Two people sign up for Perplexity Computer on the same day. Same $200 subscription. Same 19 models. Same 400+ connectors. Same everything.
A month later, one of them is shipping client research packets in fifteen minutes and quietly billing $4K a pop for them. The other is staring at an output that feels like a slightly worse ChatGPT and wondering what the fuss is about.
The tool didn’t change. The prompt did.
I’ve watched this gap show up over and over since Computer launched. It’s not about who’s smarter or more technical. It’s about one thing: do you talk to Computer like a chatbot, or do you brief it like an employee?
Almost everyone is doing the first thing. This guide walks you through the second.
Before the prompting stuff, you need to understand why prompting it like ChatGPT is such a bad fit in the first place.
Computer isn’t a chatbot. It’s a cloud-based AI worker that lives inside your Perplexity account. You give it a job. It breaks the job into pieces, picks the right AI model for each piece, and runs the whole thing in the background while you go do something else.
It runs 19 different models under the hood. Claude Opus handles the heavy reasoning. Gemini does deep research. Grok handles the fast, simple stuff. GPT-5.2 manages long documents. You don’t pick which one. Computer routes the work for you.
It also connects to 400+ apps Gmail, Slack, Notion, Google Drive, GitHub, Salesforce, Snowflake, the works. So when you ask it to do something, it can actually pull from your real data and write back to your real tools.
Every task runs on your credit balance. Simple tasks burn 30 credits. Medium tasks burn a few hundred. A messy, badly-prompted task can burn five thousand and give you garbage at the end.
That last part is why prompting matters so much. Every vague word you type costs you real money.
When you prompt Claude or ChatGPT, you are talking to one brain. It reads what you wrote, thinks, and answers. Sloppy prompts get sloppy answers, but that’s about it.
When you prompt Computer, you’re not talking to a brain. You are briefing a project manager who’s about to deploy 19 specialists on your job. If your brief is vague, all 19 specialists start guessing. They guess in parallel. You pay for every guess.
That’s why the same prompt that worked fine in ChatGPT feels expensive and underwhelming in Computer. You weren’t wrong before. The rules just changed.
There are three shifts that fix everything:
Tell it the finish line, not the route. Don’t list steps. Describe what DONE looks like and let Computer figure out how to get there. Micromanaging an orchestrator defeats the entire reason you’re using one.
Set hard rules, not friendly suggestions. Computer can send emails, edit files, post in Slack, and spend money. If you don’t tell it what NOT to do, it will pick for you. Sometimes badly.
Define the inputs and the outputs. Skip the middle. Tell it exactly where to pull data from. Tell it exactly what to deliver and where to put it. The middle is its job.
That’s the whole shift. Everything below is just applying it.
The locked section covers the full system:
the 3 mental shifts that change everything about how you prompt computer the 7 rules that separate the 10% making real money from everyone else the dispatch brief — the exact template i use for every workflow (copy + paste) 3 ready-to-steal briefs that print money: weekly competitive intel ($1-3K/month retainers), financial due diligence packets ($500-2K each), and a content repurposing system that turns one piece into ten the 5 mistakes burning your credits right now (one of them cuts your bill in half) how to stack ten of these into a one-person business
same tool. same $200 subscription. completely different bank account.
upgrade to keep reading.
