Manual
Product site My Page

SEO Insights and Rewriting

Last updated · Published · Sorabun Editorial Team

Using real Search Console data, find queries that are close to page one (with the best-ranking page), queries that just appeared, and articles whose clicks have dropped, then rewrite them. Rewrites start from the original article's headings and the search queries that bring people to it, and keep the title of articles that already rank well. Rewrites are created as drafts, and you check them section by section before applying them to the original. You can also swap in a new title for articles with a low click rate for their ranking in one click, and compare the 28 days before and after. Suggestions for what to write next and keyword data (searches, difficulty, search intent, trend, and cost per click) also appear here.

Which articles should you fix?

The "SEO Insights" tab in "Sorabun > Measure / Analyze" automatically finds three things from real Search Console data. "Analyze insights" fetches them from Site Kit again on the spot.

  • Queries close to page one (positions 5-20): Each query from the last 28 days is reduced to your site's best-ranking page for it, and queries whose best page averages position 5 to 20 are listed by impressions. They may only need a little push to reach the top of page one. When the article showing up for that search is known, "Rewrite this article" appears (you already rank for the search, so a new article would compete with your own)
  • New search queries: Queries with 10 or more impressions in the last 7 days that had none in the 28 days before. They're signs of new demand
  • Articles with declining clicks: Articles that are being read less than before. Likely causes are stale information or competitors catching up

How queries close to page one are chosen

Search Console returns one row per page a query showed up on. Where your site ranks for a query is decided by its best page, so Sorabun first picks each query's best page (the lowest average position; on a tie, the one with more impressions) and then checks whether it's between 5 and 20.

  • Queries where any page is already above position 5 are left out. Improving another page wouldn't add much traffic
  • If there were too many query and page combinations to fetch them all (5,000 rows), the screen says so. The best page may then be a different one

How new search queries are found

  • The periods are the last 7 days and the 28 days before. Search Console numbers take a few days to settle, so the periods end 3 days ago
  • Queries with search operators such as site: or filetype: are left out. Those come from tools, not people
  • If the earlier 28 days hit the row limit (5,000 queries), no rows are shown. Sorabun can't tell whether a missing query really wasn't there or was cut off, so it can't say it's new (the screen says so)
  • From the list you can go to "Create article" or "Add to rank tracking"
Credits

The rules for queries close to page one and new search queries (positions 5-20, the best page, 7 and 28 days, 10 impressions, recognizing operators) are based on the open-source SEO tool OpenSEO (MIT License, (c) 2026 Ben Senescu).

An AI can read both through the MCP tool search_console_queries.

Further down the same screen, "Title review" lists articles whose click rate is low for their ranking.

Rewrite in one click

Run "AI Rewrite" from the target article, and Sorabun rewrites the body based on the current search intent and trends among top-ranking articles.

A published article is never changed on the spot. The rewritten text is saved as a "rewrite draft" linked to the original. Open the draft and you'll see "Apply to the original" and "Discard" at the top, along with a diff against the current text, so you can read it before deciding. Applying it updates the original at the same URL and removes the draft. The old text stays in revisions, so you can still go back. In the posts list, drafts are marked "Rewrite draft" and originals "Rewrite pending." Pressing "Publish" on a rewrite draft applies it to the original instead of creating a second post. Set a future date and "Schedule" it to replace the original at that time (you can cancel the schedule from the top of the draft's screen).

Rewriting from the original article

A rewrite doesn't start over from scratch. It builds on the original article.

  • The original headings are passed along. The AI gets the original article's headings (H2 and H3) and the opening of each section, and decides for each section whether to keep, revise, add, or remove it, with a short reason for each. Kept and revised sections are written from the original section's text, so your own experiences and figures stay unless they're out of date
  • Queries that bring people in are protected. If Search Console is connected through Site Kit, the AI also gets the article's search queries from the last 28 days (the top 15 by clicks, impressions, and average position), and is told not to remove or weaken sections that answer them. Without Search Console, the rewrite goes ahead without queries (it doesn't stop)
  • AI costs stay the same. The outline step gets more material, but the number of AI calls doesn't change

If you rewrite from "Rewrite existing article" on the New Article screen with outline review turned on, the outline screen shows "Keep", "Fix", or "Add" with a reason on each heading, and a list of sections to remove below.

