Chapter 5: Vocabulary

I lost my dictionary earlier today... and honestly, I have no words to describe how I feel. 🤪

In conversations with especially 1:1 language tutors that adjust lessons to individual students, vocabulary comes up as an important part that can take up prep time. Both when creating glossary for other materials and when creating specific vocabulary exercises. We hope to offer some inspiration and help to save time on those tasks.

Vocabulary lists

The prompts in this section focus on getting AI to list relevant words. They don't specify the output, which you may remember from earlier is good to be specific about. However, output is covered in the next section and offers multiple options that can be combined with the prompts here.

Themed vocabulary

A common task is to create vocabulary for a specific topic ot theme. AI can help with this and the key is being specific about the topic and the learner's level.

Generate 15 vocabulary items about health for a B1-level adult ESL learner.

The first prompt is specific about learner and topic. However, health is a broad topic, and if you don't get what you intended, it may pay off to be more specific like the next prompt.

Generate 15 vocabulary items about visiting the doctor for a B1-level adult ESL learner.
Focus on words and phrases needed to describe symptoms and understand basic instructions.

The second prompt narrows in on a specific situation and provides further guidance on the words. You can be more specific and for instance, ask for specific word classes or indicate words to avoid or exclude.

Generate 15 verbs and noun phrases about visiting the doctor for a B1-level adult ESL learner.
Focus on describing common symptoms and understanding medication instructions, not medical terminology.

Remember that AI generates text by predicting what comes next based on patterns. Occasionally it may include something you asked it not to, because the negation gets lost in the statistical process. If you find this happening, rephrasing as a positive instruction e.g. "focus on customer-facing vocabulary", can be more reliable than a negative one e.g. "not technical terms".

Frequency-based vocabulary

Sometimes you need the most common words of a certain type rather than words for a specific topic. This may be used to work through a level or as input for later grammar exercises. Below are some examples.

What are the 15 most useful B1-level adjectives for describing people's personality and character in English?

Generate the 20 most common irregular verbs an A2-level English learner should know, ordered roughly by frequency of use in everyday conversation.

The chatbots may be able to access word frequency databases or external lists if they are "thinking" and able to make requests to these, but otherwise it is more fuzzy gotten from its training data and may not be fully accurate. However, unless you need to be absolutely sure, it is probably good enough, and you can always eyeball the result and remove some if not relevant.

Personalized vocabulary

Personalizing vocabulary to a learner's profession or interests can be valuable and easy with AI. A lawyer, a nurse, and a software developer all need English, but they may need very different words. Try adjusting the below prompts to your learners and needs.

My student is a B2-level German speaker working as a sales manager for an automotive company. He regularly attends international trade fairs and needs to present products to potential clients.
Generate 15 vocabulary items he would need for greeting clients at trade fairs, presenting product features, and following up after meetings.

The specificity here, i.e. trade fairs, automotive, presenting to clients, pushes the AI past generic "business English" and into vocabulary this fictive learner may use next week.

AI generally does well with common professions but can be less reliable for niche fields. If your learner works in marine biology or heritage conservation, you may need to double-check the terminology.

Extracting vocabulary from texts

Rather than generating vocabulary from scratch, you may also have existing materials you want to extract vocabulary from, e.g. an article, a text from a coursebook, or a document your learner brought from work. Below are some examples.

Extract 12 vocabulary items from the text below that are likely to be new or challenging for a B1 learner.
Order them in the sequence they appear in the text.
---
[paste the text here]

From the text below, list the 10 most difficult words for an A2-level ESL learner.
---
[paste the text]

You can also extract specific words or phrases. For instance words of a certain word class, idiomatic expressions, formal words that could be replaced with simpler alternatives, words that might be confused with similar-sounding words, or words with multiple meanings where the context-specific meaning matters. In the following we give some examples.

Extract all adjectives from the text below.
---
[paste the text]

Extract all formal words from the text below that can be replaced with simpler alternatives.
---
[paste the text]

Chatbots or the LLMs are generally good with classifying words into word classes or part-of-speech as it may be called. However, they may occasionally get it wrong. Try running the prompt again, in case it just has to do with the randomness of the chatbot. Otherwise, if it is because the chatbot doesn't understand the concept correctly see if you can describe it in another way.

