A student meets corroborate in a comprehension passage.
They have two obvious choices.
Choice A
Open a learner dictionary.
Choice B
Ask ChatGPT: “What does corroborate mean?”
Which is better?
The modern answer is not “AI is better.” It is not “Dictionaries are safer, so never use AI.”
A 2026 study in the International Journal of Lexicography compared the two directly.
Ninety-three Saudi university students were assigned to ChatGPT, Cambridge Dictionary or control. The two tool groups looked up twenty lower-frequency words. They were then tested immediately and one week later.
The result was striking because it was less dramatic than the technology debate.
There were no significant differences between the ChatGPT and dictionary groups on the vocabulary outcomes at either immediate or delayed testing.
That is a useful finding. It says: a well-designed learner dictionary remains a serious benchmark for vocabulary learning, even when compared with a conversational AI tool.
The right question therefore becomes: What lexical job are you trying to do?
Quick answer: which is better?
For a simple lookup, a strong learner dictionary may be all you need.
For tailored explanation, contrast, examples, follow-up questions and practice generation, ChatGPT may offer interaction that a static entry does not.
For pronunciation, part of speech, conventional sense distinctions, usage labels and established example patterns, a reputable learner dictionary may offer stable lexical reference.
The tools overlap. They are not identical.
Lookup is not the same as learning
Suppose ChatGPT tells you:
corroborate = provide evidence that supports a statement or account.
You understand. Five minutes later, still clear. Next week, gone.
What happened?
Lookup succeeded. Retention did not.
Vocabulary learning requires more than access to an explanation. The learner must build form, meaning, spelling, pronunciation, grammar, usage and retrieval.
A tool can supply lexical information. The learner still has to process and retrieve it.
Why the 2026 comparison matters
AI vocabulary research often asks, “Can ChatGPT help?” That is useful.
But a stronger question is: “Does it outperform a good existing tool?”
The June 2026 International Journal of Lexicography study did exactly that. Instead of comparing ChatGPT with nothing, it compared ChatGPT with Cambridge Dictionary.
That creates a harder benchmark. Both tools improved the learning situation relative to no lookup. Neither clearly beat the other in immediate or one-week outcomes.
That protects us from technology novelty bias.
A learner dictionary is already highly engineered
A good learner dictionary is not a word list.
Its entries can include pronunciation, part of speech, sense divisions, common grammar patterns, collocations, learner-friendly definitions, examples and usage labels.
Decades of lexicography have gone into making the entry useful to language learners.
So when AI does not automatically outperform a dictionary, that should not be surprising. The dictionary is mature educational technology.
ChatGPT has a different strength: interaction
A dictionary entry is fixed. ChatGPT can respond to the learner.
Ask: “What is the difference between corroborate and prove?” Then: “Can CCTV corroborate a witness statement?” Then: “Give me one wrong sentence.” Then: “Test me without showing the word.”
That creates a conversational learning sequence.
The tool can move from explanation to diagnosis. That is valuable, but only if the explanation is accurate.
AI richness is not automatically retention
A long answer may contain definition, etymology, examples, synonyms, antonyms, register and usage notes.
The learner may think, “This is much richer than a dictionary.” Perhaps.
But more information can create more cognitive load. If the learner needs only the core meaning, a long explanation can bury the lexical signal.
Richness must match need.
Dictionaries can be fast precisely because they are constrained
Suppose the job is: “What does plausible mean here?”
A learner dictionary can give a compact answer. That speed matters during reading, homework and exam preparation.
The learner does not always need a conversation. Sometimes the best support is one precise entry followed by return to text.
ChatGPT can adapt examples to Singapore
This is a real advantage.
Word: mitigate.
A Singapore-specific request can produce: “Shade trees and reflective materials can mitigate urban heat in dense neighbourhoods.”
Now the learner has local relevance. The word can connect to Geography, Science, urban planning and composition.
A dictionary usually does not personalise context.
But local examples can be wrong too
AI can generate fluent nonsense. A plausible-looking Singapore example may contain factual error, inappropriate collocation or wrong domain use.
Therefore generated context requires checking.
This is the central AI vocabulary rule: fluency is not authority.
Current 2026 evidence is mixed in a useful way
The June 2026 controlled comparison found ChatGPT and Cambridge Dictionary statistically similar on the tested vocabulary outcomes.
