Position eleven and position forty need different work
SEO tools display opportunities by volume, and that volume offers no indication of the effort involved. A page that sits just below the first page of search results might need only one internal link, while a page much lower in the results will need a full rewrite. When grouped by volume, they look like the same kind of task.
Diagnosis is what distinguishes them. Intent mismatch, inadequate coverage, poor internal linking, and a slow page are four separate problems, each with its own cost. You distinguish them by comparing the page with the ones ranking above it.
The only reasonable way to order the entries is by expected return. A cheap fix on a moderate-volume term beats an expensive rewrite on a high-volume one. The volume column hides that fact on purpose.
A list longer than a sprint is demotivating. Completing ten tasks is better than admiring fifty.
How the agent is built in Agent Studio
The system prompt defines the diagnostic categories, specifies the failure modes you want detected, and requires that you assign one to each recommendation.
Prompt skills reflect your site's structure and content standards, so internal link recommendations point to existing pages that align with your site's organization.
A Start Scheduled flow gathers the inputs. The HTTP Request node retrieves the position data. The Scrappey node retrieves your page and the pages ranked above it. The Agent Chat node then passes the collection along.
Notion receives an action list, arranged by expected return and limited by what a team can accomplish in a single sprint. Each item includes a diagnosis, so the writer does not need to reanalyze.
Google Sheets keeps a record of which pages have been acted on and when. The only fair way to assess whether the diagnoses were correct is to compare the pages acted on with those not acted on over a quarter.
Durable sessions let an SEO lead question a diagnosis and have the recommendation amended rather than started over from scratch.
What this agent is built from
- System prompt: states the failure modes to distinguish, and a named diagnosis for every recommendation.
- Prompt skills: carry your site structure and content standards so link suggestions are real.
- HTTP Request (Util): pulls position data.
- Scrappey (Util): retrieves your page and the pages ranking above it for comparison.
- Agent Chat (Util): a scheduled flow hands the set over for diagnosis.
- Notion (Integration): receives an action list sorted by expected return, capped at one sprint.
- When an ActionFlow is enough: A position-change report is a workflow. Diagnosis is the agent.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.