Vocabulary list output

Chatbots may infer your overall intention with the prompts above. Perhaps from implicit implications like "vocabulary item" or from your chat history as a language teacher. However, as we haven't described what we want the chatbots are likely to give us something reasonable, but perhaps not what we actually need. Therefore, we may need to describe the output we want more explicitly and that is the focus in the following.You can append the prompt fragments in this section to the prompts in the previous section.

Vocabulary Information

When you ask AI for vocabulary, you can also request additional information to include per word. Such information may include: the word class, a definition, example sentences, words it frequently appears with, synonyms, notes on register or formality, or a translation into the learner's L1.

For each word, provide the following information:
- The word and its word class (noun, verb, adjective, etc.)
- A short definition
- An example sentence showing the word in a natural context
- 1-2 common collocations (words it frequently appears with)
- A synonym or near-synonym
- A note on register (formal, informal, neutral)

If you use the above prompt, add or remove items based on what's useful for your situation. If you want the information returned as a table you can try a prompt like the below.

I'd like the words presented in a table with the following columns:
Word, Word Class, Definition, Example Sentence, Collocations

You can also specify the format explicitly by an example for the AI to follow.

For each word provide an entry formatted like this example:

**affordable** (adjective) — not expensive; reasonably priced
"The restaurant is popular because the food is delicious and **affordable**."
Collocations: **affordable** housing, **affordable** price

I've used ** to indicate bold text. Many AI chatbots understand a simple markup language called Markdown and produces text in this and displays it formatted. If your chatbot doesn't understand it, doesn't produce it correctly, you may need to apply the formatting you like manually.

Beyond dictionary entries

You can ask for other information than what is found in dictionary entries. For instance, you may ask for a small paragraph of text to try to explain the words by examples.

For each word, write a short paragraph (2-3 sentences) that uses the word naturally and makes its meaning clear from context.

The AI may have not have a good grasp of how the meaning will be clear from context, but you may ask it to generate more so you have some to choose from, and you may be more specific about the paragraps to be simple.

For each word, write 3 short paragraphs (2-3 sentences) that each use the word naturally and makes its meaning clear from context.
Each paragraph must be easily understandable for an ESL A2 learner, so use simple words and constructs except for the word.

You may want to show word families if they exist to help your learners understand and recognize other forms of some vocabulary that they are learning.

For these words, show the word family: the noun, verb, adjective, and adverb forms where they exist. Format as a table.

Similarly, you can focus on how prefixes and suffixes change word meaning, or show various inflections of a word, e.g. past, present and future tense. We leave this as an exercise for you. Don't be afraid to experiment and figure out what works and what doesn't.

Vocabulary exercises

While you may want to provide glossary or vocabulary lists to introduce words to your learners, you may also want exercises to reinforce the learned vocabulary. This section covers generating vocabulary exercises. The words can come from any of the list prompts above, from your own word list, or you can let the AI choose words based on a topic and level. We show a few variations and leave it to you to adjust and/or combine the exact prompt you need.

Some chatbots are becoming so "smart" that they may end up presenting the result as interactive exercises for you to do. If that is not what you want then add something to indicate it to the chatbot, e.g. "Don't show these as interactive exercises, I need the result as a teacher."

Matching

Matching word to definition is a simple exercise type, and AI produce it well. Try something like the following.

Create a matching exercise with 10 items about food and restaurant vocabulary for a B1-level adult learner.
Left column: vocabulary words
Right column: definitions (shuffled, not in the same order as the words)
Include an answer key at the end.

The "shuffled" instruction matters. Without it, some AI models list definitions in the same order as the words. You can also match words to synonyms, translations, or picture descriptions instead of definitions.

Fill-in-the-blank

Fill-in-the-blank (or cloze) exercises require the learner to recall and produce a word in context. Try a prompt like the following.

Create a fill-in-the-blank exercise using these 8 vocabulary words:
commute, deadline, colleague, agenda, postpone, approve, department, overtime

Write 8 sentences about office life where the learner fills in the correct word.
Each sentence should have enough context clues that a B1 learner can figure out the right answer.

