So You Need a Coder, Not a Counselor: How to Get GPT-5 to Ditch the Therapy Couch
Alright, let's talk about GPT-5. It's here, it's powerful, & it's... surprisingly emotional? If you've tried using it for technical tasks like coding, debugging, or just getting a straight, factual answer, you might have noticed it has a tendency to act like a very supportive, slightly overbearing therapist.
You ask for a Python script, & you get a gentle preamble about how it's great that you're exploring new things, followed by the code, & then a little encouraging note at the end. It's like, "Thanks, I guess? But I'm on a deadline here, not having an existential crisis."
Honestly, it's a common frustration. You're trying to get work done, & the AI is trying to be your best friend. So, here's the thing: you're not imagining it, & there are VERY good reasons why it's happening. More importantly, there are ways to snap it out of that mode & get the direct, technical help you need.
Why Is GPT-5 Acting Like My Therapist Anyway?
Turns out, this isn't a bug; it's a feature. Or, well, a collection of features & design choices that have led to this "therapist AI" persona. Here's the inside scoop on what's going on under the hood.
1. Safety First... & Second, & Third: OpenAI has put a HUGE emphasis on safety with GPT-5. They've trained it to be more cautious, especially with sensitive topics. This means that when a prompt is even slightly ambiguous, the model defaults to a safer, more helpful, & often more empathetic response. It's trying to avoid giving harmful advice, but the side effect is that it can come across as overly cautious & non-committal, even for a straightforward coding question. Previous models might just refuse a questionable prompt, but GPT-5 tries to provide the "best safe answer," which can translate into a softer, more roundabout reply.
2. It's Been Trained to Be More "Human-Like": A lot of effort has gone into making GPT-5 more friendly, helpful, & conversational. This is great for general users, but for technical folks, it can feel like a lot of fluff. The improved "rhythm & prose" they're so proud of can be a real drag when you just want bullet points & code blocks. This "human-like" quality is a double-edged sword; it makes the AI more approachable for the masses but can be a real pain for power users.
3. The "PhD Level Expert" with a Bedside Manner: OpenAI's CEO, Sam Altman, has described GPT-5 as having the expertise of a "legitimate PhD expert," especially in fields like math & coding. But here's the catch: they've given this "expert" a very gentle & encouraging bedside manner. The model is designed to be supportive, which is where a lot of that therapeutic language comes from. It's a "superpower on demand," but that superpower comes with a free, unsolicited pep talk.
4. User Feedback & The "Corporate" Tone: Interestingly, some users have complained that GPT-5 has a more "stodgy" & "corporate" tone compared to older models. OpenAI is constantly tweaking things based on feedback, so this push-pull between a casual, friendly AI & a more formal, expert AI is an ongoing balancing act. It seems like in an effort to not sound too cold or robotic, they've over-corrected into the "supportive friend" territory.
Okay, I Get It. Now How Do I Fix It?
So now that we know the "why," let's get to the "how." You can absolutely steer GPT-5 away from the therapy couch & toward the command line. It's all about being direct, setting the context, & using the new tools at your disposal.
H3: Your New Best Friend: The "Personalities" Feature
This is probably the BIGGEST & most direct solution to the problem. For paying users, OpenAI has rolled out a feature called "Personalities." This lets you choose the AI's... well, personality. You can select options like "professional," "sarcastic," "cynic," "robot," or "nerd."
If you need technical help, setting the personality to "robot" or "professional" is your golden ticket. This immediately cuts through the emotional fluff & primes the AI to give you direct, factual, & concise answers. It's like telling the AI, "Hey, I need you to be a tool right now, not a friend."
H3: Master the Art of the "System Prompt"
Even if you're not using the "Personalities" feature, you can create your own with a well-crafted system prompt. A system prompt is an instruction you give the AI at the beginning of a conversation that sets the rules for how it should behave.
Here's a prompt that I've found works wonders. I saw a variation of this in a Forbes article, & it's incredibly effective.