The title stays when the article already ranks well

Changing the title of an article that ranks well can cost rankings and clicks, even when the body gets better. So Sorabun uses the article's last 28 days of Search Console data to decide whether to keep the title. When in doubt (no data, or too little), it keeps the title.

The article's numbersTitle
The main query (the one with the most clicks) averages position 3 or betterKeep
Average position 10 or better, with 10 or more clicks in the last 28 daysKeep
Neither of the above (below position 10, or on page one with fewer than 10 clicks)New title
Fewer than 100 impressions in the last 28 daysDecided by Rank Tracking (keep at position 10 or better, new title below that). Kept if there's no rank either
No Search Console data and no rankKeep
  • On the New Article screen, under "Rewrite existing article", the "Title" setting lets you choose "Keep the original title" or "Write a new title", and your choice wins. Rewrites started from the post list, Insights, or Rank Tracking use "Decide from search data"
  • When the title does change, the AI is told to keep the words of the queries people currently use in the new title
  • When the title is kept, the AI's own title suggestion still appears on the outline screen and the rewrite draft screen. To use it, change the title before applying
  • Ranks checked more than 60 days ago aren't used

Checking the rewrite draft section by section

When you open a rewrite draft, the notice at the top shows:

  • Section comparison: how many sections were kept, revised, added, and removed. Sections that were reordered or got a reworded heading are matched by their text, so they don't count as removed or added
  • Title: whether it was kept or changed, and why (main query, average position, clicks)
  • Removed sections: sections in the original that aren't in the draft, with the reason the AI gave
  • See section-by-section changes: which original heading matches which draft heading, and the queries each section answered

If a section that answered a query people actually click (at least once in the last 28 days) was removed, a red notice appears at the very top. It also says whether the query's words still appear in another section of the draft. Decide whether to bring the original section back before applying. Queries whose words disappear from the whole article show up in the same notice.

With "Overwrite the published article on the spot" (rewrite_mode), there's no draft, so the same warning appears on the job's completion screen instead.

Matching is a text comparison

Sections are matched by how much of their headings and text overlap, without calling the AI again. A section whose heading and text were both heavily rewritten may show up as one removed and one added. A section counts as answering a query when it contains every word of the query (words separated by spaces). Words that appear in more than half of the sections (the article's main topic) aren't tied to any single section.

"Rewrite with latest info" from the post list (Pro)

In "Posts > All Posts", each post has a "Rewrite with latest info" link under its title. Clicking it researches the latest information on that post's topic and rewrites it. You can fix any post you think has gone stale, even if Insights doesn't list it.

  • A published post gets a rewrite draft, as described above. The original post doesn't change until you compare them and click "Apply to original".
  • While it's running, the row shows the current stage, such as "Rewriting (Outlining)". You can't click it again until it finishes, so the same post isn't rewritten twice and billed twice.
  • It uses your AI provider and costs money. A confirmation appears when you click.
  • If only part of a post is outdated (a model name, a price), the freshness check is cheaper. It replaces only the outdated parts, and if nothing changed it just extends the review date.
If you want rewrites applied right away

Set "Settings > Generation Defaults > Rewrite result" (rewrite_mode) to "Overwrite the published article on the spot" to rewrite without the review step, as before. Drafts and scheduled posts are rewritten directly under either setting, since no one has seen them yet.

Title review (articles with a low click rate for their ranking)

When an article shows up in search results but rarely gets clicked, the title is worth checking before the ranking. "Title review" at the bottom of this screen uses the last 28 days of Search Console data to list articles with a lower click rate than other articles on your own site at a similar ranking. Press "List articles to review", or "Analyze insights" at the top, to fill the list.

What it compares against

The baseline is not a figure from someone else's study, such as "position 1 gets X%". It is the click rate of your own site's articles. Click rates vary a lot with the kind of search, the industry, and the layout of the results page (ads, maps, AI Overviews and so on), so another site's average doesn't apply to yours.

