Most vocabulary apps stop at “match word to translation.” Real fluency needs collocations, register, and spaced production inside actual sentences. The 15 prompts below turn a general AI chatbot into a tireless private tutor for new words in any major target language.
TL;DR
- Single-word memorization fails. Always ask for collocations and a sentence at your CEFR level — that is the single biggest fix in this guide.
- Use the right model: ChatGPT for spoken drills and idiom/context nuance, Claude for natural example sentences and writing correction, Gemini for harvesting words from screenshots, PDFs, or lecture video. See the table below.
- Pair AI with a spaced-repetition app (Anki). One controlled study found learners recalled about 79.8% of target words after 10 days of spaced retrieval tasks; AI just makes richer cards faster.
- Copy any of the 15 templates, swap the bracketed
[placeholders], and paste the output straight into Anki, Notion, or your notebook.
Who this is for
Self-study learners (CEFR A2–C1), tutors building practice sets, study-abroad and pre-trip preppers, and test candidates (IELTS, TOEFL, DELE, JLPT, HSK).
Skip these prompts if you are an absolute beginner (A0–A1) who still needs a textbook for grammar foundations, or if your target language is very low-resource — general models hallucinate words and collocations far more often outside the high-data languages (English, Spanish, French, German, Japanese, Chinese, Korean).
Which AI model for which vocabulary task (June 2026)
You can run every prompt below on any of the three big chatbots, but each has a clear strength. Pricing and limits below are as of June 2026.
| Task | Best pick | Why | Notable limit |
|---|---|---|---|
| Spoken drills, accent, role-play | ChatGPT (Advanced Voice) | Most natural voice tutor; switches accents and corrects pronunciation live | Free tier gets a short daily Advanced Voice window; Plus ($20/mo) is effectively unlimited voice |
| Natural example sentences, writing correction | Claude (Sonnet 4.6 / Opus 4.7) | Output reads least “translated”; strong at keeping your voice when editing | Free tier limited Sonnet 4.6; Pro $20/mo |
| Harvesting words from images, PDFs, video | Gemini 3.1 Pro | Best at reading screenshots, handwritten notes, and lecture video | Google AI Pro $19.99/mo (formerly “Gemini Advanced”); free tier exists |
| Idioms, Asian languages, contextual nuance | ChatGPT (GPT-5.5) | Strongest on idioms and context-heavy languages | Free tier ~10 GPT-5.5 messages per 5 hours, then auto-downgrades to a mini model |
Practical default: run text drills in whichever you already pay for; switch to ChatGPT Voice when you want to say the words, and to Gemini when your source material is a photo or PDF.
What every good vocabulary prompt includes
A vocab prompt works far better when it carries five things. Drop any one and output drifts toward generic dictionary entries.
- Target language and your level. State the language and a CEFR level (A2, B1, B2, C1). Without a level, models default to roughly B2 and you lose the easy reinforcement or the stretch words you actually need.
- Native language (L1). Needed for false-friends, pronunciation traps, and translation-interference checks.
- One concrete deliverable. “10 cloze sentences,” “5 collocations,” “a 6-line dialogue” — not “help me learn words.”
- Register tag. Ask the model to label each item formal / neutral / colloquial so you do not paste a courtroom phrase into casual chat.
- Output format. Numbered list or table so you can paste straight into Anki, Notion, or Word.
15 copy-ready prompt templates
Swap the [bracketed] placeholders before sending. These are written so you can run them on ChatGPT, Claude, or Gemini without changes.
1. Word in 5 collocations
The default vocab template. A collocation beats an isolated word every time.
You are a [target language] tutor. For the word "[word]", give 5 natural
collocations a native speaker would actually use. For each: the collocation,
an example sentence at [CEFR level], and a 1-line note on register
(formal / neutral / colloquial). End with one collocation a learner should
AVOID because it sounds translated.Swap: target language, word, CEFR level.
If output is too dictionary-like: add “Prefer collocations from real spoken conversation. Skip ones that only appear in writing.”
2. Cloze drill (10 items)
Generate 10 fill-in-the-blank sentences in [target language] at [CEFR level]
using these words: [word list]. Each sentence should have only one natural
answer. Put the answer key at the end. Use everyday topics.3. False-friends warning
For my native language [L1] and target language [L2], list 8 false friends
related to the topic [topic]. For each: the L1 word, the deceptive L2 word,
the actual meaning of the L2 word, and one example showing the typical mistake.4. Register switcher
For [target language], give 6 ways to express "[concept]" ranging from very
formal to very colloquial. For each: the phrase, a context where it fits (job
interview, with a friend, on a sign), and a context where it would sound wrong.5. Word-family expansion
For the root "[word]" in [target language], list its noun, verb, adjective, and
adverb forms. For each: 1 example sentence at [CEFR level] and 1 common
collocation. Mark any forms a learner can skip because they are rare.6. Topical word harvest
Give me 25 vocabulary items in [target language] for the topic [topic] at
[CEFR level]. Group by part of speech. For each: word, gloss in English, and
1 collocation. End with 3 items a learner often misses.7. Spaced-retrieval quiz
Below are 20 vocab items I learned 3 days ago. Build a mixed retrieval quiz:
5 production items (give context, ask for the word), 5 recognition items
(give word, ask for a paraphrase in [target language]), 5 cloze, 5 collocation
gaps. Answer key at end.
