AI is more useful when the work has a place to go
Most people use AI for marketing like this: open a chat, ask for an SEO audit, get one giant answer, copy a few lines, then lose the thread. That works for quick ideas. It does not work as a real marketing workflow.
In this video, I wanted to show a different approach. I used Codex by OpenAI and ChatGPT inside Hermes Agent by Nous Research to create a marketing plan inside an actual project workspace. This is using their GPT 5.5 model, with Fast & Thinking mode, with reasoning on a mixture of High and Max. This uses my pre-existing subscription, there is no additional cost. I covered how to connect Codex to Hermes Agent in a previous video here.
The goal was not to get one perfect answer. The goal was to organize the work so the research, files, notes, and final plan all had a place to live. That is the difference. A chat answer disappears into the scroll. A project folder becomes a working system.
What the video covers
The walkthrough shows how to use Hermes Agent and Codex to build a marketing plan from scratch using a project folder structure.
The workflow includes:
- Creating a new project folder in Hermes Agent
- Using separate chat threads for different stages of the work
- Running a raw SEO and marketing audit
- Organizing research into markdown files
- Separating messy research from clean planning notes
- Building a 30-60-90 marketing plan from the organized files
- Keeping the project usable after the AI session ends
I used my own site, jasonpollakmarketing.com, as the demo target. That made the process more useful because I could show real marketing decisions instead of pretending with a random company.
Why project folders matter
The biggest problem with AI marketing work is not the first answer. The first answer is usually fine.
The problem is what happens next.
Where does the keyword research go? Where do the competitor notes go? Where do the page recommendations go? Where does the final plan live? What happens when you want to come back tomorrow and keep working?
That is where the project folder matters.
In the demo, I used a structure like this:
- Research Work
- Planning Workspace
- Final Deliverables
Research Work is where the messy audit lives. Crawls, notes, raw keyword ideas, competitor research, and anything else that might be useful later.
Planning Workspace is where the useful pieces get narrowed down. This is where the AI stops being a firehose and starts becoming a planning assistant.
Final Deliverables is where the finished work goes. In this case, that meant a 30-60-90 marketing plan.
The folder structure is simple, but it changes how the AI work feels. You are not just chatting anymore. You are building a workspace.
Why separate chat threads help
One thing I like about Hermes Agent is that the project can hold multiple conversations inside the same workspace. That means I can use one chat thread for raw research, another for keywords, another for page strategy, and another for the final plan. That keeps the work cleaner.
For example:
- One chat can focus on keyword research
- One chat can focus on page recommendations
- One chat can focus on competitor analysis
- One chat can focus on writing the final 30-60-90 plan
Each thread can still work from the same project files. The difference is that the conversation stays focused.
That matters because marketing work gets messy fast. SEO audits turn into keyword lists. Keyword lists turn into page ideas. Page ideas turn into content plans. Content plans turn into technical fixes. If all of that happens in one endless chat, it becomes hard to tell what matters.
Separate threads give the work some shape.
The important part: do not let AI make the strategy by itself
This is the part I care about most. I am not using Codex to replace strategy. I am using it to organize the raw work so strategy is easier. There is a big difference between these two prompts:
“Create a marketing plan for my website.”
And:
“Audit the site, save the raw research, separate the keywords from the page recommendations, show me the highest priority opportunities, and then help me turn those into a 30-60-90 plan.”
The second workflow gives you something you can inspect. You can see the sources. You can see the files. You can question the assumptions. You can decide what gets used and what gets ignored. That is the right way to use AI for marketing planning.
Use the model to speed up the research and organization. Do not outsource the judgment.
What the final marketing plan included
The final output was a 30-60-90 plan built from the project folder, not just a random generated answer. For my own site, the biggest takeaway was clear: the site already had strong proof, services, and positioning. The gap was structure. That means creating clearer service pages, better question-and-answer sections, stronger schema, and a cleaner connection between blog content and core service pages.
That is a much better outcome than a generic SEO checklist. The plan focused on:
- Technical SEO cleanup
- Metadata improvements
- FAQ and schema opportunities
- Service page structure
- Keyword mapping
- Competitor gaps
- Internal linking
- AI search visibility
- Content priorities
Where this Codex and Hermes Agent workflow fits
The key is that the AI is not just producing text. It is helping build the working environment around the project. That is what makes Hermes Agent interesting to me. The desktop project folder, the chat threads, the files, and the agent all sit in the same workspace. For marketing work, that is a better model than a blank chat box.
This kind of workflow is useful anywhere the work has multiple moving parts: SEO audits, marketing plans, content strategy, website restructuring, client onboarding, competitor research, agency deliverables, 30-60-90 planning, and white label marketing work. It gives the research somewhere to live, gives each decision a place in the process, and keeps the final deliverable from turning into another loose AI-generated document.
My takeaway on Codex, GPT 5.5 and Hermes Agent
The more I use AI agents like Hermes Agent for marketing work, the more I think the real advantage is organization. The model can research. It can summarize. It can draft. It can compare pages. It can build a plan. But the real value comes when the work is structured enough that you can keep using it. That is what this video is about.
Not just “AI made a marketing plan.”
More like:
“Here is how to turn AI research into a usable marketing workspace.”
That is the part most people skip. It is also the part that makes the work useful after the video ends.
FAQ
Can Codex create a full marketing plan by itself?
It can draft one, but I would not treat that as the final strategy. The better use is to have Codex and GPT 5.5 gather research in a project folder, organize files, identify opportunities, and help you turn the work into a plan you can review.
Why use Hermes Agent instead of a normal chat interface?
Hermes Agent gives the work a project structure. It also has access to your machine. You can keep files, folders, chat threads, and deliverables together instead of losing everything inside one long conversation. This is not a regular LLM platform where you just create a project folder in ChatGPT, for example. This is a project folder that lives on your machine.
Why separate raw research from planning files and can this workflow be used for real client work?
Raw research is supposed to be messy. Planning files should be cleaner. Keeping them separate makes it easier to check sources without dragging every note into the final plan. The same structure works for client SEO audits, content strategy, competitor research, and 30-60-90 marketing plans. For client work, I would keep raw research, planning notes, and final deliverables in separate folders.