RuleDetails
Articles it compares withArticles whose average position is between two thirds and 1.5 times that of the article (about 2.7 to 6 for position 4). The article itself is left out
Baseline valuesThe median click rate of those articles, and their lower quartile
Articles listedArticles whose click rate is below the lower quartile, and whose shortfall against the clicks expected at the median is larger than random variation (twice the square root of the expected clicks)
Articles left outArticles with fewer than 200 impressions in 28 days (they are neither listed nor used as a baseline), articles ranking below position 20, and articles with fewer than 5 articles to compare with
OrderBy how many more clicks the article would have had at the median click rate. This is a rough guide, not a promise that a new title will bring those clicks

The cut at 200 impressions is there because a few clicks can swing the rate a lot when impressions are low. At a 3% click rate, 200 impressions mean about 6 expected clicks. At 100 impressions it's 3, and a single click moves the rate by a third.

The list only shows articles you can edit. The baseline uses every article on the site.

Swap the title in one click

Each listed article shows title options. Click one and confirm, and Sorabun updates the post title and the title in your SEO plugin (when a per-post title is set in Yoast SEO, Rank Math or SEO SIMPLE PACK).

  • The options are the ones saved when the article was created. Sorabun keeps the title options the AI suggested while outlining, including the ones you didn't use (up to 8 per article). See Title patterns that get clicked for how they're written
  • For articles with no saved options (for example, ones written before you installed Sorabun), "Have AI suggest 3 options" writes them from the search queries the article actually appears for. This uses your AI provider and costs money (one AI call)
  • You can also type your own title
  • If the SEO plugin title looks like "old title | site name", only the old title part is replaced. Titles written with variables such as %%title%% follow the post title on their own, so they are left alone. Per-post titles in All in One SEO are not changed
  • The content isn't edited, so the article's modified date stays the same
  • Each swapped article keeps a record of when it changed, from what to what, and the numbers for the 28 days before. The swap also goes into the Activity Log
  • While the before/after comparison is running (until day 30 after the swap), the same article can't be swapped again, because that would make the comparison meaningless. To try a different title, click "Undo" first
  • Articles without Search Console numbers for the last 28 days can't be compared, so they aren't swapped here. Edit the title in the editor instead

Compare the 28 days before and after

"Swapped articles (28 days before and after)" lists the articles you changed. It sets the 28 days starting the day after the swap next to the 28 days before it. Search Console takes about 2 days to settle, so the comparison is ready on day 30 after the swap. Open this screen after that date and the numbers are fetched automatically.

What you'll seeMeaning
The click rate went up / went downThe difference is larger than random variation (the gap between two rates, checked against twice its standard error). Even so, it isn't necessarily the title's doing
Within random variationYou can't say whether the new title helped or not
The average position moved a lot, so this is hard to compareThe average position changed by 20% or more (for example, from 5 to 6). Click rates depend heavily on position, so the title's effect can't be separated out
Too few impressions, so this is hard to compareOne of the two 28-day periods has fewer than 200 impressions
Don't read the difference as the title's effect

The before/after difference also includes changes in ranking, season, news and how people search. Google also sometimes rewrites titles instead of showing them as written, so the headline in search results may not change even after a swap.

"Undo" restores the title from before the swap. The SEO plugin title is restored too, as long as it is still what Sorabun wrote. If someone changed the title after the swap (in the editor, for example), Undo doesn't overwrite it. Articles that were undone partway through aren't compared, because the 28 days after would include days with the old title.

When Search Console isn't connected (Site Kit is missing or not set up), the panel explains how to connect it. The record of swapped articles and "Undo" work without the connection.

Suggestions for what to write next

Sorabun analyzes patterns from articles that performed well and automatically suggests keywords to write about next, once a week. It's a way of extending your winning patterns to new articles.

Suggestions take your site description and existing articles into account. They're filtered with the same checks as the AI Editor-in-Chief (overlap with existing articles, whether you already rank, search volume, autocomplete, the strength of the top results, keyword difficulty, and search intent), then ordered by how promising they are.

