It’s been 1,259 days since OpenAI launched ChatGPT on November 30, 2022.
And I still get asked the same question:
“Are you actually using this stuff? Or just talking about it?”
The answer is yes, yes, we are genuinely using it.
But probably not in the way most AI content online suggests.
As someone leading marketing within a strategy-focused firm, I’m less interested in AI as a trend and more interested in whether it actually improves execution, thinking, and team workflows in a meaningful way.
And when we decided to integrate AI into our workflows, my team didn’t transform overnight.
We ran experiments, dropped tools that didn’t fit, and slowly built workflows that actually stuck.
And even now, we’re constantly iterating, refining workflows, and adapting as the tools continue to evolve.
This Review is a record of exactly what those workflows look like today.
No predictions. No hype.
Just what we do.
Here’s what you’ll learn:
- The exact tools on our stack and the specific job each one does
- Our 3-step writing workflow (with the real prompts we use)
- How we generate 25 ad variations in under 10 minutes
- Our modular video system for short-form and long-form content
Let’s get into it.
The AI Tools We Actually Use in Our Workflow (And Why They are Still Relevant)
Over the last three years, I’ve added a lot of AI tools to our stack.
Most didn’t last.
These are the ones that did, the ones that are open on someone’s screen pretty much every day.
| Writing and Research | Visuals | Video and Audio |
| ChatGPT Perplexity Claude | Midjourney Freepik AI / Canva AI ChatGPT image generation | Runway Sora / Kling / Pika ElevenLabs |
I’m not listing these to overwhelm you.
I’m listing the tools because each one has a specific job in a specific workflow.
Tools without a purpose don’t last on our stack.
Our AI Workflow for Writing
This is the workflow we use most consistently.
It covers blogs, social captions, and ad copy.
Content Writing Workflow
This is how we go about creating our content…
Step #1: Research
We split the research across two tools intentionally.
Perplexity for facts and ChatGPT for thinking.
While Perplexity helps us scan sources and verify claims faster, ChatGPT helps explore angles, find counterarguments, and surface subtopics we might have missed.
Here’s the exact prompt I use at this stage:
Step #2: Outline
Once we have a direction, we generate an outline.
But we don’t use a generic ‘write me an outline’ prompt.
That produces generic outlines.
We give it structure instructions instead.
Here’s how I’d do it:
Step #3: First Draft
This is where we switch to Claude.
Claude (Opus 4.6) follows prompt constraints more precisely than other tools for this specific task and produces a cleaner, more structured output when provided with detailed formatting instructions like:
Step #4: Editing
The output becomes our working draft.
A human editor from the team, rewrites anything that sounds off, adds real examples, and cuts anything generic.
The draft saves us from starting at zero.
The editing is where the actual content gets made.
Our Ad Copy Workflow
Campaign testing needs volume.
You can’t test one headline when running ads.
You need 20–30 variations to find what performs.
Before AI, writing 30 headlines manually was a half-day task.
Now it’s 10 minutes.
74% of marketers are now using at least one AI tool at work — up from 35% just a year earlier.
HubSpot, 2024
Here’s my current approach to ad copy generation:
From those 25, I pick 8–10 to refine manually.
Cut them down to 5 strong options and then launch the test.
AI simply solves the volume problem.
The strategy: knowing what to test and why, still comes from me and the team.
Our AI Workflow for Image Creation
Two years ago, our team was either paying for stock photos or waiting on a designer.
Now we generate base visuals ourselves using Canva AI and Freepik AI for social graphics and templates and Midjourney for campaign imagery
Design review is now for refinement, not creation.
Our current workflow:
- Describe what you need
- Generate 4-6 variations
- Pick the strongest concept
- Refine in Canva or pass to the designer for final polish
It seems pretty straightforward, but one of the most common aspects that people get wrong is being too vague.
Asking simply for “A marketing dashboard” more often than not will generate garbage.
Specific beats general every time.
So, we try something like this:
Pro Tip: For blog featured images, go conceptual, not literal. If the article is about AI in marketing, don’t generate a robot holding a megaphone. Generate something that communicates the feeling of the topic: efficiency, scale, clarity.
Our AI Workflow for Video Production
Video production took the longest to build into a real workflow.
The tools were taking too long to become reliable.
That changed in the last 12 months.
We now produce short-form video for social and longer structured video for YouTube using AI generation as the base layer.
The main tool we use is Runway, which recently released its Gen-4 model with significantly improved visual consistency.
We currently follow a two-pronged approach to create visually consistent AI-generated videos:
1. The Modular Approach
We never try to generate a full video in one pass.
AI video tools perform better on short clips, as shorter durations make it easier to maintain visual consistency, smooth motion, accurate context, and overall output quality.
60 to 90 seconds is our sweet spot.
So we plan every video in modules.
A typical 4 to 5-minute tutorial looks something like this:
- Clip 1 (60–90s): Introduction, what the video covers and why it matters
- Clip 2 (60–90s): The core concept explained
- Clip 3 (60–90s): A real example of the concept in use
- Clip 4 (60–90s): Conclusion and takeaway
Each clip is generated separately. Then assembled.
The same clips get cut individually for Reels, Shorts, and TikTok.
One production run. Multiple formats.
2. The First Frame / Last Frame Approach
This is the technique that improved our visual consistency the most.
Instead of letting the model decide where to begin and end, we define it.
Step #1: Create the opening frame
Generate or select what the clip looks like at second 0.
Be specific…a marketing dashboard, a person at their desk, a conceptual animation scene.
Step #2: Create the ending frame
Generate the final frame separately.
This defines how the motion should end.
Step #3: Generate the Video
Feed both frames into Runway or Kling.
The model generates the motion between them.
This gives us control over continuity between clips.
When clip 1 ends on a specific visual and clip 2 starts from a matching frame, the assembled video feels coherent, not like random clips stitched together.
Voiceovers
ElevenLabs handles all our voiceovers.
We write the script first, a stripped-down version of the blog or briefing doc, then generate the audio.
Pro Tip: Use one consistent voice across all videos. Consistency builds recognition faster than perfection.
What AI Workflow Tools Changed (And What Didn’t)
The biggest change isn’t quality.
It’s speed at the front end of production.
Before AI workflow tools, every piece of content started at zero.
My team would open a blank doc and spend the first hour just getting to something workable.
That hour is mostly gone now.
We now start with a generated draft or outline and immediately move into editing mode.
91% of marketing leaders say their teams now use AI to assist in their jobs.
HubSpot
What hasn’t changed: judgment, positioning, and editing.
AI does not decide what we say or why.
It handles the mechanical first pass.
We handle everything that requires understanding our audience, our brand, and our strategy.
The honest number: AI saves us roughly 40–60% of the time on early-stage production.
The back half, i.e., editing, refinement, distribution strategy, still takes the same human attention it always did.
What’s Coming Next in This Series
This Review covered production: how we create content faster without sacrificing quality at the editing stage.
The next Review covers distribution.
Specifically: how AI is changing SEO and GEO (Generative Engine Optimization), why being cited in AI answers is becoming as important as ranking in search, and what we’re actually doing differently because of it.
If there’s a specific workflow you want me to go deeper on, prompts, tool choices, how we measure output quality, leave a comment or reach out directly.
I’ll work it into the series.
