Man Pretends to Hallucinate in AI Job Interview, Sending the Bot Into a Tailspin

Applying for an office gig these days often means the first round isn’t a human at all, but a pixelated avatar powered by a large language model. HR departments worldwide have rolled out these bots to sift through the flood of applications that modern recruiting generates.

Enter Graham Zip, a satirical influencer who decided to see just how far he could push one of these digital interviewers. In a video that’s now gathering a fair amount of attention, he logs into a mock interview and is greeted by a virtual recruiter named “Dana Whitfield.” The bot’s opening line is the classic, “Tell me about yourself.”

Instead of delivering a polished career summary, Graham launches into a deliberately garbled spiel that sounds almost plausible—if you’re not paying close attention. “Hi Dana, it’s great to meet you. I’m Chick Haversplat, or Chick Bongo for short. And uh, Q3 pipeline, Salesforce, Kayak margin.” He tosses in a handful of buzzwords, then slips into pure nonsense.

The AI, designed to mirror back what it thinks it heard, immediately goes into its rehearsed script. “It’s a pleasure to meet you, Chick Bongo. So it sounds like you’ve been deep in the weeds with Salesforce, hygiene, and managing those Kayak margins during the third quarter.”

When prompted to expand on his background, Graham doubles down on the absurd: “freight, cadence, walnut, a little bit of Richard, sprucewood.” The bot, still trying to stay helpful, replies, “Six years of sprucewood—actually seven, if you count Richard.” The conversation quickly derails.

Graham keeps feeding the model bizarre fragments—“palette forward, a little bit of denim,” and even the cryptic “every pharmacist is a hallway.” The AI, now clearly out of its depth, responds with increasingly generic affirmations: “It sounds like you have a unique way of navigating the palette‑forward and denim aspects of your work.”

At one point Graham mentions he’s looking for a role with “two plumes of gauze” and wants to follow “the raccoon protocols.” The bot dutifully parrots back, “It sounds like you’re looking for a position where you can really lean into those raccoon protocols, and the plumes of gauze.” By the end of the skit, the virtual recruiter is summarizing a completely fabricated gap in a “drum counting process” involving a work truck and a cousin named Francis.

What’s striking isn’t just the comedy—it’s how effortlessly Graham commandeered the entire interview. The bot didn’t challenge the nonsense; it simply tried to keep the conversation moving, mirroring the very purpose of many AI recruiters: to screen, not to deeply understand. Whether Graham’s fake interview would have advanced him to a real hiring round remains unknown, but the bot’s cheery sign‑off—“It has been a pleasure chatting today, Chick Bongo… I will be in touch very soon”—suggests the system thought everything went perfectly.

The stunt underscores a growing concern among tech observers: LLM‑based hiring tools can be coaxed into “hallucinating” or producing nonsensical outputs when fed unconventional prompts. As companies continue to lean on AI to manage candidate flow, the need for robust safeguards and human oversight becomes ever more apparent.

In short, Graham’s performance is a reminder that, for all their polish, AI interview bots are still just sophisticated pattern‑matchers. Push them far enough outside their comfort zone, and they’ll gladly follow you down the rabbit hole.