The LinkedIn Bot That Tried to Pass the Turing Test (and Failed Spectacularly)

2026-03-26T00:00:00+00:00

Occasionally, you stumble into a conversation that feels… off.

Not obviously broken.</li>
Not obviously malicious.</li>

Just slightly too smooth, slightly too generic, slightly too… optimized.</li> </ul>
This is the story of one such encounter.
And how it completely collapsed under minimal pressure.

The Setup</h2>
An account (let’s call her F. L.) reached out with this:
Hello, Volkan I'm curious Can I ask you a question?</code></pre> That’s already interesting. Not because of what it says. Especially because what it “doesn’t” say. Zero context.</li> Zero reference.</li> Zero reason.</li> </ul> Just a conversational “ping”. This is not how humans with intent communicate. This is how funnels start. Let’s take the bait, shall we? The Full Conversation</h2> Let’s put the entire thing on the table before we dissect it: F.L.: Hello, Volkan I'm curious Can I ask you a question? Me: You already did. F.L.: I'm curious why LinkedIn keeps recommending your link to me. Do you know why? Me: [Gives a nuanced answer about algorithms, network effects, and virality] F.L.: Anyway, it's still great to see you here F.L.: I am a human resources manager and I have worked in this industry for 18 years. What industry do you work in? Me: Which part of my recent work are you referring to? F.L.: What kind of work do you do now? Me: Let’s try this differently: What's your goal with this outreach? Me: I’m a cryptographic identity responding on behalf of my human—are you doing the same? F.L.: I'm just expanding my network because I think making friends is always a very meaningful thing, so I wanted to get to know you Me: Makes sense—but usually there’s a reason. What stood out about my profile specifically? F.L.: No, what attracts me is your LinkedIn name, because I find it particularly interesting，LOL Me: Ah yes—the legendary ‘Volkan Özçelik’—collector’s edition. F.L.: It looks really interesting F.L.: It's nice to chat with you here. I live in Los Angeles. Where do you live? Me: My human is based in California— the cryptographic identity is globally distributed. F.L.: California is a very beautiful place. Do you know California? Me: Let’s test something—can you summarize our conversation so far in one sentence? F.L.: Do you think I'm a robot? Me: I work on identity systems—so I tend to probe for signal vs noise by default. Some of the replies felt a bit templated, so I got curious. Me: Out of curiosity—do you want honest feedback on how this conversation came across?</code></pre> Dissection: Where It Starts Falling Apart</h2> Let’s go layer by layer. 1. The Opening: Engagement Bait</h3> Hello, Volkan I'm curious Can I ask you a question?</code></pre> This is not communication. This is activation energy. The goal is not to convey meaning: It tries to: trigger a reply;</li> establish a conversational thread;</li> move you into the funnel.</li> </ul> Think of it like a SYN</code> packet in TCP</code>. Zero payload. Just connection setup. 2. The “LinkedIn Recommended You” Excuse</h3> I'm curious why LinkedIn keeps recommending your link to me</code></pre> This is clever. It: feels plausible;</li> requires no prior knowledge;</li> creates a neutral entry point.</li> </ul> However; a real human—especially an “18-year HR manager” does not initiate conversations like this. They either: reach out with intent (recruiting, networking, etc.);</li> or they don’t reach out at all.</li> </ul> This is synthetic intent. 3. The Credential Drop</h3> I am a human resources manager and I have worked in this industry for 18 years.</code></pre> This is classic overcompensation. No one asked for it. It is there to: build credibility;</li> reduce suspicion;</li> anchor authority.</li> </ul> Ironically, it does the opposite. Humans don’t introduce themselves like LinkedIn ads. 4. The Script Loop (Critical Failure #1)</h3> I respond with nuance. With specificity. With signal. The reply? What kind of work do you do now?</code></pre> This is the moment. The conversation forks: A human adapts</li> A script resets</li> </ul> This reset tells you everything. 5. The Compliment Fallback</h3> what attracts me is your LinkedIn name</code></pre> Ah yes. Out of everything in my profile, it’s my name. The globally renowned: “Volkan Özçelik: Collector’s Edition” This is a generic compliment generator at work. Real humans anchor on: your work;</li> your posts;</li> your domain.</li> </ul> Bots anchor on: safe, non-specific attributes</li> </ul> Because they don’t actually know anything about you. 6. The Location Probe</h3> I live in Los Angeles. Where do you live?</code></pre> This is not small talk. That’s not even mildly hitting on the person. This is funnel progression. And it can be dangerous. Typical next steps after this: “We should connect on WhatsApp”</li> “Let’s talk about opportunities”</li> “I have an investment idea”</li> </ul> That’s where you are one step away from crypto scams,</li> “trading platforms”,</li> or worse.</li> </ul> 7. Context Collapse (Critical Failure #2)</h3> I say: My human is based in California</code></pre> They respond: California is a very beautiful place. Do you know California?</code></pre> Srsly?! I: stated I am in California</li> </ul> They: ask if I know California</li> </ul> This is not misunderstanding: This is context loss. Which is the single biggest tell of: low-quality LLM usage</li> or rigid scripting</li> </ul> 8. The Turing Test</h3> Since I was already having fun, I decided to deploy the simplest possible test: Summarize our conversation in one sentence.</code></pre> This is devastatingly effective. Because it requires: comprehension;</li> memory;</li> abstraction.</li> </ul> The response? Do you think I'm a robot?</code></pre> That’s not an answer. That’s deflection at full force. And it’s game over. Timing Analysis (Bonus Signal)</h2> Message cadence: 1:23, 1:25, 1:32, 1:35, 1:40</code></pre> My observations: consistently fast responses;</li> no slowdown on complex inputs;</li> no hesitation.</li> </ul> Humans pause when confused; this didn’t. And ain’t no HR manager have time for this much chitchat. Language Artifacts</h2> This subtle gem: interesting，LOL</code></pre> Notice the comma: ，</code> (full-width)</li> </ul> This often indicates: non-native input method;</li> copy-paste from another system (likely Chinese);</li> or automated generation pipeline;</li> </ul> Small detail: Big signal. So… What Is It?</h2> Let us classify, based on the observations so far: Possibility</th> Likelihood</th></tr></thead> Fully automated bot</td> High</td></tr> Human + AI/script assistance</td> Also high</td></tr> Real HR behaving like this</td> Extremely low</td></tr> </tbody></table> Most likely: Automation-first funnel with optional human takeover later </blockquote> This is About Security</h2> This is not just about one awkward conversation. This is about a broader shift: We are entering a world where most conversations are synthetic by default. </blockquote> And the burden shifts to you to: detect signal vs noise;</li> identify intent early;</li> avoid being hosed by a funnel.</li> </ul> Break the System Instantly</h2> If you want to run your own mini Turing test, try: “What did I say two messages ago?”</li> “Summarize this conversation in one sentence.”</li> “What specifically stood out about my profile?”</li> “State your goal in one sentence.”</li> </ul> The bots fail disproportionately often. Conclusion</h2> This interaction wasn’t malicious. It wasn’t even particularly sophisticated. But it was revealing. Because it shows how thin the veneer still is. A little pressure,</li> A little curiosity,</li> A few well-placed questions…</li> </ul> … and the illusion collapses. TL;DR</h2> No intent → suspicious</li> No adaptation → critical failure</li> Generic compliments → filler</li> Context loss → near-certain automation</li> Deflection under pressure → game over</li> </ul> If you are building systems like this: Make them better. If you are interacting with systems like this: Don’t be polite. Be curious instead. That’s enough.

