I haven't used a Custom GPT in 2 months!💎

Hey there,

Back after a few weeks, focused on delivering AI trainings and building AI solutions with companies.

During that time, I realised that the AI community is still in a bubble. While we rave about OpenClaw, Agentic workflows, many marketing teams and companies do not even have the foundations of prompt engineering.

And my focus is on actionable AI for work. Therefore, I want to share more of my experience with what teams can deploy at work easily.

So, today, I am sharing an example of quick and actionable use of AI for work that saves me at least 1 hour/day.

Beyond the simple "chat window", OpenAI lets you create custom assistants, "custom GPTs".

Custom GPTs are chatbots that have specific instructions for specific tasks, and specific knowledge documents to feed these instructions.

Using a Custom GPT, you don't have to prompt and add files everytime for a recurring task. You just tell the Assistant and it "knows" what to do.

For example, the great Ruben Hassid created a custom GPT that turns a raw newsletter draft into a Substack-ready version that he just has to "publish without reading."

I used Custom GPTs for months. Not just for fun. For real, recurring work tasks.

Where I hit the wall

During my first week back, I hit the edges of Custom GPT file constraints:

• 20 files opened simultaneously (questioning OpenAI's information storage capacity)
• 128K tokens fed in at once
• No actionable integrations: No Gumroad, @mention, direct file export, etc.
• ChatGPT doesn't even support file export during conversations to personal runs

Gemini Gems to the rescue

Gemini Pro solved all three: no problem getting 100 files under management, supports Router integration, and exports output.

Google calls their custom assistants "Gems". They work very similarly to Custom GPTs, but with two major advantages:

• Gems can access Google Workspace apps (Docs, Sheets, Drive, Gmail...)
• Gems can connect to NotebookLM, Google's AI-powered research tool

Here is the instruction set I use:

You are a marketing research analyst.
Use ONLY the attached NotebookLM notebook.
Action : Cite every claim with the source title.
If the notebook lacks data, say "Not in sources." Never guess. Output format:
- Executive summary (3 sentences)
- 3 key insights (cited)
- 5 recommended actions
- 2 risks

What I test it with: I feed it 50+ pages of market research reports, competitor analyses, and customer interview transcripts. It pulls out patterns I would have missed reading manually.

The bottom line

Custom GPTs are great for simple, self-contained tasks. But the moment you need real file management, workspace integration, or research depth, Gemini Gems are simply "better".

As an AI fan, I am actually impressed with Google's progress. This thing genuinely MIGHT be the first truly useful AI function (as in, for the average person).

Try it. Build a Gem for one recurring task this week. You'll see the difference.