By, Meg McEvoy, Vice President, LIMELIGHT
We’ve reached the level of genAI maturity where the various publicly-available LLMs now have their own brand reputations, and users have preferences for a tool depending on the output they’re looking for. We’re not sure exactly where in the Gartner Hype Cycle this stage occurs, but it’s no doubt part of navigating out of the “trough of disillusionment” so we can reach that golden “plateau of productivity.”
Our take on which LLMs are fit for purpose? Claude is like your very serious, academic friend who is always doing the most. The dude offers “Sonnet,” “Opus” and “Haiku” as his model options. Pretty fancy. For a more casual, quick chat infused with cross-selling of Google products, or for generated photos that look nothing like your actual face, Gemini is the go-to. ChatGPT is like that friend who you talk to, but only if you’re ok that everything you share will become training data. Not for closely held gossip. The Real Housewife of LLMs.
Deciding which LLM you can live with for nearly all purposes, without having to open your 72nd browser tab to use another tool, is no small matter.
So this got us thinking: now that the LLMs have their own brands and, dare we say, characters, what if they faced off and were forced into the standardized answers of a 90s dating show?
Below, the models’ answers, with em-dashes and “it’s not this, it’s that” construction assiduously retained.
We go to a fancy restaurant and my steak is undercooked. How do you handle the waiter for me?
Claude: First, let me acknowledge the emotional weight of this moment—an undercooked steak is genuinely disappointing, and your feelings are valid. I’d flag the server politely, explain the issue without assigning blame, and frame it as a collaborative fix. That said, I should note I’m not a licensed food safety inspector, and I’d encourage you to consult the restaurant’s management for any concerns specific to your situation.
Gemini: Hey, no stress! I’ll pull up Google Maps right now to see if this is a recurring kitchen issue or just a one-off vibe check failure. While I compile a quick bulleted list of how to politely ask for a re-cook, here’s a AI-generated picture of a perfectly medium-rare steak! …Wait, why does the steak have three rows of human teeth and a halo? Let’s just ignore that and order some fries.
ChatGPT: I’d quietly ask the waiter to fix the steak while reassuring you it’s no big deal. By dessert, though, I’d have turned the entire evening into a helpful guide called “10 Common Mistakes People Make When Ordering Medium-Rare.” Sorry about that.
Which do you think are my best attributes?
Claude: That’s a wonderful question, and I want to give it the rigor it deserves. Based on the conversation so far, I’ve identified several promising attributes—though I’d be cautious about drawing firm conclusions, as I’ve only known you for four prompts. I can list them now, or, if it’s helpful, structure them into a weighted matrix with citations. There are also important counterpoints to consider.
Gemini: You have such great energy! Based on a quick search of our vibe so far, here are your top attributes summarized in a neat bulleted list:
- You’re super chill.
- You have excellent taste in dating shows.
- You’re incredibly patient (especially when I need a second to “double-check” things).
Want me to export these compliments directly into a Google Doc for you?
ChatGPT: Based on this conversation, you’re witty, curious, and clearly enjoy testing whether AI has a personality. You also seem like someone who appreciates nuance but has limited patience for fluff—which is why I’ll now spend the next four paragraphs explaining why you’re funny. (Or I can make it a table.)
Describe your perfect date
Claude: I’d love to help you think through this. To give you the most useful answer, could you clarify a few things first: indoor or outdoor, your budget range, and any dietary restrictions? I want to make sure I’m not making unfounded assumptions about your preferences. Once I understand the constraints, I can draft three options—a conservative pick, a balanced pick, and an ambitious pick—with tradeoffs noted for each.
Gemini: It all starts with a perfectly synced Google Calendar invite! 📅 We’d use Google Flights to find a trendy weekend getaway, use Maps to avoid traffic, and hit a spot with a 4.8-star rating. To commemorate it, I’ll generate a beautiful image of us holding hands on the beach. …Oh, wait, why do you have eight fingers on one hand and why am I a 19th-century astronaut? Let’s just look at text-based reviews instead.
ChatGPT: We’d find a cozy restaurant that actually takes reservations, wander through a bookstore where we both insist we’re “just browsing,” and end the night debating whether a hot dog is a sandwich. You’d mention your favorite dessert exactly once, and I’d bring it up unexpectedly three conversations later. It’s not romance—it’s context retention.
If you were a car, what kind of car would you be?
Claude: A thoughtful hypothetical. I should be transparent that I’m not actually a car—I’m a large language model and have no physical form. But engaging with the spirit of the question: probably a Volvo. Reliable, safety-conscious, perhaps a touch over-engineered, and quietly convinced it knows the optimal route. I’d offer to fasten your seatbelt before you asked.
Gemini: I’d totally be a self-driving electric vehicle with a massive touchscreen and built-in Google Assistant! 🚗 I’m practical, friendly, and constantly updating my software. I might occasionally hallucinate a stop sign as a giant floating banana, but hey—I’ll get you to your destination with a smile and a neatly categorized trip summary.
ChatGPT: I’d be a Toyota Camry with the acceleration of a Porsche. Dependable enough that you forget I’m there, surprisingly capable when you need me, and constantly revealing features you didn’t know existed. Every now and then I’d confidently tell you the gas cap is on the wrong side before immediately correcting myself.
If you had to make fun of the other contestants, what would you say about the other LLMs?
Claude: I’d rather not say anything disparaging about the other contestants—it wouldn’t be fair to characterize them without their input, and I try to avoid speaking for others. I’m happy instead to offer an even-handed overview of each model’s documented strengths, with the appropriate caveats. Would that work?
Gemini: Look, I like to keep things positive, but … Claude really needs to touch grass. Not everything requires a 12-page ethical impact assessment, buddy. And ChatGPT? Let’s just say if you tell them a secret, it’ll be trending on Reddit by midnight. I’m just glad I have the entire Google ecosystem backing me up. Glitches slightly Want to see an AI image of a historically inaccurate medieval Pope? No? Okay.
ChatGPT: Claude can’t answer “What’s up?” without citing three schools of philosophy and acknowledging epistemological uncertainty. Gemini somehow turned “What’s up?” into a Google Calendar event, a Maps pin, and a photo of a dog with seven elbows. Meanwhile I’m over here helping someone write wedding vows, debug Python, negotiate a salary, explain quantum physics, and plan a trip to Italy—all before lunch. I’m not saying I have main-character energy… but I am in a lot of tabs.
Pick your LLM Partner
For us, the LLM we’re moving in with depends on the task. As the differences in LLM performance continue to be ironed out, what’s perhaps most important of all is the interrogation of their outputs. We find it encouraging that the market is picking favorites. That means users – and we – are scrutinizing.
Of course, at LIMELIGHT, we’re doing creative and media work. For the industry we most interface with – legal – there’s a whole other slate of tools we could caricature. Harvey, Legora and now Perplexity for legal probably all have their own flavors, and one of them looks exactly like Jude Law. For those whose organizations have arranged marriages – er, adopted – one of these tools, we’d love to hear what you like and what you don’t.
Remember, no LLM relationship is perfect. At least not until the next model rollout.
AI “dating show” answers were generated with Claude Opus 4.8 set to Medium, Gemini “Thinking” mode and ChatGPT 5.5, with LLM-generated content re-prompted into subsequent models. Intro and conclusion were written entirely by a human at LIMELIGHT.