Simplified Technical English: we tested it on our own blog

What Simplified Technical English is

The full name is ASD-STE100. ASD is the AeroSpace, Security and Defence Industries Association of Europe. The airlines of Europe asked for the standard. The association that came before ASD, AECMA, wrote the first version.

The standard has two parts. The first part has 53 writing rules. The second part is a dictionary of about 900 approved words. Each approved word has one meaning and one part of speech. For example, "close" is a verb that means "to move a door or a cover". It does not mean "near".

The current issue is Issue 9. ASD published it in January 2025. With Issue 9, ASD-STE100 became an international standard. A group in ASD, the STEMG, keeps it up to date. The group started in 1983.

The standard is free. You can get it from ASD. ASD owns the copyright and the trademark.

The main rules

These are the rules that change a text the most:

• Write short sentences. An instruction has a maximum of 20 words. A description has a maximum of 25 words.
• Write one idea in each sentence. Give one instruction in each sentence.
• Use the active voice. Write "Open the valve", not "The valve is to be opened".
• Use simple verb tenses. Use the present, the past and the simple future.
• Do not use the -ing form of a verb. The rules permit a technical name such as "landing gear".
• Use approved words. Use one word for one meaning. Use the same word for the same thing each time.
• Keep noun clusters short. Put a maximum of three nouns together.
• Keep paragraphs short. A paragraph has a maximum of six sentences.

Here is an example. The first text breaks four rules:

Who uses it now

The standard started in aerospace. Now most users are in other industries. ASD reports that 64% of the people who got the standard up to December 2024 came from outside aerospace and defence. These users make cars, wind turbines, medical devices and medicines. They have the same problem as the airlines. Many readers do not speak English as a first language, and a mistake is expensive.

Test 1: our own blog

We wrote a checker for the rules that a program can check. Then we gave it every paragraph from our 39 blog posts. That is 2,125 sentences and 34,305 words.

One sentence in six is too long. The average is good, at 16.1 words. The problem is in the long tail. One sentence in ten has more than 28 words.

Our newer posts get the lowest scores. The post about Cloudflare Workers AI has a score of 68%. The post about the free GPT-OSS API has a score of 91%. The newer posts give more measurements and more opinions. We put both into long sentences.

Test 2: can AI fix it?

We took the first failed paragraph from 24 posts. We sent each paragraph through our own gateway, with the rules as instructions. Then the same checker examined the result.

We also counted the numbers. A rewrite that changes "12.7" to "12" changes a fact. So we compared every number in the old text with the numbers in the new text.

The AI kept every number in both versions. That is the most important result. But the first prompt had three problems, and the table shows them.

Problem 1: too many sentences in a paragraph

The first prompt left five errors. All five were paragraphs with too many sentences. The AI cut long sentences into short ones, but it kept them in one paragraph. One paragraph had 17 sentences.

We did not ask the AI to fix this. A program can fix it, so our code splits the paragraph and does not change a word. This change removed all five errors.

Problem 2: sentences that are too short

The average sentence fell from 19.4 words to 7.5 words. The rules give a maximum, not a target. A text of seven-word sentences is correct, but it is hard to read.

In the second prompt, we told the AI that a sentence of 12 to 20 words is normal. The average went up only to 7.9 words. This instruction did not work well. The AI still cuts sentences that it does not need to cut.

A person who writes to these rules has the same problem. The average sentence in this article has 8.9 words. Only two sentences fail, and they are the bad examples that we show on purpose. Short sentences are easy to get. Sentences of a good length need more work.

Problem 3: lost opinions

One of our posts says that a product is "one of the nicest small products released this year". The first prompt changed this to "a good small product". The meaning became weaker. The rewrite also removed 12.5% of the words.

In the second prompt, we told the AI to keep opinions, comparisons and qualifications. The result was "a very good small product from this year". The rewrite removed only 5.8% of the words. This instruction worked.

One problem stayed. The AI changed some descriptions into instructions. Our original text said "you pick a model, you type". The rewrite said "Select a model. Type your text."

We told the AI not to do this, and it still did. Read each rewrite before you use it.

What a program can check, and what it cannot

A program can count words in a sentence and sentences in a paragraph. It can find most passive verbs. These checks are exact or close to exact.

A program cannot easily check the dictionary. The dictionary controls the meaning of each word, and the meaning depends on the sentence. "Close the door" is correct. "The pump is close to the wall" is not correct, because "close" must be a verb. To find this error, a program must know the part of speech of each word.

The -ing form has a similar problem. "The setting is wrong" uses a noun. "He is setting the valve" uses a verb. Our checker shows each -ing word as a warning. It does not count the warning as an error.

Try it

We made the checker into a free tool. Paste your text, and the tool marks each sentence that breaks a rule. If you sign in, the tool can also rewrite your text. The rewrite goes through our gateway and uses your free quota.

You can also call the check from code. It needs no key:

curl https://dreamprompting.com/api/v1/ste/check \
  -H "Content-Type: application/json" \
  -d '{"text": "The valve is opened by the operator."}'

What we learned

• One sentence in six on our blog is too long for the rules.
• AI can remove almost all rule errors, and it kept every number in our test.
• AI makes sentences shorter than the rules require. The result is correct but hard to read.
• An instruction to keep opinions works. An instruction to keep descriptions as descriptions does not work well.
• A program can check sentence length and passive verbs. Only a person or a full language tool can check the dictionary.

ASD-STE100 and Simplified Technical English are trademarks of ASD. This article and our checker are not made or approved by ASD. Our checker does not certify that a text conforms to the standard.