A ready AI prompt for every SEO report card

Short answer. A good analysis prompt has three parts: a role scoped to the data in front of it, rules that forbid guessing (every number with source and period, rates as percentages, small samples named as such, no settings changes), and an output format (findings → actions → verification). seops attaches such a prompt to every report card; one click copies the card's data plus the prompt into Claude, ChatGPT or Gemini. Save your variants to a library and reuse them over MCP.

The failure mode

Paste a Search Console table into an assistant and ask “what should I do?” and you’ll get a confident list: write more content, build links, improve CTR. None of it is wrong and none of it is from your data. The model wasn’t told what counts as evidence, so it used general knowledge.

What a card prompt contains

Every report card in seops has an AI prompt button. It copies the card’s data and a prompt built from three blocks.

Role, scoped to the card:

You are the SEO and web-analytics analyst for this panel. Your data source is the table below and, if the seops connector is available, the same report over MCP.

Rules, the part that does the work:

• No guessing. Every number you cite comes with its source and period.
• Rates arrive as 0–1; show them as percentages.
• Under 50 clicks or 500 impressions in 28 days is a small sample; say so.
• Don’t propose actions that conflict with the site’s topic or content policy (unverifiable claims, misleading comparisons).
• You cannot change panel, Cloudflare or GA4 settings; give copy-paste text instead.

Output format: findings (with numbers), actions (owner: user or ai), and a verification line for each action. If the seops connector is present, open the actions as tasks with add_task.

The rules are the same across cards; the “requested analysis” line differs. For the opportunity-queries card it asks which queries at position 4–20 have the most impressions and which page to improve; for the AI-visibility card it asks which prompts are cited by Perplexity but not ChatGPT and what the cited sites have in common.

Why a library

After a month you’ll have changed the wording: your sites have a tone, your clients have a vocabulary, some cards need a stricter sample threshold. The library keeps those variants:

• Scope: this card type only, or every card.
• Visibility: your workspace; every site in it sees the template.
• Source: written by you, or drafted by the AI from a description (through AI Gateway, costs a credit).
• MCP: list_prompt_templates and save_prompt_template, so an agent can pick the right template without you pasting anything.

Starter prompts for the empty states

Two cards are empty on day one: rank tracking (no keywords) and AI visibility (no prompts). Each empty state has a starter prompt that asks the assistant to build the list from Search Console data and brand terms, returning it in the format the Add dialog accepts. That is how most workspaces get their first 30 keywords and 15 prompts in five minutes.

Prompt history

Every run the agent makes ends with log_prompt: the request, a technical one-liner, the summary, tasks and commits. The history page is where you see which prompts produce useful work and which produce noise, and where next week’s session picks up context.

Try it

Open any card in the demo, press AI prompt, paste into your assistant. Then compare with the same question asked without the prompt. The difference is the rules.

Fragen