Recently, the expression “You can’t lick a badger twice” suddenly went viral on social networks. The nonsense sentence– which was most likely never ever said by a human before recently– had actually ended up being the poster kid for the freshly found method Google search’s AI Overviews comprises plausible-sounding descriptions for fabricated idioms(though the principle appears to precede that particular viral post by a minimum of a couple of days ).

Google users rapidly found that typing any prepared expression into the search bar with the word “meaning” connected at the end would create an AI Overview with a supposed description of its idiomatic significance. Even the most ridiculous efforts at brand-new sayings led to a positive description from Google’s AI Overview, developed right there on the area.

In the wake of the “lick a badger” post, many users gathered to social networks to share Google’s AI analyses of their own fabricated idioms, frequently revealing scary or shock at Google’s take on their rubbish. Those posts frequently highlight the overconfident method the AI Overview frames its idiomatic descriptions and periodic issues with the design confabulating sources that do not exist.

After checking out through lots of openly shared examples of Google’s descriptions for phony idioms– and creating a few of my own– I’ve come away rather pleased with the design’s nearly poetic efforts to obtain significance from mumbo jumbo and make sense out of the ridiculous.

Speak to me like a kid

Let’s attempt an idea experiment: Say a kid asked you what the expression “you can’t lick a badger twice” methods. You ‘d most likely state you’ve never ever heard that specific expression or ask the kid where they heard it. You may state that you’re not acquainted with that expression or that it does not truly make good sense without more context.

Someone on Threads observed you can type any random sentence into Google, then include “significance”later on, and you’ll get an AI description of a well-known idiom or expression you simply comprised. Here is mine

[image or embed]

— Greg Jenner(@gregjenner. bsky.social)April 23, 2025 at 6:15 AM

Let’s state the kid continued and actually desired a description for what the expression implies. You ‘d do your finest to produce a plausible-sounding response. You ‘d browse your memory for possible undertones for the word “lick” and/or symbolic significance for the worthy badger to require the idiom into some form of sense. You ‘d reach back to other comparable idioms you understand to attempt to fit this brand-new, unknown expression into a larger pattern (anybody who has actually played the exceptional parlor game Wise and Otherwise may be acquainted with the procedure).

Google’s AI Overview does not go through precisely that type of human idea procedure when confronted with a comparable concern about the exact same stating. In its own method, the big language design likewise does its finest to produce a plausible-sounding reaction to an unreasonable demand.

As seen in Greg Jenner’s viral Bluesky post, Google’s AI Overview recommends that “you can’t lick a badger twice” methods that “you can’t trick or deceive someone a second time after they’ve been tricked once. It’s a warning that if someone has already been deceived, they are unlikely to fall for the same trick again.” As an effort to obtain significance from a useless expression– which was, after all, the user’s demand– that’s not half bad. Confronted with an expression that has no fundamental significance, the AI Overview still makes a good-faith effort to respond to the user’s demand and draw some possible description out of troll-worthy rubbish.

Contrary to the computer technology truism of “trash in, trash out, Google here is taking in some trash and spitting out … well, a practical analysis of trash, at the minimum.

Google’s AI Overview even enters into more information describing its idea procedure. “Lick” here suggests to “trick or deceive” somebody, it states, a little bit of a stretch from the dictionary meaning of lick as “comprehensively defeat,” Most likely close sufficient for an idiom (and a possible model of the idiom, “Fool me once shame on you, fool me twice, shame on me…”. Google likewise describes that the badger part of the expression “likely originates from the historical sport of badger baiting,” a practice I made certain Google was hallucinating till I looked it up and discovered it was genuine.

Zeus camouflaged himself as what

The worst examples of Google’s idiomatic AI uncertainty are ones where the LLM slips previous possible analyses and into large hallucination of entirely imaginary sources. The expression “a dog never dances before sunset,” Did not appear in the movie Before Sunriseno matter what Google states. “There are always two suns on Tuesday” does not appear in The Hitchhiker’s Guide to the Galaxy movie regardless of Google’s persistence.

Literally in the one I attempted.

[image or embed]

— Sarah Vaughan (@madamefelicie. bsky.social)April 23, 2025 at 7:52 AM

There’s likewise no indicator that the fabricated expression “Welsh men jump the rabbit” stemmed on the Welsh island of Portland, or that “peanut butter platform heels” describes a clinical experiment producing diamonds from the sticky treat. We’re likewise uninformed of any Greek misconception where Zeus disguises himself as a golden shower to describe the expression “beware what glitters in a golden shower.” (Update: As lots of commenters have actually mentioned, this last one is in fact a recommendation to the greek misconception of Danaë and the shower of gold, revealing Google’s AI understands more about this prospective significance than I do)

The truth that Google’s AI Overview provides these entirely fabricated sources with the exact same self-assurance as its abstract analyses is a huge part of the issue here. It’s likewise a relentless issue for LLMs that tend to comprise news sources and point out phony legal cases frequently. As normal, one ought to be really cautious when relying on anything an LLM provides as an unbiased reality.

When it concerns the more creative and symbolic analysis of rubbish expressions, however, I believe Google’s AI Overviews have actually gotten something of a bum rap just recently. Provided with the uphill struggle of describing nigh-unexplainable expressions, the design does its finest, creating analyses that can verge on the extensive sometimes. While the reliable tone of those actions can in some cases be irritating or actively deceptive, it’s at least entertaining to see the design’s finest efforts to handle our useless expressions.