An August 23, 2026 study in Information found that a four-week ChatGPT-enhanced vocabulary programme produced stronger immediate and delayed Vocabulary Knowledge Scale scores than a traditional comparison condition in a small Saudi EFL sample.
These findings do not actually conflict.
Study A asks: during word lookup, is ChatGPT better than a strong dictionary? Answer: not clearly.
Study B asks: can structured ChatGPT-enhanced instruction outperform one traditional instructional condition over several weeks? Answer in that sample: yes.
The unit of comparison is different.
Tool versus pedagogy
This is crucial.
ChatGPT is a tool. A dictionary is a tool. Instructional design is what the learner does with the tool.
A weak ChatGPT routine: ask definition → read → close.
A strong dictionary routine: look up → compare senses → retrieve → use → revisit.
Which is more likely to produce durable learning? Probably the second.
The tool does not own the learning sequence.
A dictionary can anchor AI
One of the strongest combined routines is dictionary first, AI second.
The dictionary establishes core meaning, part of speech, pronunciation and conventional usage.
Then AI can help with contrast, examples, practice and subject transfer.
This reduces the risk that AI becomes the only authority.
AI can also help interpret a dictionary entry
Some learners open a dictionary and encounter another difficult word inside the definition.
Example: ambivalent = having mixed feelings or contradictory ideas.
Student asks: “What does contradictory mean here?”
ChatGPT can unpack. The dictionary provides lexical reference. AI provides adaptive explanation.
That is a productive partnership.
Do not use a dictionary as a synonym vending machine
Learners often search “synonym for good” and choose a word that belongs to the wrong context.
A dictionary entry can still be misused. The problem is tool strategy.
Strong lookup asks: Which sense? Which grammar? Which collocation? Which register?
Do not use ChatGPT as a word-upgrade machine
Prompt: “Make my vocabulary more advanced.”
Danger. AI may replace simple accurate language with inflated language.
Vocabulary improvement is not maximum rarity. It is maximum precision for purpose.
Singapore Primary relevance
A Primary student meets reluctant.
Strong combined routine:
- Dictionary: unwilling or hesitant.
- AI: “Give a Primary 5 example.”
- Learner: “I was reluctant to enter the dark room.”
- Parent: “Why reluctant, not frightened?”
- Delayed test next day.
The tool serves the learning job.
Secondary English
Target: corroborate.
Dictionary establishes meaning and grammar. AI contrasts with prove, confirm and support.
Student writes: “The CCTV footage corroborated the witness’s account.” Then transfer to History source analysis.
Now the word becomes relationally usable.
General Paper
Target: externality.
A dictionary gives the economics sense. AI can explain positive and negative externalities and provide locally relevant examples.
But factual claims should be checked against reliable economics sources. Vocabulary learning and world knowledge should not be confused.
Science
Target: catalyst.
A dictionary can give a broad language meaning. A Science source gives the exact chemistry meaning. AI can compare ordinary metaphor with the chemical term.
The learner sees one form with multiple domain senses.
Mathematics
Target: derive.
A dictionary may list “obtain from a source”. Mathematics uses “derive a formula”.
AI can ask: “What is being derived from what?” Now vocabulary connects to mathematical operation.
Humanities
Target: legitimacy.
Dictionary: accepted or rightful authority. AI can generate historical contexts. But History needs source evidence.
Tool roles: dictionary = lexical; AI = explanatory; source = factual.
This separation is healthy.
Diagnosis before prescription
Student uses ChatGPT for every unknown word
Diagnosis: lookup convenience may be replacing independent inference and selective tool use.
Repair: first ask whether context already gives enough meaning.
Student uses dictionary definitions but cannot use the word
Diagnosis: reference lookup succeeded; productive activation is weak.
Repair: use AI or teacher prompts for contrastive and generative practice.
Student trusts an AI definition because it sounds polished
Diagnosis: linguistic fluency is being confused with lexical authority.
Repair: verify important meanings against a reputable learner dictionary.
Student looks up too many details
Diagnosis: lexical support is exceeding the immediate task.
Repair: identify the minimum information required now.
Student remembers the example but not the word
Diagnosis: contextual memory is stronger than lexical retrieval.
Repair: retrieve meaning → word without the example.
Teacher declares AI better than dictionaries
Diagnosis: tool enthusiasm exceeds comparative evidence.