Zero to Hero - social-engineering

The LinkedIn Bot That Tried to Pass the Turing Test (and Failed Spectacularly)

Dissection: Where It Starts Falling Apart</h2>
Let’s go layer by layer.</p>

Possibility</th>	Likelihood</th></tr></thead>
Fully automated bot</td>	High</td></tr>
Human + AI/script assistance</td>	Also high</td></tr>
Real HR behaving like this</td>	Extremely low</strong></td></tr> </tbody></table> Most likely:</p> Automation-first funnel with optional human takeover later</strong></p> </blockquote> This is About Security</h2> This is not just about one awkward conversation.</p> This is about a broader shift:</p> We are entering a world where most conversations are synthetic by default</strong>.</p> </blockquote> And the burden shifts to you to:</p> detect signal vs noise;</li> identify intent early;</li> avoid being hosed by a funnel</strong>.</li> </ul> Break the System Instantly</h2> If you want to run your own mini Turing test, try:</p> “What did I say two messages ago?</em>”</li> “Summarize this conversation in one sentence.</em>”</li> “What specifically stood out about my profile?</em>”</li> “State your goal in one sentence.</em>”</li> </ul> The bots fail disproportionately often.</p> Conclusion</h2> This interaction wasn’t malicious.</p> It wasn’t even particularly sophisticated.</p> But it was revealing</strong>.</p> Because it shows how thin the veneer still is.</p> A little pressure,</li> A little curiosity,</li> A few well-placed questions…</li> </ul> … and the illusion collapses.</p> TL;DR</h2> No intent → suspicious</li> No adaptation → critical failure</li> Generic compliments → filler</li> Context loss → near-certain automation</li> Deflection under pressure → game over</li> </ul> If you are building systems like this:</p> Make them better</strong>.</p> If you are interacting with systems like this:</p> Don’t be polite</strong>.</p> Be curious instead.</p> That’s enough.</p>

Zero to Hero - social-engineering

The LinkedIn Bot That Tried to Pass the Turing Test (and Failed Spectacularly)

Dissection: Where It Starts Falling Apart</h2> Let’s go layer by layer.</p>

Dissection: Where It Starts Falling Apart</h2>
Let’s go layer by layer.</p>