Inside the AI System That Plans a Client’s Year of SEO
Every new SEO client used to start the same way: a strategist spending days pulling exports, sorting keywords, guessing at a page plan and rebuilding a task board by hand. Multiply that by dozens of clients and you have a team that spends its best hours on setup instead of judgment.
So I built a system that does the setup. A strategist feeds it a client’s intake form, crawl export and keyword data, and it returns a scored opportunity list, a tracked keyword set, a 12-month page plan sized to the client’s package and a task board ready to import. Here is how it works and what I learned building it.
What goes in
- The intake form: services, cities, differentiators and anything the client has confirmed about their business.
- A site crawl: every URL, title, heading and status code.
- Keyword exports: current rankings, competitor gaps and local search volume.
- The live site itself. More on that below.
What the system does
- Normalizes the data so exports from different tools speak the same language.
- Scores every opportunity by intent, local relevance, difficulty and whether an existing page could win it with a rewrite instead of a new build.
- Picks the tracking set: the keywords that actually represent revenue for this business, tagged by service and city.
- Plans the pages month by month, within the number of pages the client’s package includes.
- Builds the task board with owners and due dates, ready to import into the project management tool.
Then a strategist reads the top of the scored list and makes the real decisions. The system’s plan is a starting grid, not a strategy.
The rules I wrote into it
The most useful part of the system isn’t the scoring. It’s the rules it won’t break.
- Never invent client facts. Prices, guarantees, licenses, years in business, response times. If a page needs one and the client hasn’t confirmed it, the system writes a confirmation flag instead of a guess. A made-up certification on a live page is a liability, not content.
- Never exceed the package. Page counts live in one configuration file. Overdelivering quietly becomes next year’s expectation, and when policy changes, it changes once for every client.
- Always check the live site. Exports miss things. Checking rendered pages has surfaced links pointing at a staging domain, title tags naming the wrong city and tenure claims nobody could back up.
- Forecast in ranges. Every projection is anchored to the client’s real baseline. No promised positions.
What it catches that people miss
Because the system looks at everything every time, it finds the boring problems that quietly cap results: a homepage indexed at three different URLs, five pages competing for the same topic, a rank tracker set to national volume for a business that only serves one county, and a brand-new site that launched with search engines blocked. None of these are exciting. All of them matter more than the next blog post.
What stays human
- Choosing which opportunities fit the client’s business goals.
- Rewriting titles and keywords with real decisions before anything reaches a client.
- Every claim on every page, approved by a person.
- The relationship. No system explains a plan to an owner as well as someone who understands their business.
If you want to build your own
- Write down your service tiers first. AI can’t systematize a package that isn’t defined.
- Pick the single most repeated setup task and automate only that.
- Put your rules in data, not in someone’s memory.
- Make the system flag unknowns instead of filling them in.
- Keep a person accountable for every output that reaches a client.
The payoff isn’t that AI writes the strategy. It’s that strategists get their week back to do the part only they can do.
More on the systems I build: AI Systems.