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 made several useful GPTs (prompt-making, Gamma Slide builder, Reddit Search).
Ruben’s Custom GPTs.
I built Custom GPTs and Projects for everything. Blog drafts. Competitor analysis. Email replies. Persona research.
For a while, they worked.
Then my competitor folder hit 73 files. The GPT took 20. Created another project
Then I fed a 200-page industry report. ChatGPT read 47 pages and made up the rest.
Then I needed to draft a client email. So I wrote it in ChatGPT, copied it, switched tabs, pasted it into Gmail, realised I forgot the CTA, switched back, copied again.
The tool that was supposed to save me time had become the slow part of my workflow.
Because Custom GPTs max out at 20 files and 128K tokens. They don’t have access to my custom instructions.
Then Google Gemini 3.0 came out.
I started using Gemini more. And more. Until I decided to rebuild my “custom assistants” (some AI influencers would even call them “AI Agents” 🤦🏻‍♂️) in Gemini Gems.
Three agents. 147 sources. No tab-switching. That was two months ago.
Gemini Gems:
connect to NotebookLM (300+ sources),
work inside your Google Workspace
I moved my entire stack of custom assistants. Setup: 10 minutes each.
I have not opened my Custom GPTs since. (And in all honesty, I open ChatGPT less and less every day. But more on that soon)
What actually changed
As I explained: Custom GPTs (and projects) are personalised ChatGPT agents. You write instructions, upload up to 20 files, and the GPT repeats the instructions when asked.
Useful for contained tasks. Limited for anything research-heavy.
Gemini Gems are the same idea inside Google’s ecosystem. Same persistence. Same custom instructions. But they now connect to NotebookLM (Google’s research tool, basically a 300-source filing cabinet your agent can search - 50 for the free version). And they run inside Gmail and Docs without leaving the app.
Same concept. But different ceiling.
Google Gemini Gem Manager
Here’s how they compare on the things that actually matter for marketing work:
Knowledge base: Gems get 300+ sources via NotebookLM. GPTs max at 20 files.
Auto-sync: Add a source to NotebookLM, the Gem sees it. With GPTs, you re-upload manually every time. I update a case study once. The Gem knows.
Context window: 10M tokens on Gemini Pro vs. 128K on GPTs. I fed it a 200-page report. It read every page.
Multimodal: Screenshots, PDFs, images all work natively. Competitor ad analysis in one prompt.
Workspace: Gmail, Docs, Sheets native. I draft emails without leaving my inbox.
Cost: Included with Google Workspace (and a free NotebookLM holds up to 50 sources). Custom GPTs require a paid plan to create.
The gap that decided it for me: I needed 80+ updated sources and Workspace integration.
How I set up a Gem (<5 minutes, for real)
Go to gemini.google.com. Sign in.
Click “+” in the sidebar. “New Gem.”
Name it something specific. “Competitor Intel Pro” works. “Marketing Helper” does not.
Write instructions: what it is, what it does, what it avoids, how it formats output.
Add knowledge. Upload files directly. Or click “+” > “NotebookLM” to connect an entire notebook.
Test it. Run a real prompt.
Save. Every future session starts with full context.
No code. No API keys.
Now here are 3 examples of Gems for Product Marketing.
Note: You can start with Google’s preloaded Gems for different tasks (Copy, Learning, Outreach, Storytelling…)
Subscribe now
Gem 1: The research hub
This one replaced my most frustrating GPT.
I had a research agent with files. Every time I found a good report or competitor teardown, I had to delete something to make room. It felt like packing a suitcase for a two-week trip with a carry-on.
Now I run a NotebookLM notebook with 120+ sources. Industry reports. Competitor pages. ICP interviews. Podcast transcripts. Reddit threads. The Gem searches across all of them and cites every claim.
Access to NotebookLM is a winning feature vs ChatGPT
Here is the instruction set I use:
You are a marketing research analyst.
Use ONLY the attached NotebookLM notebook.
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: “What objections do enterprise buyers raise about AI training? Cite sources.”
If any insight comes back without a citation, I ask it to re-check. If it still cannot cite, I delete the claim. No exceptions.
Research briefs that took 90 minutes now take about 10 minutes. Every claim traceable. No hallucinated stats. I actually trust the output enough to send it to clients.
Gem 2: The competitor scout
I used to screenshot competitor ads, download their blog posts, save their pricing pages. Then analyse them one by one in ChatGPT. Slowly.
This Gem takes everything at once. Screenshots, PDFs, landing page copy. Returns a SWOT table, gap analysis, and counter-moves. One prompt.
Analyse marketing content I upload.
For each: extract headline, hook, CTA, keywords, visual style.
Rate weaknesses 1 to 10 with reasoning.
Identify 3 messaging gaps.
Suggest 5 counter-moves I can ship this week.
Output as a SWOT table.
Be specific. "Their CTA is vague" is useless.
"Their CTA says 'Learn More' with no urgency or benefit" is useful.
The Gemini multimodal piece matters here.
I upload a competitor’s landing page screenshot and the Gem reads the image. Layout, copy, CTA placement, colour contrast. A Custom GPT barely handles this.
One thing to watch: the Gem analyses what you upload. It does not browse live sites. If competitor content changes, you need fresh screenshots.
I found a positioning gap this way that became the angle for a campaign. Took 25 minutes instead of the usual three hours.
Gem 3: The email draft assistant
This is the one that saved me the most daily time. And it is the simplest.
Gemini is embedded in Google Workspace Apps now. And Gems come along.
This means that, in any Google app (Gmail,Sheets, etc) you can summon your gems directly.
GEMS are available directly withing Google Apps
In my case, I was drafting emails in ChatGPT, copying them, pasting them into Gmail, editing them, realising the tone was wrong, going back. Twenty times a day. The context-switch tax was absurd.
This Gem lives inside Gmail.
Draft professional emails for a B2B marketer.
Match my voice: direct, warm, concise.
Keep under 150 words.
Every email needs one clear next step.
Never use: "Hope this finds you well," "Just following up," "Circling back."
I uploaded 50 of my best sent emails as tone examples and negociation frameworks to apply.
Enabled Workspace integration (Settings > Connected apps > Google Workspace).
Now I open Gmail. Click the Gemini icon. Select the Gem.
Describe the task on the email i want directly: “Draft a reply. The Prospect is price-sensitive. Anchor on value, not cost.” Edit. Send.
One thing I still do: review every reply before sending.
The Gem does not know my relationship with the person. It does not know we had a tense call last Tuesday.
Context it cannot see, I add manually.
Email time dropped by roughly 40 minutes a day.
My replies are tighter and I stopped sounding like a chatbot pretending to be polite.
Where this breaks
I would be lying if I said Gems are perfect. A few things to know.
NotebookLM caps at around 300 sources per notebook.
For most PMMs this is plenty. If you run a massive research operation, you will need multiple notebooks.
Gems do not browse the web live. They work with what you give them. If your competitor changes their messaging next week, your Gem does not know until you upload fresh material.
The Workspace integration works on desktop. Mobile is limited. If you live in the Gmail app on your phone, you will still tab-switch for now probably.
And Gemini’s output still needs editing. It is better than ChatGPT for my use cases, but “better” does not mean “publish without reading.”
I treat every output like a first draft from a sharp but literal-minded intern.
ALWAYS KEEP A HUMAN IN THE LOOP!
Thanks for reading Vibe Product Marketing! This post is public so feel free to share it.Share
Bottom line
Custom GPTs taught me what persistent AI agents could do.
Gemini Gems showed me what happens when you remove the walls.
300 sources instead of 20. Auto-sync instead of re-upload. Google Workspace-native instead of tab-switching.
Context is key with AI. And having access to more context means more relevance, more personalisation.
Gemini works within the Google ecosystem. You can leverage all the contextual elements from your different files.
NotebookLM is the best AI knowledge hub available. You can feed 100s of pages to your Gem assistant. On any topic.
Tailored assistants for each task means more relevant instructions, fitted for the task.
Bonus: you can set a Nano Banana Pro Gem to create tailored images with your design (that’s how I craft my Linkedin Carousel slides)
I went from decent Custom GPTs to context-rich Gemini Gems. From static assistants to dynamic, multimodal, embedded partners. I am not going back.
That’s a wrap for today.
Let us know what Gems you built for Product and Marketing. Curious to find more use cases to explore.
