A customer told me this month that her online store gets four to five times the visitors it did two years ago. Then she said the part that mattered: they come, but they do not buy.
She had paid for keyword data for two years and never really used it. The most important keywords for her columns came from someone else. She opened Search Console now and then, and did not know where to look.
If that sounds like you, this post is for you. It is not about rankings. It is about what to look at when the numbers go up and nothing else changes.
Rankings Were the Wrong Thing to Watch
For years, checking on your website meant one thing: are we ranking for our keywords? That number is easy to read and easy to feel good about. On its own, it also tells you almost nothing.
A ranking is one dot. It does not tell you who is above you, or why. It does not tell you what your competitors rank for that you do not. It does not tell you which of your pages bring in people who stay, and which bring in people who leave in five seconds. And it does not tell you what changed since last month.
The store owner had good rankings. She still could not answer the only question she cared about: what do I do next?
The Problem Was Never the Data
Here is what I learned from sitting with customers this September. Nobody lacked data. Everybody lacked a reader. More data does not fix that. Better decisions do, which is the point of growth depends on smarter decisions, not more data.
Keyword tables are made for people whose job is to read keyword tables. Most small businesses do not have that person, so the data just sits there. One customer put it plainly: he never got to use the good parts these data would have bring.
That is what changes in Kafkai Ver.4. You do not read the data. Your AI does. You connect Kafkai to the ChatGPT or Claude you already pay for, and you ask a question in ordinary words. Your AI pulls the real numbers from Kafkai and answers you. This kind of connection is called MCP, and it works like a plugin. The Ver.4 announcement explains it in one paragraph.
The plumbing is not the point. The point is that, for the first time, you can ask the four questions below and get an answer.
The Four Questions
These four questions turn a pile of numbers into a next step. Ask them in this order.
1. Who is beating me, and on what?
You have probably looked up a competitor on Similarweb. It tells you how much traffic they get. It does not tell you which searches send it. The question is not "where do I rank?" It is "where do they rank that I don't?" Kafkai sorts every keyword that your site and your competitors' sites appear for into four groups: the ones you both compete on, the ones they own and you do not, the ones you already rank for but could do better on, and the ones nobody has taken yet. Your next article comes from the second group. You do not even need to know who your competitors are. Give Kafkai your domain and ask.
2. Who links to them and not to me?
Every site that links to two of your competitors but not to you is a site you can approach. That list is where link building starts, and no free AI can see it. Ask for it once, and you have a real outreach list instead of a guess.
3. What did my visitors actually do after I published?
This is the question the store owner needed. Connect your Search Console once, then ask what changed this week. You will see which pages gained clicks, which lost them, which show up in search but get no clicks, and which ones Google crawled but never indexed. You will find pages that bring visitors who leave, and pages that bring the right people. Fix the first kind. Write more of the second kind.
4. What moved since last month?
One snapshot tells you where you are. Two tell you where you are going. Every time Kafkai refreshes, it keeps the old data. After a few months you can ask which keywords rose, which links you gained or lost, and which competitor is moving fastest. That is the difference between a photo and a map.
Ask these four, and the answer is not a table. It is a short list: fix this old article, write this new one, contact these three sites. Each item comes with a reason.
We Ran This on Our Own Site
It would be easy to write the four questions and stop there. So this week I asked them about kafkai.ai, our news site. Here is what came back. Every number comes from Search Console, read through Kafkai, and each one is labelled with its source.
In the 28 days to 28 September, Google showed kafkai.ai (our media platform) 10,239 times and 78 people clicked (Search Console). In the 28 days before that, it was shown 16,227 times and 64 people clicked. Impressions fell by more than a third. Clicks went up. Average position moved from 23.9 to 17.0. Look only at impressions and you panic. Look only at position and you celebrate. Look at all three and you see a site that is shown less but chosen more.
Four pages told me what to do.
An article about Neil Patel and SEO was shown 757 times and clicked zero times (Search Console). It ranks at 18. Nobody clicks because the title does not match what people searched for. That calls for a rewrite, not a new article.
An article on the McKinsey State of AI report jumped from position 28 to 7.8 in one month (Search Console). It was shown 898 times and got 3 clicks. Google moved it to page one, but the title did not earn the click. Fix the snippet now, before the page slips back.
