Manual
Product site My Page

AI Director

Pick a single obstacle from real measurements, set a prediction before acting, then check the results on the due date

Don't stop at "we made an improvement"

"AI Director (Improvement Cycle)," the top tab under "Sorabun > Leave It to AI," is a feature that runs the improvement process itself (Pro).

From actual measurements of rank, search data, read rate, and CTA performance, it picks out the single biggest thing currently holding you back, decides on a countermeasure and a prediction ("this number will look like this in 28 days") set before taking any action, and then runs it. On the due date, it automatically checks the outcome against real measurements, and carries what it learned from the gap between prediction and reality into the next cycle.

AI is not used to judge the results

If you ask an AI "did this work?", it will almost always answer "yes, it did." Judgment here is based only on actual measurements, so a lack of real results can never get waved through as "it worked."

How it proceeds

This is built on the improvement kata approach (target state → current state → a single obstacle → experiment → learning). The screen is laid out in that same order.

1. The obstacle to tackle (only one)

Fixing several things at once makes it impossible to tell what actually worked. Sorabun narrows it down to one.

2. The action to take

The screen distinguishes between what Sorabun can execute itself (generating or rewriting articles, etc.) and what requires manual work. Manual items are labeled "This is manual work (takes a few minutes)," so it's clear which parts can't simply be delegated.

3. The prediction (decided in advance)

You settle on "◯◯ now → ◯◯ after ◯ days" up front. This is so the bar can't be moved after the fact.

Runs only after you approve

Articles are created as drafts and only run after you approve them. You can also choose to have this run automatically.

The results check still runs even with automatic execution OFF

Only the results check for an experiment whose due date has arrived runs automatically. This is so an experiment never just gets left unevaluated.

To improve accuracy

The screen shows "things that would improve accuracy if set up." With no ranking keywords registered at all, or read rate not being measured, there isn't enough information to work with. Filling in what's listed here improves the accuracy of the obstacle Sorabun selects.

Related features: