#11 - From Hands-free Prompting to Custom GPTs

From dictation to custom GPT (incl. a messaging evolution analyst) | ChatGPT or automated AI workflows? | More Deep Research Credits | Hubspot's AI expectations of their leaders

If you only check one highlight this week: How to create prompts hands-free with o3. If you already know that, learn how to create a custom GPT (it’s super easy).

- François

🛠️ From Handsfree Prompt Creation to Custom GPT

Reasoning models, especially OpenAI's o3, have become so good that they create better prompts than most humans.

Like humans, they need a good brief.

So, unless I'm in a rush, I always start with o3 (even for copy writing).

And I love starting with dictation to go faster.

Here's my workflow:

  1. Open the ChatGPT mobile app (works in browser and desktop client too)

  2. Select o3. It's their best reasoning model. It will “think” and structure things better

  3. Press the dictate button (mic icon), NOT advanced voice mode (black sound wave icon to the right of the mic). This way you can read and confirm your brief and see the resulting prompts.

  4. “talk” your brief and what you want the prompt to do: goal, context, instructions, deliverable format, etc. Example below. Specify you want a prompt for a reasoning model

  5. Ask ChatGPT to recap your ask so you can review and adjust before it writes the prompt

  6. Fine tune if needed (I often realize I forgot something when reading the recap)

  7. Tell it to write the prompt once things look right

  8. Review, update and test the prompt

Pick o3 as the model and dictate with that “mic” option. Don't use advanced voice mode.

Like your prompt? Think you will use it often?

Then save it in a Custom GPT.

It’s so easy, I’m always surprised so few people have ever created one.

  1. Go to “Explore GPTs in the top left panel when in your bowser (can’t create a GPT in apps yet)

  2. “+Create” button at the top right

  3. Work in “Configure” mode in the left panel

  4. Drop the prompt in instructions (you can add conversation starters relevant to your project - but that’s optional)

  5. Select capabilities

  6. Publish your GPT

You can share it with friends and colleagues now.

💡 Tip: publish it to the GPT store (don't worry, hardly anyone will find it: that GPT “app” store is a bust ) for better with web search access (thanks Guillaume for the tip).

Here’s an example of a brief I dictated into ChatGPT to analyze the evolution of a company's messaging. Remember, this is NOT the final prompt, just the initial brief from which o3 will write the actual prompt.

🗣️🎙️ “Help me write a prompt that will be used for your reasoning model with deep research.

Go to the Internet Archive and analyze the evolution of messaging for a given company based on how their messaging and positioning evolved on their website in the last 3 years.
Analyze snapshots from 2 years ago and then every 6 months since then, up to the most recent.

He’s the deliverable format I need:
- First, an executive summary of the evolution of their messaging in the last two years and what it is today.
- Second, in a table, the messaging for each period you analyzed, who you believe the target audience was for each period, and why you believe the messaging evolved based on what you see.

Before writing the prompt, recap my ask.”

Interested in using the resulting prompt for your own analysis?

Here's the prompt after light edits, and here's my resulting “Messaging Evolution Analysis” GPT. I did all that under 7 minutes.

Give it a try pls email me he outcome + what worked/didn’t work so I can fine tune it.

My selection of tips, news and workflows

🛠️ When should we use ChatGPT vs. an AI agent/workflow tool?

Jacob Bank shares his quick answer:

“1. Use ChatGPT (or another chatbot like Claude or Gemini) if:

- It's a one-off use case that you aren't doing very frequently

- If the input and output is primarily text

- If you want to chat and refine the output

For example, if you want to create an annual marketing strategy, ChatGPT is perfect. You don't do it very often, you want a text report as output based on your text prompt as input and you want to go back and forth and refine the report with the AI.

2. Use an AI agent/workflow tool (like Relay.app, Zapier, Lindy, or Gumloop) if:

- It's a frequently repeated use case

- If it needs to access structured data from your tools

- If it needs to automatically take action in your tools

For example, if you want to create a weekly marketing performance report, this is a perfect fit. You do it every week. You need to pull data from spreadsheets, emails, docs, etc. And you need to put the output automatically in a doc or Slack.”

💡I’ll add that there is a middle option, which consists in creating a custom GPT, as explained in the section above if you are going to reuse the same prompt often.

🛠️ Deep Research Credits in ChatGPT increased

Nicole explains what we get:

👊 Five ways Hubspot execs build an AI-first mindset on their teams

This is Hubspot’s CEO’s mandate to her leaders. They need to:

tl;dr:
- Use AI daily: Lead by example
- Apply constraints: Give clear, focused challenges
- Establish tiger teams: Empower small, agile groups
- Be a learn-it-all: Foster continuous learning
- Measure progress and share it openly

Final Words

âťť

Every single exec has to have an AI plan. It’s the CEO’s OKR. She has her name on it, and we all need to support it.

A CMO at an event I co-organized last week

Thanks for sharing these highlights with busy marketing execs around you.🙏 

Someone forwarded you this email? You can subscribe here.

François | LinkedIn 

I'm a CMO, advisor, and "CMO Wingman". Yes, that's a thing :-). Ask my clients: in this AI era, things move so fast that CMOs need a strategic and supportive advisor more than ever. I’m former CMO at Twilio, Augment Code, Apollo GraphQL, Decibel, Udacity and Head of Marketing for LinkedIn Talent Solutions.