Example Prompts

Last updated: September 7, 2026

Kafkai hands your site's data to your agent. It does not write, and it does not decide. The agent you connect does both.

The prompts on this page are built around that split. Three things they have in common:

  1. Name the project (the site).
  2. Ask for a decision, not a list. Do not stop at "show me"; say "decide", "rank them", "tell me which".
  3. Say what the result is for: a draft, a report, a post published as a draft.

Read example.com as your own project and competitor-a.com as a competitor already registered in it. Tool calls that return data spend credits. Checking your balance is free.

Some examples need a connection set up first: Search Console, analytics, or a publishing destination. Each of those examples says so above its prompt, so nothing fails by surprise.

A. Decide what to write next

Example 1: Pick the next three articles, with evidence

For the example.com project, propose the next three articles to write, in priority order. For each one, show which 4C strategy and which keyword it rests on, and which competitor page it has to beat.

What the agent does: reads the project summary and the strategy with the most opportunities (recommended_strategy), then pulls that strategy's keywords. Catch-up keywords come with the competitor that ranks, its position and its page title. The agent weighs volume, difficulty and the competing page, and narrows to three.

What you get: three topics. Each carries its keyword, its strategy, and the URL and title of the page to beat.

Example 2: Choose winnable keywords, checked against the live results

List 10 keywords where competitor-a.com ranks and example.com does not, with real search volume and low difficulty. For the top 3, check today's search results and tell me what kind of article would win.

What the agent does: pulls Catch-up keywords and filters by volume and difficulty. For the top three it reads the per-keyword SERP data: who ranks where, the titles and descriptions of the top pages, and their page types (article, product page, listing).

What you get: a table of 10, and a reading for 3. Something like "the top results are all comparison posts, so a step-by-step guide stands apart".

Example 3: Choose one topic a single article can cover, from clusters

Read the keyword clusters for example.com and pick one topic that a single article can cover in full. Build the heading outline so that every keyword in the cluster gets a heading.

What the agent does: pulls the semantic clusters, each with its average volume, difficulty and member keywords. It picks one and maps the members onto headings.

What you get: an outline for one article, with the keyword and volume behind each heading.

Example 4: Keep only the untouched topics that fit the site

From the Complement keywords (the ones neither example.com nor its competitors target yet), keep the ones that fit example.com's niche. Drop the rest, and give a reason for each one you drop.

What the agent does: reads the niche and site description from the project summary, then pulls Complement keywords and tests each one against the niche.

What you get: a keep list and a drop list. Every drop has a reason.

Example 5: For one keyword, decide between writing and fixing

For "[keyword]", tell me who ranks where today, where example.com and the competitors sit, and what moved recently. Decide whether we should write something new or fix an existing page.

What the agent does: reads the per-keyword SERP data: the ranking domains and pages, your position, the competitors' positions, and the change since the last measurement. If your site already ranks, it also reads the Search Console page rows to see that page's impressions and click-through rate.

What you get: one of "write new" or "fix existing", with the reason. For a fix, it names the page and what to change.

B. Read the competition

Example 6: Target the keywords a competitor is losing where you sit close behind

Of the keywords competitor-a.com lost positions on in the last 90 days, list the ones where example.com sits just behind. For each, suggest what to add to example.com's page to overtake.

What the agent does: pulls the competitor's ranking changes for 90 days, movement "declined". Each row carries the competitor's current and baseline positions with their dates, plus your own rank on the same keyword. It keeps the small gaps and compares the two pages.

What you get: a target list. Each keyword shows how far the competitor fell, where you sit now, and what to add.

Example 7: Read where every competitor is investing now

Across all registered competitors, list the keywords they started ranking for in the last three months. For each competitor, tell me which topics they have started putting effort into.

What the agent does: pulls ranking changes for "all" competitors, movement "new". The response includes a per-competitor breakdown. The agent groups the new keywords by topic.

What you get: the topics each competitor has begun to take, with the ones you have not written about marked.

Example 8: Find what a competitor's linking sites have in common

Read competitor-a.com's backlinks and tell me what the linking sites have in common. What kind of page would example.com need to get mentioned by the same kind of site?

What the agent does: reads the competitor's backlinks, strongest referring domain first. Each row carries the referring domain and its strength, the target page, the anchor text and the platform type. The profile totals say how large the whole profile is and how much of it the rows cover.

What you get: the pattern in the linking sites (trade media, personal blogs, tool directories) and the kind of competitor page that collects the links, with a matching page proposal for your site.

Example 9: Compare link-building pace across the project

Line up the backlink gains and losses for example.com and every competitor over the last six months. Which competitor is accelerating? Is example.com keeping pace?

What the agent does: pulls backlink changes for "all" sites over 180 days. Each site row carries new, lost and net counts for the period, and a month-by-month breakdown.

What you get: a table per site and a reading of the monthly trend, such as "competitor-b.com's net gain tripled from April".

