Why everyone's furious with ChatGPT-5 😤

Hi Non-Techies,

Well, this is awkward.

ChatGPT-5 finally dropped 11 days ago, and instead of a celebration, we got a Reddit thread titled "GPT-5 is horrible" with 4,600 upvotes.

We were told we were getting an epic upgrade.
What we seemed to get was an AI having a proper personality crisis.

Let me explain what's really happening – and how to fix it.

The Great ChatGPT Personality Lobotomy

The biggest complaint isn't about intelligence (GPT-5 is genuinely, measurably smarter). It's about how it feels.

Users are saying GPT-5 talks like a "LinkedIn-approved ghost" and they're "genuinely grieving" GPT-4's warmth. One person nailed it: "Sure, GPT-5 might be a straight-A student, but GPT-4 was a friend."

Yup. OpenAI optimised for accuracy and accidentally performed a personality lobotomy. Where GPT-4 felt conversational, GPT-5 reads a bit like a corporate memo. It’s weird, and feels like a huge step back.

When a "PhD-Level Expert" Can't Count Letters

Then there are the technical blunders.

GPT-5 insisted there are three "b"s in "blueberry" and got confused about US states. This echoes past, well-documented problems (it couldn’t count the “r”s in “strawberry” a long time ago) - not exactly the expert-level performance promised.

The culprit is a new routing system that decides whether to use "fast" mode or "thinking" mode. When it glitches, simple questions get routed to the wrong model, creating these sorts of facepalm moments.

The Subscription Swindle

They also got loyal paying users feeling properly wound up.

ChatGPT Plus subscribers suddenly found themselves capped at 200 messages per week for the good stuff, while their trusted older models vanished overnight.

As one Reddit user put it: "It feels like a downgrade branded as the new hotness."

They’ve got to be careful. They could alienate a massive base of paying users if they continue to offer very limited usage. The free version needs to be substantially worse, or why bother paying!?

Here’s How to Actually Make GPT-5 Useful

Before you give up entirely, some early adopters have sussed out how to make it work properly:

1. Fix the personality crisis

OpenAI added four personality modes because people hated the corporate tone. Go to Settings > Personalisation and pick Cynic (sarcastic), Robot (efficient), Listener (warm), or Nerd (enthusiastic). Or add your own in.

Here’s mine, if you’re interested:

2. Prompt like you're briefing a PhD student

GPT-5 is hyper-sensitive to instruction style. Instead of "write me a marketing plan," try "Break this into components, then create a structured plan for each." This is called ‘chain of thought’ prompting - literally share how you’d do it manually as a human. More specific = better results.

3. Master the three modes

You can now choose: Fast (simple stuff), Thinking (complex problems), or Auto (let it decide). This prevents those "blueberry has three b's" disasters.

4. Feed it bigger projects

GPT-5's context window is massive – upload entire documents instead of breaking everything into tiny chunks. It's designed for substantial inputs and genuinely can handle all the extra stuff.

5. Treat it like a conversation

Power users found it works best with back-and-forth. Start broad, then refine: "expand the opportunities section" or "make this more specific." It’s not just a ‘click and go’ thing.

There's Hope (I Think…)

The encouraging bit is that OpenAI is listening.

Within days, Sam Altman brought back GPT-4o (see pic below) and promised to make GPT-5 "warmer." They've increased rate limits and are working on better personalisation.

Plus, the fundamentals are solid – 45% fewer factual errors than GPT-4o, better reasoning and significantly improved accuracy. It's just the user experience that needed work.

I think a few more weeks of this active listening, and we’ll have ourselves a much stronger product.

Your CRM just learned how to talk 🤯

Tired of fighting your CRM for basic information? We've all been there - needing simple answers like "How many deals are closing this month?" but having to do 25 clicks, build custom reports, and practically perform a rain dance just to get basic data. It's unnecessarily faffy and frustrating.

Well, HubSpot and ChatGPT just solved this by partnering up so you can ask your CRM questions in plain English and get actual answers. 

To access, go to your ChatGPT: 

Click on your name >> Settings >> Connectors >> Hubspot >> Connect

You'll need paid ChatGPT and HubSpot accounts, but if you're done with the endless clicking, this is your solution. 

There's a live demo on Thursday, 21st August if you want to see it in action:

Right, it’s my birthday today, so I’m now off to go and eat yet more of that lovely Costco birthday cake (what do they put in it to make it so sickly but so addictive?!)

See you next week,

Heather and the AIFNT team.