Skip to main content

Stop Using AI Like a Beginner (2026 Framework)

•47:51

By , founder of Genial Agency

AI Framework

You will have tried the three levels of AI use I lay out: getting more out of ChatGPT, Gemini and Claude, automating a real workflow with Make or n8n, and handing a browsing task to an agent in Antigravity.

What you need

  • A paid ChatGPT account (I mention the $20 plan)

  • A Google account for Gemini

  • A Claude account

  • A Make.com or n8n account for level 2

  • Antigravity installed for level 3 (optional)

Free downloadAI levels cheat sheetThe prompts, settings, workflow examples and five skill questions from my three levels of AI use, on one page.
Download

Step by step

  1. Use Deep Research instead of plain chat

    In ChatGPT, turn on Deep Research for questions that need more than the first page of Google. Try one of my two uses:

    • Product comparison: name two products and describe your own use case in detail.
    • Sales research: research a prospect, a company or its decision-makers before a meeting. Start by running it on yourself.
  2. Stop ChatGPT from keeping your data

    Open Settings, go to Data controls and turn off Improve the model for everyone. As I explain in the video, this setting means they keep your data, so switch it off as soon as you get your subscription.

  3. Build a custom GPT for FAQs or team rules

    Write a short text document about your services and policies (AI can write it for you). Then describe to ChatGPT what the custom GPT should do, test it until you like the answers, publish it and share the link. The people who use it don't need to pay. A version holding your brand guidelines helps a marketing team check its own work.

    Test questions I ask my FAQ bot
    What are the services that this company offers? What are the return policies? Do you offer student discounts?
  4. Give Gemini long documents, then turn the results into visuals

    Upload a long report (I use a 150-page Meta financial report) and ask Gemini to pull out the figures you need. Gemini gives the page and line for each one, so you can check anything that looks wrong. Then ask for an image. If it doesn't use NanoBanana Pro, click Tools, then Create images.

    Report prompt
    Here is a report from Meta. I want you to go into this report, analyze it, and get all of this information out there.
    Image prompt
    Can you create an image of the financial KPIs year on year that you extracted from that report, please?
    Slide prompt
    Now I want you to create a slide using the KPI data.
  5. Build a small app with Claude artifacts

    In Claude, go into apps and websites and describe a simple app. When Claude asks, confirm with "Create it". Once it works, ask for more features, such as multiplayer or scores. You can use the same approach for dashboards and client portals.

    Coin toss prompt
    I just want you to generate a very simple coin toss game with heads and tails.
  6. Automate a written-down process first, without AI

    Write out the steps of one workflow you repeat (your SOP). Then automate it in Make.com. My example: list every invoice you send in a Google Sheet. Until you remove a row, the client gets a reminder after 5, 7, 14, 21, 30 and 60 days, and each email is firmer than the last. This step uses no AI.

  7. Add one AI decision step to a workflow

    In n8n (or Make), add one AI step where a person would normally decide yes or no. In my job post analyzer, the workflow collects job posts that contain AI keywords. A ChatGPT step decides whether each company is worth talking to, and the good ones go into a database.

  8. Try an agent for a narrow, repetitive conversation

    Build a small agent (mine has six nodes) for one job. My WhatsApp agent for a coach asks new leads what motivates them, what their budget is and when they want to start. Once it has the answers, it sends the coach a summary. Keep the task narrow, because agents don't always do the same thing every time.

  9. Hand a browsing task to an agent in Antigravity

    Open Antigravity. It can use your browser, the web and the files on your computer. Ask it to browse a site and report back, then keep working while it runs. I also use it to test websites (accessibility, responsiveness, bugs) and to build dashboards.

    Browsing prompt
    Can you go to the Apple website and give me the latest products that they released, please? Just create a list for me after you browse it.

Watch out for

  • Many useful automations don't need AI at all. My invoice reminders, onboarding and client portals work without it.

  • You can't automate a process nobody has written down. Document your steps (SOPs) before you build anything.

  • AI is probabilistic, so agents don't do exactly the same thing every time. Weigh that before relying on one for work where consistency matters.

  • Unreviewed AI output becomes "AI slop". Have someone with good taste check what the AI produces.

Watch next

Related tutorials

Upskilling: Hands-on AI training for your team

Want your team to move past beginner AI use?

We train your team across all three levels, from chat tools to workflows and agents. The next step is a call.