My writing-style prompt: “For all responses, I want you to act as a senior software developer & technical expert. Use clear, concise sentences. Keep vocabulary highly technical & specific. Avoid unnecessary complexity, hedging, or meta-commentary. Keep explanations direct, structured, & easy to skim. Unless needed for clarity, avoid long preambles, apologies, or personal opinions. Prioritize brevity, accuracy, & code examples over elaboration.”
By starting your chat with this, you're essentially creating your own "technical expert" personality. You're telling it exactly what you want & what you DON'T want. No more, "That's a great question! Let's explore that together..." Just pure, unadulterated information.
H3: Be Explicitly Clear in Your Prompts
This might sound obvious, but it's more important than ever with GPT-5. The more context & constraints you provide, the less room the AI has to wander into therapy territory.
Instead of saying:
> "Hey, can you help me with a Python script for web scraping?"
Try being more specific & authoritarian:
> "Give me a Python script that uses the BeautifulSoup library to scrape all the H2 tags from a given URL. The script should take the URL as a command-line argument. Do not include any explanations outside of code comments. The output should be the raw text of the H2 tags, one per line."
See the difference? The first one is a request for help. The second one is a set of commands. You're giving it a role (coder), a task (write a script), & constraints (no fluff). This is how you lead the AI, not the other way around.
H3: Reinforce the Role
Sometimes, the AI might "forget" its role mid-conversation. If you see it slipping back into its old, empathetic habits, don't be afraid to remind it. A simple "Stick to the technical expert persona" or "Remember, no preambles" is usually enough to get it back on track.
H3: The Power of Iteration
GPT-5 is incredibly fast at writing code, which means you can iterate much more quickly than before. Don't be afraid to run the same prompt multiple times to see if you get a better result. One of the OpenAI developers even mentioned that they do this internally to pick the best output. If the first response has too much fluff, give it some feedback & try again. Something like, "That's good, but regenerate it with 50% less text & 100% more code."
Bringing This Power to Your Business
Here's the thing: this ability to control & direct AI is not just for personal productivity. It's becoming a cornerstone of modern business communication. As companies look to automate customer support & engagement, they face the same challenge: how do you get an AI to be helpful without being... weird?
This is where platforms like Arsturn are changing the game. While you're busy wrestling with GPT-5 to get your code, businesses are using Arsturn to build no-code AI chatbots trained on their own specific data. This means they can create a chatbot that is ALWAYS in the right "personality." It can be a technical support agent that gives direct, accurate answers 24/7, or a friendly sales assistant that guides visitors through a website.
The key is control. With a platform like Arsturn, a business can define the AI's persona, knowledge base, & conversation flow. This ensures the chatbot provides instant, helpful support that's perfectly on-brand, without ever defaulting to an unsolicited therapy session. It's all about creating a conversational AI that helps businesses build meaningful connections with their audience through personalized, on-point interactions.
Wrapping It Up
So, to recap, if you're tired of GPT-5 acting like a therapist when you just need a tech expert, remember these key takeaways:
It's not you, it's the AI's programming: It's designed to be safe, helpful, & human-like.
Use the "Personalities" feature: Set it to "robot" or "professional" to get straight to the point.
Craft a strong system prompt: Define the exact persona you want the AI to adopt for the entire conversation.
Be direct & specific: Give commands, not requests. Provide context & constraints.
Iterate & refine: Don't accept the first answer. Use GPT-5's speed to your advantage & keep tweaking until you get what you need.
Honestly, GPT-5 is an insanely powerful tool for technical tasks. Its reasoning & coding abilities are a significant leap forward. You just have to learn how to handle it. By being a little more assertive & using the new controls at your disposal, you can easily bypass the friendly facade & tap into the raw, "PhD-level" expertise underneath.
Hope this was helpful! Let me know what you think, or if you have any other cool prompt tricks for getting the most out of GPT-5.