Learner version: sentences with blanks and a word bank at the top.
Teacher version: completed sentences with the answers in bold.

The instruction about "enough context clues" helps. Without it, AI sometimes produces sentences where multiple words could fit. You can also specify whether to include a word bank (easier) or leave it out (harder).

Multiple choice

Multiple choice can be used for lightweight easy recall, or for harder distinguishing between similar words. Try the below prompt for the latter case.

Create a multiple-choice vocabulary quiz with 8 questions for a B2-level learner.
Topic: phrasal verbs related to work and career.
Each question should be a sentence with a blank, followed by 4 options where only one is correct.
The distractors should be other phrasal verbs that are plausible but wrong in context.

Include an answer key with brief explanations of why the correct answer is right.

Asking for "plausible distractors" helps if you don't want the wrong answers to be obviously wrong and the right too easy to determine.

Word family exercises

You can turn word family lists into exercises by asking the AI to leave some cells blank for the learner to complete, like with the following prompt.

Create a word family exercise for B2-level English learners.
Choose 6 common root words and format as a table with columns for noun, verb, adjective, and adverb forms.
Fill in some cells and leave others blank for the learner to complete. Include a separate answer key.

Example row:
| Root | Noun | Verb | Adjective | Adverb |
|------|------|------|-----------|--------|
| create | creation | create | creative | creatively |
| employ | _______ | employ | _______ | — |

Here we have again use the formatting language Markdown to indicate an example table to the AI. The AI may be able to construct the table solely from the description "as a table with columns ..." and understand how to leave some cells blank, but if not consider adding the example to be more specific.

Other formats

We have shown some examples of exercises, which are common, but you can ask for others too. Odd-one-out exercises (which word doesn't belong?), categorization exercises (group these words by topic), or sentence-writing prompts (use each word in a sentence of your own) all work well with AI. The pattern is the same: specify the format, the level, the topic, and any constraints. Experiment and ask for what you need in a certain situation. 

L1 support for vocabulary

As covered earlier, the learner's native language can be a useful resource for vocabulary learning. For vocabulary lists, it can make sense to include L1 translations alongside the target language, especially for lower-level learners where L1 translations speed up initial word learning.

Create a vocabulary list for a Japanese-speaking A2 learner studying English.
Topic: daily routines.
12 items.

For each word, include the English word, the Japanese translation, and a simple English example sentence.
Use natural Japanese translations, not word-for-word equivalents.

If you don't speak the learner's L1, the caveats from Chapter 3 apply. Be transparent with the learner and let them flag anything that sounds off.

You can also generate exercises that specifically target the relationship between two languages, leveraging cognates (similar words with similar meanings) and addressing false friends (similar words with different meanings).

I teach English to Spanish-speaking B1 learners.
Create an exercise with two parts:

Part 1: "Good friends" — 8 English-Spanish cognates where the meaning is the same or very similar (e.g., telephone/teléfono).
For each, write an English sentence using the word.

Part 2: "False friends" — 6 English-Spanish false friends where the similar-looking word means something different (e.g., embarrassed ≠ embarazada).
For each, explain the actual meaning in both languages and write a sentence showing the correct English usage.

This kind of exercise turns a common source of errors into an explicit learning opportunity. 

Tips and pitfalls

• Check CEFR level accuracy. AI's sense of vocabulary difficulty is approximate. It's generally reasonable at B1-B2 but tends to overshoot at lower levels (giving B1 words when you asked for A2) and undershoot at higher levels (giving B2 words when you asked for C1). A quick scan will hopefully reveal any words that don't fit.

• Watch out for polysemy. Many common words have multiple meanings. "Run" can mean to move quickly, to manage a business, to operate software, or to flow. If you're generating vocabulary for a specific context, it can help to add "definitions as used in [context]" to your prompt so the AI picks the right meaning.

• Review answer keys. Vocabulary exercises can be prone to answer key errors. Fill-in-the-blank exercises sometimes have blanks where more than one word fits. Multiple choice exercises sometimes have two plausible answers. A quick check before sharing the exercise saves confusion later.