[paste word list]8. Mini-dialogue practice
Write a 6-line dialogue in [target language] at [CEFR level] between
[personas] that naturally uses these 6 words: [word list]. After the dialogue,
list each word with its line number and a 1-sentence explanation.9. Pronunciation focus
For these [N] [target language] words: [list], mark the difficult sound for an
[L1] speaker and give one minimal pair to drill it. Add 1 tongue-twister-style
sentence per word.Run this one in ChatGPT Advanced Voice and read each minimal pair aloud for live correction.
10. Idiom + literal trap
Give 6 common idioms in [target language] on the topic [topic]. For each:
idiom, literal English translation, actual meaning, and one situation where a
learner would misuse it.11. Word-of-the-day deep dive
Build a "word of the day" sheet for "[word]" in [target language]: etymology
in 1 sentence, 3 senses with examples, 4 collocations, 2 derived forms, 1
common mistake, 1 sample paragraph using the word naturally.12. Reading-text harvest
Below is a passage in [target language]. Pull out 10 words that would be useful
for a [CEFR level] learner to add to their active vocabulary. For each: the
word, the sentence it appeared in, a 1-line gloss, and 1 collocation.
[paste text]This is the highest-leverage template: it harvests words you have already seen in context, which beat random frequency lists for relevance.
13. Writing-correction with vocab upgrades
Below is my paragraph in [target language]. (1) Correct grammar in line.
(2) Suggest 5 vocabulary upgrades where I used a basic word; show the upgrade
plus 1 alternative collocation. Keep my voice.
[paste paragraph]Claude tends to keep your original voice best on this one.
14. Image / scene description drill
Describe a scene that requires using the words [word list] in [target language]
at [CEFR level]. After describing, give me 5 prompts to describe my own version
of the scene using as many of those words as possible.15. Weekly review report
Based on these 30 words I studied this week: [paste], give me a 1-page review:
5 words I likely already know, 10 I should drill more, 10 collocations I am
still missing, 5 ready-to-use sentences combining them, and one mini-essay
topic that would force me to use 15 of them.Turn AI output into Anki cards
General chatbots are great at generating rich material but will not remind you to review it. Hand the scheduling to a spaced-repetition app. A fast loop:
- Run template 1 or 12 to get words plus a collocation and a sentence each.
- Add “Output as a CSV with three columns: front (word + collocation), back (sentence + register), tag.” to the prompt.
- Import the CSV into Anki and drill on the standard schedule.
The evidence is solid: spacing reviews over days beats cramming, and one classroom study reported about 79.8% recall of target words after a 10-day spaced-retrieval routine. AI does not replace the schedule — it fills your deck with collocations and sentences instead of bare translations.
Common mistakes
- Memorizing single words without collocations — you will sound translated.
- Ignoring register; formal vocabulary in casual conversation sounds stiff.
- Skipping production drills — recognition is far easier than recall, so test both.
- Letting AI invent rare words; cross-check unusual outputs against a dictionary, especially outside the high-data languages.
- Locking to one CEFR level — a B1 learner needs occasional A2 reinforcement and B2 stretch.
- Reviewing everything on the same day; spaced retrieval (roughly 3–5–7 day intervals) beats a single cram.
FAQ
- How many words per week can I really learn? Around 15–25 active words is realistic for a working adult, 40+ for an immersion student. Pushing past that hurts retention more than it adds coverage.
- Which AI is best for vocabulary? ChatGPT for spoken practice and idioms, Claude for natural sentences and writing correction, Gemini for pulling words out of images and PDFs. Most learners pick one paid plan ($20/mo tier) and switch tools only for voice or image tasks.
- Is AI accurate for any language? Strong in high-data languages (English, Spanish, French, German, Japanese, Chinese, Korean). Weaker in low-resource languages, where it hallucinates words and collocations — verify against a dictionary.
- Should I use AI instead of Anki? Use both. AI generates richer cards (collocations plus sentences); Anki runs the spaced-repetition schedule.
- How long until a word feels active? A word usually needs roughly 8–15 spaced retrievals plus 2–3 production attempts before it feels active. AI lets you cycle through those reps faster.