Example 10: Find competitors you have not registered (without adding them)

Find competitors that example.com does not track yet. Pick three direct competitors of a size example.com can win against, and tell me whether to add each one and why. Do not add any yet.

What the agent does: pulls competitor candidates, built from the keywords your site actually ranks for. Each candidate carries its relationship (direct, adjacent), winnability, shared keyword count and overlap share. The agent keeps direct, winnable ones.

What you get: three candidates with reasons. Adding one is a separate paid step, so the agent waits for your word.

C. See what happened after you published

Every example in this section reads Search Console, so setup comes first: verify the domain and grant Kafkai read access. Example 19 walks the agent through both steps.

Example 11: Queries just off page one, with the fix

Setup needed first: Search Console. Verify the domain and grant Kafkai read access (Example 19). Sent before that, this prompt returns the missing steps instead of the data.

From example.com's Search Console, list the queries that are just short of page one. For each affected page, give concrete changes to the title and the body.

What the agent does: pulls Search Console insights of type striking_distance: queries at positions 8 to 20 with real impressions, ranked by the extra clicks expected from reaching position 5. It then reads the page's numbers and the other queries that page appears for.

What you get: a fix per page: the target query, the current position, the expected clicks, the title to change and the content to add.

Example 12: Fix queries that rank well but do not get clicked

Setup needed first: Search Console. Verify the domain and grant Kafkai read access (Example 19). Sent before that, this prompt returns the missing steps instead of the data.

Find queries that rank well but are not getting clicked. Look at the current title and description of each page and write three alternatives for each.

What the agent does: pulls insights of type ctr_gap: queries whose click-through rate sits far below what that position normally earns on your site, ranked by clicks lost. That is a title and snippet problem, not a ranking problem. The agent reads the current title and description from the SERP data and rewrites them.

What you get: per query, the current title and description, three rewrites, and the clicks at stake.

Example 13: Merge pages that compete for the same query

Setup needed first: Search Console. Verify the domain and grant Kafkai read access (Example 19). Sent before that, this prompt returns the missing steps instead of the data.

Show me where two or more of example.com's pages compete for the same query. Decide which page to keep and which to fold in, with the reasoning.

What the agent does: pulls insights of type cannibalisation: one query split across two or more pages, each taking a real share of impressions, ranked by impressions at stake. The agent compares each page's clicks, position and other queries, and picks the keeper.

What you get: per query, the page to keep, the page to fold in, and why. This is the Consolidate strategy of the 4Cs in practice.

Example 14: Check whether last month's article reaches the intent you wrote for

Setup needed first: Search Console (Example 19), and analytics for the visits and traffic sources. Kafkai support connects analytics once the domain is verified. Without it, the agent answers from Search Console alone.

For "[article URL]" published last month, check the queries it has started to appear for in Search Console, and its visits and traffic sources in analytics. Judge whether it reaches the search intent we wrote it for.

What the agent does: reads the Search Console page rows filtered to that URL, then the new queries (new_query). With analytics connected, it also reads the page's visits and referrers. It sets the queries against the article's intent.

What you get: the queries the page appears for and where they diverge from the intent. If they diverge, how to change the title and the opening.

Example 15: Refresh rankings, then report the moves and who overtook you

Update the rankings for example.com. When it finishes, report what went up and what went down since the last measurement. For the keywords that fell, find out which competitor overtook us.

What the agent does: queues the update. It takes a few minutes, so the agent watches the project summary for completion. Then it reads your ranked keywords (with the improved, declined and new tallies) and the ranking changes with movement "declined". For each fallen keyword it checks the SERP data for who moved in.

What you get: the tallies, and per fallen keyword, who overtook you. Rankings refresh only when you ask, so this prompt works as a weekly or monthly routine as it is.

D. Find link opportunities

Example 16: Sites that link to two or more competitors but not to you

List the sites that link to two or more competitors but not to example.com. Order them by domain strength with low spam scores first, and for each write an approach: which of our pages to ask them to mention.

What the agent does: pulls the link gaps. One row per referring domain, with the link it gives each competitor (target page, anchor text). The agent looks at what competitor page gets mentioned and picks your matching page.

What you get: an outreach list. Per site: strength, spam score, what of the competitor it mentions, and which of your pages to offer.

Example 17: Links a competitor lost, where the source is still strong

Of the links competitor-a.com lost in the last year, list the ones whose source is still strong. What page should example.com build to be mentioned there instead, on the same topic?

What the agent does: reads the competitor's backlinks with status "lost". Each row names the source, the page it used to link to, and the date it was last seen. The agent filters by source strength and checks what the lost target page covered.

What you get: the sources and the topics of the lost target pages, with the page your site should build to match.

E. Start from nothing

Example 18: Create a project without competitors, then get the picture and candidates

Create a project for https://example.com targeting Japan. Do not register competitors yet. When the research finishes, show me the full picture of what example.com ranks for today, and the competitor candidates.

