If you only check one highlight this week: understand how o3 works and then play with it. And again.

- François

🛠️ The Reasoning Models Improved and that’s a Big Deal

It's worth truly understanding how reasoning models such as o3 and o4-mini (available in ChatGPT Plus and Pro) work and how to use them.

Given that some reported “feeling the AGI” when using ChatGPT o3...

These models have access to tools. They can plan. They select which of these tools to use and when, in order to hit their goals:

• Browser – pulls info from the web in real time

• Python / Advanced Data Analysis – runs code to crunch data, build charts, transform files

• File Reader – ingests and reasons over anything you upload (PDFs, slides, CSVs, code, images)

• Vision – “looks” at and “thinks with” images, zooms-in, rotates, OCRs, and uses what it sees in its reasoning. Can even find where pics are taken sometimes (with metadata removed)!

And while you couldn’t use these with o1, these are available with o3:

• Image Gen – creates or edits images with the right text and consistent styles

• Advanced voice mode - talk to o3 in both desktop and mobile apps

o3 has great memory since OpenAI turned on memory by default two weeks ago. The improvement in quality and relevance when it remembers all my chats is impressive. You can tell it to add to its memory when you see or create something you want to reuse (after creating a writing and illustration style that I like, I simply told it to “save that to your memory and use it going forward unless I indicate otherwise” and it works).

The more you work with it, the more it memorizes, and the less you have to give it context or super detailed prompts (btw, these models create great prompts for you, lowering the prompting bar).

It also reasons in plain sight (like most reasoning models): you see its “chain of thought” unfold as it thinks and plans.

I used to ignore that and let a task run in the background.

Mistake if the stakes are high.

Ethan Mollick’s reminds us: even super-smart models fail spectacularly at weird stuff, like completely misreading a twist on a classic brain teaser because it had “seen” the original version too many times in training. You still need to test, guide, and check them.

It's like working with an associate. You want to verify how they work, the steps they take, if they actually use tools and what sources they use.

When you get it right, it’s like having your own super fast business analyst AND creative studio. It can not only create full reports with analyses, images, graphs, but also generate new ideas and craft full new marketing strategies and campaigns (style guides, emails, billboards, landing pages, social posts) and more.

Kipp Bodnar (Hubspot CMO) was impressed when o3 generated strong new ideas for leveraging image gen capabilities for marketing, including suggesting to use it for ultra local campaigns.

Ethan Mollick also reported that “You can now launch a full cheese shop—logo, financials, competitive analysis, website—using just one prompt”.

True, it’s possible. But I don't recommend using it like that. After using o3 for a while, I recommend breaking down the steps and iterating in the same chat, instead of doing “just one prompt”. i.e. don’t go in one fell swoop from:

• craft prompt to

• research to

• plan to

• execution

Instead, first, review the prompt it generates for you, like you would review the strategic brief of an associate on your team. Then, pause again to review and refine the analysis and plan before moving to execution.

You are the expert. The tool works for you and requires your guidance, taste and judgement. Otherwise, we’ll just create more “AI slop”. The world doesn't need that now.

I hope you'll have as much fun using and pushing o3 as I’m having.

My selection of tips, news and workflows

🧠 Conversion Rate and Landing Page Optimization with ChatGPT o3 + its Vision

You can use o3 for conversion rate optimization. Nicole Leffer shares a good workflow.

tl;dr:

1. Use o3 in ChatGPT with Vision enabled

2. Start with a detailed expert persona (“You are…” prompt)

3. Upload a screenshot, not a link, of your page

4. Ask it to describe what it sees before analyzing

5. Feed it rich context: goals, audience, traffic source, etc.

6. Prompt it for a full CRO analysis—skip Code Interpreter for this

» Marketers can now get page-specific, context-aware CRO audits that rival what you'd pay an agency for. In minutes.

🛠️ Tim Dalrimple’s guide to Marketing Attribution

Tim Dalrimple shares his playbook for marketing attribution in Growth Unhinged.

tl;dr:

Attribution isn’t a nice-to-have, and “our data’s messy” isn’t an excuse. It’s a risk.

We need:

• to set campaign-level attribution on every key growth metric. Not just leads.

• to have it across channels. Paid, organic, outbound, partner—everything.

• to own the model. Don’t rely on default tools. Customize it to fit your business and buying motion.

• to push back when someone says “that’s not possible with our data.” It is. It just takes effort.

• to have the right breakdowns. Campaign data is a start. But real insights come from layering internal segments—persona, deal size, velocity.

To get started, audit your stack, pick your metrics, and get attribution in place.

📅 Bringing your Google Workspace context to Gemini and Claude

Models need context to create relevant output. For many of us, a lot of that context is in our G-Mail, Google Calendar, Drive, and Docs.

Good thing: we can now give both Gemini and Claude access to our Google Workspace - and not just to docs one by one (see screenshots below to activate that in settings).

Oddly, Claude extracted a lot more context from my calendar than Gemini. It was able to analyze with whom I spent my time in April 2025 vs 2024, with what companies and get a report including pie charts and findings. And that’s just one use case…

Final Words

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

François | LinkedIn 

I'm a CMO, advisor, and "CMO Wingman." Yes, that's a thing :-). Ask my clients: in this AI era, 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.