Keep, update, consolidate, build.
The model is mapped against your site. Every existing page gets a decision; every missing fact gets a page only if it needs one. Then the fixes are specified down to the section, the schema and the sentence.
What an action plan is
An action plan is one run of Step 3 against one category: the prompts that matter, the gaps Model found, and the on-site work that closes them — page by page, with the evidence behind each recommendation and the impact we expect.
It is not a content calendar. It is the list of what should exist, what should change, and what should go.
Inside a plan
- Overview — the category, the topics and prompts covered, the models run, and the recommendation status.
- Components — each recommendation as a unit of work: a new page, a rewrite, a consolidation, a retirement, a schema change.
- Analytics and competitors — where you're absent, who's recommended instead, and why.
- Recommendations — the on-site fixes, prioritised, with expected impact and the evidence that supports each one.
- Evidence — the model outputs and citations the recommendation is based on.
Example, shared with permission: Pro Max Fence, commercial fencing — 10 topics, 100 prompts, five models. One component: a video comparing chain-link and welded-wire fencing for commercial properties, because AI systems answer that comparison with generic manufacturer content today; the plan specifies the installations to show, the cost comparison to state, and why a model would cite it.
Every page has a reason to exist — or a reason to go
| Decision | When | What you get |
|---|---|---|
| Keep | The page is right and complete | Nothing — page count is not the goal |
| Update | Right topic, thin or stale | The sections to add, the facts to state, the schema to emit |
| Consolidate | Two pages compete for one intent | Which page becomes canonical and what moves into it |
| Build | A fact with no home | A page brief: audience, sections, unique data, proof, schema, CTA |
| Retire | Nothing true left on it | A redirect or removal recommendation |
Illustrative: keep 112 · update 24 · new required 9 · retire 3.
Templates, schema and implementation specs
On Agency, every recommendation ships with what a team needs to execute it without a second conversation: page templates by type, the schema to emit (Organization, Product, Person, FAQPage, Article, and what each property must match on the page), and implementation specs a developer or writer can work from. Freshness alerts tell you when a page's facts have drifted from the model.
Your team or ours
Self-serve: you run the plan. Managed program: our team writes, structures and marks up the pages, and pushes one verified change to every page, schema object and FAQ it touches. In both cases, publishing is a person's decision — agents draft, check, compare and recommend; people approve and publish.
Pro: 5 action-plan runs a month, up to 50 fixes per plan. Agency: 15 runs a month, plus templates, schema and implementation specs, freshness alerts and benchmarking. Managed program: our team runs and executes the plans.
Questions
Start with the score.
Enter your website. In 24 hours: your GEO Score, how each model describes you, where you're missing, and the first three fixes.