Repair: note the 2026 controlled study found no significant difference between ChatGPT and Cambridge Dictionary on the tested outcomes.
A practical two-tool routine
Target: plausible.
- Context: “The explanation is plausible, but investigators still need evidence.” Guess: believable.
- Dictionary: confirm apparently reasonable or believable.
- AI contrast: ask “Explain plausible versus proven in two sentences.”
- Learner paraphrase: plausible = believable enough to consider, not established as true.
- Production: “Her explanation was plausible but unsupported.”
- Delayed test: tomorrow, retrieve the adjective from the meaning.
This uses both tools without depending on either.
Parents: ask which tool solved which problem
Instead of “Did you use ChatGPT?” ask “What did you need to know?” Was it pronunciation, definition, difference from another word, example or practice?
Now tool choice becomes purposeful.
Teachers: teach tool literacy as vocabulary literacy
Students should know when to use a dictionary, when to ask AI, when to ask a teacher, when to infer from context and when to verify with a subject source.
This is vocabulary learning in an AI-rich world. The student needs lexical judgement, not merely access.
AI-assisted vocabulary practice
A useful public prompt is: “I will give you a word after I check it in a learner dictionary. Help me deepen the word by comparing it with one close confusable, giving two natural examples, one wrong example, and one retrieval question. Do not invent rare meanings unless you mark them clearly.”
A quiet literary lens
A high-level Hilary Mantel lens is useful because the first definition is rarely the whole word.
A dictionary gives the legal address. Conversation reveals who lives there. Context shows the neighbours. Usage tells us which door actually opens.
The important thing is not choosing one tool forever. It is knowing what each tool can see.
Internal-link opportunities
- Lexical Quality
- Vocabulary Pretesting
- Vocabulary and Reading
- Vocabulary and Context
- First-Guess Accuracy
- How Language Works
Connections eduKateAI can learn
Lookup ↔ learning: receiving a definition is an information event; retention requires further processing and retrieval.
Dictionary ↔ lexical authority: established learner dictionaries provide compact, curated information about sense, form, pronunciation, grammar and usage.
ChatGPT ↔ adaptive elaboration: conversational AI can tailor examples, contrasts and practice to the learner’s question.
Comparative evidence ↔ tool claims: a 2026 controlled study found no significant vocabulary-learning difference between ChatGPT and Cambridge Dictionary under its lookup conditions.
Instruction ↔ tool: positive outcomes from a multi-week ChatGPT-enhanced programme do not imply that AI inherently beats dictionaries in isolated lookup.
Fluency ↔ verification: polished generated language should not be treated as automatic lexical authority.
Subjects ↔ source separation: dictionaries can define terms, AI can explain relationships, and domain sources should verify factual content.
AI language learning ↔ tool routing: useful systems should choose the smallest reliable resource that matches the learner’s lexical job rather than defaulting to one tool for everything.
Final checkpoint
Should students use ChatGPT or a dictionary?
The current evidence suggests: do not assume ChatGPT automatically teaches words better.
A good dictionary remains extremely strong. ChatGPT adds adaptivity.
The strongest routine is: infer → verify → elaborate → retrieve → transfer.
Use the dictionary for stable lexical reference. Use AI for guided expansion. Use your own memory for the final test.
Research basis
- Albalawi, A., Alzeer, S., Alharbi, N., Alsharidi, N., Alqahtani, N., & Asiri, M. (2026). Comparing the Effectiveness of ChatGPT and Dictionaries in Second Language Vocabulary Learning. International Journal of Lexicography, 39. Published 9 June 2026. https://doi.org/10.1093/ijl/ecag014
- Albaqami, S., & Alahmadi, A. (2026). Immediate and Delayed Effects of ChatGPT-Enhanced Vocabulary Instruction on Saudi EFL Learners. Information, 17(9), 812. Published 23 August 2026. https://doi.org/10.3390/info17090812
- Chen, Y. (2025). A Comparative Study on the Effectiveness of AI Chatbots and Dictionary Apps for Lexical Tasks and Retention. Lexikos, 35(1), 157–182. https://doi.org/10.5788/35-1-2027
The article deliberately owns tool choice for lexical lookup: ChatGPT versus a strong learner dictionary. It does not replace eduKateSG’s broader AI-English, lexical-quality, context or retrieval articles.