. For business writing, 80% lands in the right place: clean without being mechanical.
Your First Real Run
Pick anything you have struggled to get a clean explanation of from your AI tool. A concept you are going to explain to a client. A process you are documenting for your team.
Add "following ASD-STE100 about 80% of the way" to the end of your prompt. Read the output. Compare it to what you normally get from the same tool.
The Mistake That Makes It Fail
Adding it to a vague prompt. "Explain my business following ASD-STE100" gives you nothing useful because the topic is not defined. The standard sharpens precision. It does not add specificity that is not in the prompt. If the input is vague, the output is vague and clean.
Step Two: Diagrams
A cleaner written explanation is still text. Text takes time to read.
If the goal is comprehension, not writing, skip the paragraph entirely and ask for a diagram.
Copy: The Diagram Prompt
Copy this.
Give me a diagram of [TOPIC].
The model produces a text-based diagram: a table, an ASCII flow, a markdown structure. For anything you want to scan instead of read, this is faster. A comparison between two options takes three times longer to process in prose than in a table. A workflow with five steps is clearer as a numbered flow than as a paragraph describing each step.
This format pairs especially well with the writing step. Write the explanation first to get the precision. Then ask for the diagram to get the structure. Use whichever version serves the audience.
Your First Real Run
Take the last three-paragraph explanation an AI gave you. Ask for a diagram of the same content. See which one you can explain back faster after thirty seconds with it.
The Mistake That Makes It Fail
Using it for topics that require nuance. A sensitive client situation does not go in a diagram. A negotiation is not a flow chart. Use this for processes, comparisons, sequential steps, and decision trees where clarity is the goal and depth is not.
Step Three: Web Pages
The models are increasingly capable at front end. When an output format supports HTML, the result is interactive rather than static.
Copy: The HTML Output Prompt
Copy this.
Build me [TOOL OR CALCULATOR] in HTML. Include [INPUT FIELDS] and output [RESULT]. No external libraries. It should run in a browser from a single file.
The key example: instead of asking for an explanation of a calculation, ask for the calculator. With sliders. That responds when the user changes a number.
A concrete version:
Copy: Client ROI Calculator
Copy this.
Build me a client ROI calculator in HTML. It has two inputs: monthly hours saved (a slider from 1 to 100) and hourly rate (a text field). It outputs the monthly value recovered and the annual value recovered. No external libraries. Single file.
You get a working page, not a paragraph describing how the math works. The output has inputs, outputs, and logic. You can open it in a browser and hand it to a client.
This changes the value of what you produce. Before, the deliverable was an explanation. Now it is a tool. The client does not need to trust your math because they can see it run on their own numbers.
Your First Real Run
Think of the last time you explained a calculation or decision framework to a client. Build the HTML version instead. Open it in a browser and check if it communicates the same thing faster.
If the first version is not quite right, give it a correction: "The annual output is showing the monthly number times 10. Multiply by 12." The model fixes it and returns the corrected file.
The Mistake That Makes It Fail
Asking for HTML on something that is not interactive. A narrative does not become better as a web page. This is for tools, calculators, and decision trees where the user needs to enter their own values and see a result. For static content, this format adds nothing.
Also: test the output file in a browser immediately. The model occasionally produces HTML that looks correct but has a logic error in the JavaScript. Catching it takes ten seconds. Sending a broken tool to a client takes longer to explain.
Step Four: Explainer Videos
This is the most capable format and the one that requires the most setup.
You can ask a model to build a fully custom explainer video on any topic. The structure follows visual math explainer style: concept introduced, illustrated, narrated, resolved. The narration renders automatically with an ElevenLabs API key.
Copy: The Explainer Video Prompt
Copy this.
Build a 3-minute explainer video on [TOPIC] in the style of a visual math explainer (3Blue1Brown style). Use my ElevenLabs API key [YOUR_ELEVEN_LABS_API_KEY] for narration. Keep each scene under 10 seconds. Output the full generation script and a production structure.
What you get back is a complete production script and a generation structure. With an ElevenLabs account and API key, the narration renders in the voice you have configured there. The visual animation requires a code environment (Claude Code or Codex) to generate, but the model handles the generation logic from the prompt.
This is throwaway custom software. Before this year, building a one-off animated explainer cost a designer, a voiceover actor, and a half-day of back-and-forth. It now costs the ElevenLabs API call plus the time it takes to write the prompt. The result is a finished narrated video, not an outline.
Your First Real Run
Pick a topic you explain to prospects repeatedly. A framework. A process. Something you have explained in writing or on a call a dozen times and that always takes ten minutes to get through.
Build the explainer video version. Use it in a sales follow-up once and check if it moves the conversation faster than a written summary.
The Mistake That Makes It Fail
Skipping the ElevenLabs setup. Without the API key in place, you get a script and production structure but no audio. The value of this format is in the finished narrated video, not the script. Set up the API key first. The ElevenLabs free tier covers the testing volume.
Also: test on a simple topic before committing to a client deliverable. The visual quality depends on the model you are using and the complexity of what you are asking it to animate. A one-concept explainer is a good first run. A ten-concept explainer is a second or third run.
Why This Matters Beyond These Four Formats
These four formats share one logic: the model can produce more than text. Most prompts ask for text. Most outputs are text. That is the path of least resistance for the model and the least value for you.
Ask for a diagram and you get structure instead of prose. Ask for HTML and you get a tool instead of a description. Ask for a video and you get a deliverable instead of an outline. Ask for ASD-STE100 writing and you get precision instead of padding.
Intelligence is cheap now. The model does more of the work than most prompts ask it to. The job on your side moves to specifying the right output format and making sure the result is correct.
Honest: When These Do Not Work
The writing format delivers the most consistent results across all models. ASD-STE100 is well-documented and in the training data for every major model. The 80% rule is reliable.
The diagram format works well for factual comparisons and linear processes. It breaks on topics where the nuance matters as much as the structure. Do not force a conversation or a negotiation into a flow chart.
The HTML format depends on the model's coding ability. Claude Code and Codex produce better results on complex interactive tools than the web apps. For a simple calculator with two inputs and one output, the web app models handle it without issue. For a multi-step decision tree with conditional logic, use a code environment.
The video format requires the most setup and produces the least consistent results at first. The narration quality depends on your ElevenLabs voice configuration. The visual quality depends on the model and the complexity of the animation. Plan for one or two test runs before the version you use with a client.
None of these formats replace judgment about what the audience needs. A client who wants a quick answer does not need a four-minute explainer video. A team that needs to scan a document on a call does not need a ten-paragraph explanation. Match the format to the situation.
The One Thing to Try Today
If you use AI tools regularly and you have not tried the ASD-STE100 writing format, try it in the next hour.
Pick one explanation you are about to write. Something you will send to a client, put in a document, or use to brief your team.
Add "following ASD-STE100 about 80% of the way" to the end of your prompt. Read the output. That is the full test. Five seconds to add it. The difference is immediately visible.
The other three formats require more setup. They also deliver more value at that investment. Start with the writing format. The others follow.