A practical playbook
How to build an SEO content flow that keeps you and real data in the loop.
Search content matters and never gets the time. This playbook runs the legwork and keeps every piece tied to your strategy.
The problem
Why SEO stalls
SEO has gone from essential to easy to neglect.
Then paid media offered a faster route. You could buy visibility, see results quickly and shift budget when something stopped working. SEO kept going in the background.
AI has changed the conversation. People now use search and AI tools to compare products, understand options and build a shortlist before they speak to a company. Being visible at that point matters. But there's no shortcut that guarantees it. You still need useful, accurate content that answers the question well, and evidence of that value beyond your own website.
The problem is that most teams don't have someone whose job is to keep that work moving. An agency can research the opportunity and define the technical fixes. Then comes the harder part: who writes the content, checks it, publishes it and looks at what happened next?
I've seen strong articles start to climb, then sit untouched for months. The person who owned the work moves on. Other priorities take over. Later, someone suggests, “We should write a blog about this,” without checking what already performs or whether a new article could compete with it.
A simple, repeatable process can help. Put time in the calendar to review a data-backed brief. Add your own angle, experience and evidence. Review the draft, give feedback, then approve the finished page and its social post.
That could mean around 90 minutes a month for one article, or three hours for two. The research is ready when you need it, but the judgement stays with you. The strategy gets reviewed each quarter, so it can respond as the evidence changes.
SEO doesn't usually fail because teams lack ideas. It fails because nobody has the time to carry those ideas through.
How it works
Four steps, then it repeats
You make four calls, all in the first two steps: the brief, then the draft, the images and the proof. The agents do the rest.
- 01
Brief
Search data flags a gap. You brief the piece in a voice note.
- 02
Approve
You approve the draft, pick the images and sign off the proof.
- 03
Agent publishes
Live on your site, linked into its cluster. Social posts scheduled and going out.
- 04
Search validation
Rankings, AI citations and demo requests feed the next brief.
↺ Results feed the next brief.
Prerequisites
What you need first
Strategy is the foundation. The engine doesn't set it. It follows it, and shows you where it's working.
- 01
A search strategy you've validated. Which terms drive volume, which signal intent to buy and which build brand visibility. Keep those buckets separate.
- 02
A tone of voice guide. How you talk, what you do and what you never do.
- 03
A site the agent can publish to. A CMS it can connect to, a page template you've validated and clear content clusters.
- 04
Access to your social scheduling tool.
- 05
Tracking that maps to the goal. Demo requests, brand search, AI citations. If the data can't show it, fix that first.
Under the hood
The same flow, in six stages
For the team that builds it. Each stage hands a checked output to the next. The routine work runs on a schedule. A person signs off at four points.
System diagram
How the content engine works
- Trigger
- Data collection
- Agent
- Human review
- Executive dashboard
Executive dashboard
Weekly view of traffic, search, citations and demos
Executive dashboard
Weekly view of traffic, search, citations and demos
-
Weekly data
Pulled, cleaned and checked
-
Data checks
Records what changed, ready or blocked
- Shapes brief
Strategy agent
One topic a month, from search results
- Approves draft & images
Draft and preview
Written to your guide, checked in private
- Approves proof
Publish
Linked, checked, released
- Approves posts
Social
Tagged posts with the live URL
Results flow back tagged to each piece, so every brief learns from the last
Every piece gets an ID when it's recommended. The same ID goes on the live page, every conversion event and every social link. Plain code pulls and checks the numbers. A language model never decides what they are.
Design choices
Why it's built this way
It measures traction, not output. Each cycle shows which pieces rank, which get cited in AI answers and which lead to a demo request. Not only revenue this month. The next recommendation leans on what's working, for search and for AI answers alike.
Agents take the busywork. People keep the real work. Around 30% of a working day goes on the job you were hired for. The other 70% goes on everything around it: pulling numbers, formatting, publishing, linking, scheduling, chasing. Agents take that repeatable 70%, so the brief, the draft, the images and the proof get a person's full attention.
Context lives outside the model. The strategy, tone guide, data models, publishing and deployment flows, and the code are all documented in one workspace, Notion here. Every agent reads from it, so each one can run on the model that fits its job, balancing tokens and cost, and you can switch tools at any point. Run out of credits halfway through a task? Point another tool at the same pages: here's the system, follow the path.
One source, many hands, guardrails built in. The same pages work for people. Save a prompt, an action or a task as a page, and a colleague can share the link and say “run task 1”. The documentation does the rest. Core pages are read-only, and every change goes through a log page, so no tool or person can pull the system off course.
The stack
The tools behind it
Any comparable stack works. This is the one behind this playbook.
Context
Notion holds the strategy, tone guide, image gallery, data models, and the conversion and structured-data requirements. Every agent knows what to fetch.
Data
Google Analytics, Google Search Console, DataForSEO, and Metricool for social results.
Agents
Claude runs the data, strategy and brief agents, scheduled in the cloud on GitHub Actions.
Publishing
Codex and GitHub, into Sanity as the CMS.
Social
Claude schedules posts into Metricool for Facebook, Instagram, LinkedIn and YouTube.
What breaks
Three failures to design for
An automated deploy ships an old version. Publish only an exact build you've verified, from clean source, and lock production between releases.
An agent reports a fix as live when it isn't. Built and tested is not deployed. Check the live system before you believe "done".
A data job fails silently for weeks. It can log success while pulling nothing. Give every job a real exit status, timeout, retry and alert.
What's next
Extending it
The same shape takes more channels. Paid media, email or CRM become new lanes alongside content and social. Each lane acts on the same checked data and sends results back into the same data layer, never directly to another agent. Extend the join key as you go: content ID today, campaign and contact IDs when those channels arrive. That shared data layer is the same one the Revenue Command Centre reads from, so content results sit next to pipeline and revenue.
FAQ
Questions
How much of my time does it take?
Around 90 minutes a month per article, across four short decisions.
What happens when the data is wrong?
The checks block it and the strategy agent stops. A missed cycle costs less than a confident brief built on a broken number.
Why aren't images automated?
They can be, later. Choosing the image that fits takes two minutes and it's the first thing a reader sees. Start manual, build a pool, then let an agent suggest from it.