The Japanese version of our article on Japanese companies entering Malaysia went from 3 clicks to 13, and from 56 impressions to 292 (Search Console). The numbers are small, but the direction is clear. Write more like it.
The Japanese version of our article on why a keyword difficulty of zero does not guarantee a top ranking was shown once in August and 210 times in September (Search Console). It sits at position 28. Demand appeared, and the page is on page three. Fix this one first.
That was Search Console alone: two titles to rewrite, one topic to repeat, one old article to fix. Search Console tells you what happened on your own site. It cannot tell you what to write next, or who is in your way. That is what Kafkai's other three sources are for, so I asked them too.
Rankings. Kafkai's own sample found kafkai.ai ranking for 7 keywords in Japan (Kafkai). Only one is near page one: 検索 意図 (search intent), at position 20, with 320 searches a month. The page that ranks is the Japanese version of our search intent article. That is the cheapest win on the site: one rewrite for one page-one position. The other six sit between positions 35 and 94, where almost nobody looks. Two of them, both about third-party cookies, get 1,000 and 720 searches a month. The page exists. It is just buried.
Advertiser value. Kafkai also returns what advertisers pay per click for each keyword. Of the seven, only AIの課題 (AI challenges) has advertisers paying for clicks at all (Kafkai). When a keyword has a price, someone is making money from it. That is the one worth the extra effort.
Competitors and the 4C. I gave Kafkai one competitor, graffer-aistudio.jp. Kafkai compared the two sites and told me something I did not want to hear: their site has 2.4 times our authority, so fighting them on their own keywords is a poor bet (Kafkai). It then sorted 591 keywords into the four groups of the 4C framework:
- Catch-up, the ones they own and we do not: 55
- Compete, where we both rank: 3
- Consolidate, where we rank and they do not: 4
- Complement, where nobody ranks yet: 522
For a site like ours, the answer is the last group. Nobody owns DX推進 (digital transformation, 4,400 searches a month), データ活用 (data use, 1,900) or AI 業務効率化 (AI for business efficiency, 1,300) yet (Kafkai). Those are our next articles. The question is, how do we fit those topics into what we have? That is where you use your AI agent to suggest to you something ater feeding it data about yourself.
The list is not perfect. Its top entry was a medical subsidy term that has nothing to do with us. Your AI sorts. You choose.
Backlinks. kafkai.ai has 316 links from 82 sites. The competitor has 1,278 links from 92 sites (Kafkai). About the same number of sites, four times the links. Kafkai can list the sites that link to two competitors and not to us, but that needs two competitors, and I gave it only one. So this month the honest answer is: add a second competitor, then ask.
Put together, the four sources give a list that Search Console alone could not:
- Rewrite two titles.
- Fix one buried article.
- Push 検索 意図 onto page one.
- Write about DX推進 first, because nobody owns it.
- Add a second competitor to get the link list.
All of the above came from six questions and cost ¥15 in Kafkai credits.
What Kafkai Does Not Do
I want to be honest about the limits, because you will find them anyway.
Kafkai holds search data: keywords, rankings, backlinks, competitors, and your own Search Console. It does not hold facts about your industry, and it does not check whether an article is true. Your AI does the writing and the fact-finding, and you still read what it writes before you publish. Kafkai publishes as a draft unless you say otherwise, so nothing goes live without you.
Kafkai also does not yet tell you why a visitor did not buy. It tells you which pages bring which visitors. That is the start of the answer, not the end of it. Often the rest of the answer is trust, which is the subject of traffic is nice, but trust is what converts.
One rule makes all of this safe to use: tell your AI to label where every number came from, Kafkai or the web. Then you can check any figure you doubt.
What This Costs
There is no subscription. One credit is one yen, and a month in which you use nothing costs nothing. Researching your site and one competitor costs under ¥1,000. Asking the four questions above costs a few dozen yen each time. The kafkai.ai check in this post cost ¥15. The store owner's session this month, reading her Search Console, cost ¥62. Writing the article costs nothing on Kafkai's side, because your own AI writes it. The pricing page has the full price card.
One Small Step
You do not need to set everything up. Register, take the 1,000 free credits, and ask one question about your own site. If you do not know what to type, the Example Prompts page has ready-made prompts. Copy one, change the domain, and send it.
If the answer surprises you, you have found the thing to look at.