What the agent does: creates the project. The country is required; the agent never guesses it. Research takes a few minutes, so it watches the summary. Then it reads your ranked keywords (with top 3 and top 10 counts) and the competitor candidates.

What you get: the ranking picture and a candidate list to choose the first competitors from.

Example 19: Connect Search Console

I want Kafkai to read example.com's Search Console. Give me the steps in order, and issue the DNS record.

What the agent does: starts domain verification and returns the DNS TXT record to add (name and value). Then it gives the Search Console step and the reader address to add with Restricted permission. Once the record is in, you can ask the agent to run the check. Both tools are free.

What you get: a TXT record to copy and the Search Console steps. Once verified, the prompts in section C work.

F. Write the monthly report for a client

Every number Kafkai returns is dated, and every source sits in one project: keyword positions over time, Search Console, backlinks. So one prompt can assemble the report a project manager builds by hand each month. Kafkai returns the numbers; the agent writes the report. Rank history needs a point at the start of the period and one at the end, so run an update at both (Example 15).

Example 20: Write the monthly client report from every source

Setup needed first: Search Console (Example 19) for the traffic part of the report. Without it, the agent reports rankings and backlinks only.

Write the monthly report for example.com for [month]. Cover four things: which keyword positions improved and declined over the last 30 days and what that means for the site; Search Console clicks and impressions against the previous 28 days, with the queries and pages behind the change; backlinks gained and lost for example.com and each competitor; and what the competitors did. End with the recommendation for next month and why. Write it for a client who does not read data tables: one page, plain language, numbers only where they change a decision.

What the agent does: reads your ranked keywords (with the improved, declined and new tallies) and the ranking changes over 30 days. Then the Search Console totals with the change against the prior window, the query and page rows behind the change, and the new-query and decay insights. Then backlink changes for all sites over 30 days, and the competitors' ranking changes with movement "new" and "improved". It writes the report from those numbers.

What you get: a one-page monthly report in three parts: what changed, what it means, what to do next month. Every number carries its measurement date, so "as of when?" has an answer.

Example 21: Separate what is working from what is not

Setup needed first: Search Console. Verify the domain and grant Kafkai read access (Example 19). Sent before that, this prompt returns the missing steps instead of the data.

Look at example.com over the last 90 days and split its content into what is working and what is not. Working: pages whose clicks grew and keywords that moved up. Not working: pages whose clicks decayed, keywords we lost, and articles published this quarter that do not appear for any query yet. For each item that is not working, say whether the cause looks like ranking, click-through rate or competition, and what to do about it.

What the agent does: reads the Search Console page rows with their change fields, and the insights of type decay, ctr_gap and new_query. Then your ranking changes over 90 days with movement "declined" and "lost". For each lost or declined keyword it checks the competitors' ranking changes for a competitor that gained on the same keyword. It sorts the causes into the three kinds.

What you get: two lists. Every item on the not-working list carries its likely cause and the action.

Example 22: Explain a traffic drop to the client, with evidence

Setup needed first: Search Console. Verify the domain and grant Kafkai read access (Example 19). Sent before that, this prompt returns the missing steps instead of the data.

The client says traffic fell in [month]. Use the daily Search Console trend to find the day it turned. Then check which keywords lost positions around that date, whether a competitor gained on those same keywords, and whether example.com lost backlinks in the same weeks. Write the explanation for the client in plain words with the evidence, and say what we will do.

What the agent does: reads the Search Console trend view (the whole daily series) and finds the turning day. Then your ranking changes with movement "declined" and "lost", matching each measured_at against that day. Then the competitors' ranking changes with movement "improved", looking for the same keywords. Then your backlinks with status "lost", matching last_seen against the same weeks.

What you get: an explanation in order: the turning day, the keywords that moved around it, the competitor that gained, the links lost. When the evidence does not point to one cause, it says so and lists the candidates.

From writing to publishing

Kafkai does not write. The agent does. The agent can then hand the finished piece to Kafkai to publish.

Example 23: Write for the chosen keyword and publish as a draft

Setup needed first: a publishing destination. Ask the agent to add your WordPress site, with an application password, or a webhook. Without one, the agent writes the article but cannot publish it.

Write an article for the first keyword picked in Example 2. Before writing, read the top three ranking pages and decide the angle example.com can add that they lack. When done, publish it to WordPress as a draft and give me the URL.

What the agent does: takes the top three URLs from the SERP data and reads them on the web. It fixes the angle, then writes. It sends the piece to a registered destination as a draft and receives the post URL. Kafkai keeps only a snapshot of what it sent.

What you get: the draft URL. Review it on your own site before it goes live.

When it does not work

  • Keyword counts stay at zero: research is still running. Wait a few minutes and ask again.
  • No ranking changes: comparison needs at least two updates. Run one with Example 15.
  • Search Console cannot be read: either verification or access is still missing. Check the steps in Example 19.
  • The agent added a competitor on its own: adding is paid. Say "do not add any yet" as in Example 10 and the agent waits.