Test of AI's to suggest mnemonics

Ever since ChatGPT became available in late 2022, creating a mnemonic is no longer exclusively a human effort. Tricks people have been using for centuries still work. But when no good mnemonic comes out of our brain, we now have one more tool to try and check.

The French word arnaque means "scam" or "rip-off". Other than by rote memory (with flash cards, e.g.), there's no good way to remember this word, which means "scam", "rip-off". Etymologically, it probably comes from harnais (“to harness”) in its informal sense of “to dress up ridiculously”. But that requires digging out quite a bit of history to find out why that led to the sense of "scam", not to mention the obscure sense of "dress up ridiculously". Time to think up a mnemonic.

So I ask 9 AI tools (10 if counting Google twice), "Give a mnemonic to help remember the French word 'arnaque'." The following are their answers, abbreviated:

Chatgpt.com: "A Real Nasty Act, Quite Underhanded and Evil"
DeepSeek.com: "The ARtful NAKed con artist scammed me."
Perplexity.ai: "Arnie’s knack is tricking people"
Mistral.ai: "An Ape Never Aquires Quality Electronics"
Copilot.microsoft.com: "A Really Nasty Act Quickly Uncovered Eventually"
Claude.ai: "Arnold's HACK"
Grok.com or X.ai: "A Rude Nasty Act, Quite Unfair, Exploits."
Qwen.ai: "A rat never asks quietly, unless it's a scam!"
Gemini.google.com: "A really nice antiquarian quickly understands exactly what's a scam", "All Roberts Need A Quick Un-doing of Everything. Arnaque means scam."
Google.com (directly, not through Gemini): "AR NO-AQUe (no water)"

We can see that many try to break up the word into single letters, much like those mnemonics school children use to remember the names of the planets in the solar system ("My Very Educated Mother Just Served Us Nachos"), or the order of the biological classification ("Dear King Phillip Came Over For Good Soup"). But honestly, this type of mnemonics are not the best to help remember a single word in a human language, because it's less effective if you completely break up the word. It's better to work on chunks of letters of the word. DeepSeek suggests "ar-" and -nak-", Perplexity "ar-" and "knack", Claude "Arn-" and "-ack", and Google "ar" and "n...-aqu-". After reviewing all these, I'd like to suggest a combination of DeepSeek's and Perplexity's mnemonics, e.g. "artful knack" (and reject "naked" as it's semantically far from "scam", and "Arnie" as it's not a common name). Claude's "Arnold's hack" is a good one, too, but "h" of "hack" somewhat blocks a mental association of "n" of "Arnold" with "a" of "hack". On the other hand, if the English word "knack" is not in your active vocabulary, "artful knack" may be inferior to "Arnold's hack".

Other than completely breaking up a word into single letters, AI's seem to have another tendency compared to human effort: it rarely incorporates the ending of a mnemonic word. Take as an example Spanish bruja, meaning "witch". I created the mnemonic "broom ninja", which vividly depicts a scene in which a witch sits across a broom and moves or possibly flies like a ninja. AI can come up with "broom" but will be hard pressed to think of "ninja" because "-ja" is the ending of a word.

AI has drastically improved our ability to do lots of things. This posting may be the first to explore and discuss its capability in generating mnemonics for foreign language vocabulary studies. Future work may be to fine tune a prompt so that AI can generate better mnemonics suitable to your own need.