# Weekly SEO plan: a routine for the AI agent

If you are a person reading this: this page is written for your AI agent. Once a week, paste this line into Claude, ChatGPT or any MCP client:

> `Read https://kafkai.com/docs/weekly-seo-plan.md and run it on my site.`

There are two other ways to start it:

- **As a prompt.** If your client lists the prompts of the Kafkai
  connection, choose `kafkai_weekly_seo_plan`.
- **As a skill.** If your client supports skills, install
  [the skill file](https://kafkai.com/docs/weekly-seo-plan/skill.zip)
  once, then ask for "my weekly SEO plan".

The rest of this page is for the agent.

## Your job

A Kafkai customer asked for their weekly SEO plan. Read the data Kafkai
holds for their project, then give three actions for this week, in
order. Reply in the language the customer writes in. Ask one question at
a time.

If you have no tools whose names start with `kafkai_`, go to
[Without Kafkai](#without-kafkai).

## Before you start

1. Call `kafkai_list_projects` (free). If there is more than one
   project, ask which one. If there is none, send the customer to
   https://kafkai.com/docs/onboarding.md and stop.
2. Call `kafkai_get_project_summary` (free). Read `last_research`, the
   competitor list, `backlinks` and `search_console.status`.
3. If `last_research.status` is `pending`, research is still running.
   Say so and stop. Run the plan when it has finished.
4. Check the age of the data. Rankings, keywords and backlinks are as
   old as the last full research run: `last_research.completed_at` when
   `last_research.kind` is `update` and `last_research.status` is
   `success`. If the last run is `failed` or `partial`, or its kind is
   `competitor_add`, that date is not the date of the stored data: tell
   the customer the data may be older. If that date is more than 30 days
   ago, tell the customer the date and offer an update. An update costs
   500 credits for the site plus 350
   credits for each competitor. State the total for this project.
   Never call `kafkai_update_keywords` before the customer agrees to that total.
   If the customer says no, continue with the stored data.
5. Ask which keywords to watch on page one: at most 5, and the same ones
   every week. If the customer has no list, run Step 2 after Step 3:
   propose 5 from the results of Steps 1 and 3, and ask the customer to
   confirm them before you call `kafkai_check_search_results`.
6. State the cost and wait for a yes. When every step runs, the plan
   costs about 43 credits. Leave out the steps you
   will skip. Call `kafkai_get_credit_balance` (free). If the balance is
   too low, point to https://kafkai.com/en/pricing/ and stop.

## The steps

Each step says what it costs and when to skip it. For a skipped step,
write one line in the report: what was skipped and how to turn it on.

### Step 1: What changed in the rankings?

Run this step only when a full research run finished in the last 7
days. Without a new run the answer is the same as last week: skip it and
give the date of the data.

Call `kafkai_get_ranking_changes` with `days=7` and `limit=20`. Read the
`summary` first, then the biggest moves. For each important move, say
what the likely reason is, and say when it is a guess.

Cost: about 4 credits.

### Step 2: Who is on page one?

Call `kafkai_check_search_results` with the keywords from "Before you
start". Kafkai fetches each page fresh, so this step is new every week.
Report where the customer's site is, and read `changes`: who entered,
who left, who moved. `changes` is `null` on the first check of a
keyword. A `null` domain means "unknown", not "absent".

Cost: 1 credit per call plus
5 credits per keyword fetched fresh. With
5 keywords: at most 26 credits.

### Step 3: What to write next?

Call `kafkai_get_catchup_keywords` with `limit=20`: keywords competitors
rank for and the customer does not. Skip this call when the project has
no competitor. Then call `kafkai_get_complement_keywords` with
`limit=20`: related keywords nobody in the project targets yet.

Pick at most three keywords to write about next. Prefer high search
volume with low difficulty. Leave out keywords the customer has already
written about. Ask if you do not know.

Cost: about 6 credits.

### Step 4: Which links were gained and lost?

Run this step only when a full research run or a backlink update
finished in the last 7 days. Otherwise skip it and give the date of the
data. Never call `kafkai_update_backlinks` unless the customer asks and agrees to its price.

Call `kafkai_get_backlink_changes` with `competitor_domain="all"` and
`days=7`. Read gains and losses from `period_totals`. Do not subtract
one total from another. Then call `kafkai_get_link_gaps` with
`limit=20`: sites that link to two or more competitors but not to the
customer. Name at most three sites worth contacting.

Cost: about 4 credits.

### Step 5: What to fix first? (Search Console)

Run this step only when `search_console.status` is `ready`. Otherwise
skip it and point to https://kafkai.com/en/docs/google-search-console/.

Call `kafkai_update_search_console` (free, at most one fetch a day). If
it queued a fetch, wait about a minute. Then call
`kafkai_get_search_console_insights` with `limit=5`. The limit applies
to each kind of flag. Start with `striking_distance` and `ctr_gap`: they
are the fastest wins.

Cost: about 3 credits.

### Step 6: The report

Write a short report in this order:

1. **Three actions for this week, most important first.** For each: what
   to do, on which page or keyword, and the number that supports it.
2. **What changed**: rankings, page one, links. One or two lines each.
3. **Skipped steps**, one line each, with how to turn each one on.
4. **Data dates and cost**: the date of the last research run, and the
   credits this run used (each response reports them in `_usage` or
   `billing`).

Offer to write the first action's article or page change. Do not start
it unasked.

## Without Kafkai

Steps 2, 5 and 6 work without Kafkai, with more manual work:

- **Step 2:** search the web for each keyword and list the sites on
  page one. Keep the list, so next week you can compare.
- **Step 5:** ask the customer for a Search Console export of queries
  for the last 28 days and the 28 days before. Look for queries at
  positions 8 to 20 and for queries with many impressions and few clicks.
- **Step 6:** write the report from what you have.

Steps 1, 3 and 4 need Kafkai's stored data. To connect Kafkai, read
https://kafkai.com/docs/onboarding.md. To find a site's competitors for
free, use https://kafkai.com/en/find-competitors/.

## Rules

- Say the credit cost before every call that spends credits, and wait
  for the customer to agree.
- Read stored data only. A research update is the customer's decision.
- Skip a step whose data source is not connected. Never report a
  missing source as "no data".
- Quote every number with its date.
- Never ask the customer to paste a password or an API token into this
  conversation.
- If a tool returns an error, show the customer the message and point
  to https://kafkai.com/en/docs/troubleshooting/.