With a DataForSEO key, each suggestion shows labels for difficulty, search intent, and cost per click, plus small bars for the 12-month trend. If you turn on "Settings > Search data source > Add real-data candidates" (labs_candidates), Sorabun also adds real search keywords found from your site's genre and categories to the AI's ideas before choosing. Suggestions added this way say "A search keyword found in real search data (DataForSEO)" in the reason column.

Keyword data (search volume, difficulty, intent, trend)

Below "What to write next," Keyword data lets you look up keywords yourself with DataForSEO's search data. To use it, enter your DataForSEO login and password (dataforseo_login / dataforseo_password) in "Settings > Search data source."

ActionWhat it doesTypical cost per run
Find candidates from a seed keywordFinds up to 40 keywords people actually search for, starting from a seed keyword (for example, ふるさと納税)About $0.029. For seeds with few results, related searches are added, up to $0.046
Look up numbers for a listOne keyword per line, up to 100. Shows the numbers for eachAbout $0.024 for 100 keywords. Asking Google Ads about keywords DataForSEO Labs doesn't have adds $0.09

Both make a request only when you click. Before you click, the estimated cost of one run (in US dollars) appears next to the button and in the confirmation. Keywords and seed keywords you looked up in the last 30 days use the saved numbers, so they cost nothing.

The table has these columns.

ColumnWhat it shows
Monthly searchesMonthly searches on Google in Japan. Keywords searched so rarely that there's no figure are left blank. They aren't counted as 0
DifficultyHow hard it is to reach the top 10 (0 to 100). DataForSEO estimates it from things like the backlinks of the top pages. 0 to 29 is low, 30 to 59 is medium, and 60 or more is high
Search intentOne of four: research, comparing options, buying or signing up, and going to a specific site
12-month trendSmall bars for monthly searches. The rightmost bar is the latest month. You can see whether a keyword rises with the seasons
Cost per clickWhat one ad click costs (in US dollars). The higher it is, the more likely the keyword leads to a sale
ChecksWhether it targets the same search as an existing article, and whether this site already ranks for it (Search Console)

Searches you already rank for get a "Rewrite this article" button, and the others get "Create article." Keywords that target the same search as an existing article get no button.

How candidates are found

Sorabun takes half of the candidates from keywords that contain the seed (DataForSEO Labs keyword_suggestions) and half from keywords in the same field (keyword_ideas), then alternates them and removes duplicates. Either source alone misses close terms that don't contain the seed, or drifts toward terms that are too broad. Only when there are fewer than five keywords besides the seed itself does it add keywords found by following "People also search for" (related_keywords).

How numbers are fetched, and what they cost

Other screens (the AI Editor-in-Chief, keyword suggestions when creating an article, the AI agent, search volume in rank tracking, and so on) also get keyword numbers from DataForSEO Labs. It costs $0.012 per call plus $0.00012 per keyword (about $0.024 for 100 keywords), which is cheaper than the Google Ads search volume Sorabun used before ($0.09 per call). Only keywords Labs doesn't have (often long-tail ones) are sent to Google Ads. Keywords with no figure from either source stay "unknown" instead of being counted as 0, so they aren't removed.

Daily limit

There's a daily limit on the total Sorabun spends on DataForSEO ("Settings > Search data source > DataForSEO daily limit," labs_daily_usd; $1 by default, $50 at most). Once the limit is reached, Sorabun stops making requests for the day and carries on with the numbers it fetched in the last 30 days. Set it to 0 to make no requests at all.

What this is based on

How candidates are found (alternating sources and adding more when there are too few), how DataForSEO is called, and the cost estimates are based on the design of the open-source SEO tool OpenSEO (MIT License, (c) 2026 Ben Senescu), rebuilt to run inside WordPress.

Keywords you send from "Keywords only your competitors rank for" in the Technical Audit with "To article ideas" are listed at the top here too (Pro). The reason shows the competitor's position, monthly searches, difficulty and your own position. Sent keywords survive the weekly automatic refresh.

Rewrite before you write something new

Pushing an article that's already showing up in results onto page one is often faster than ranking a brand-new article from scratch. Start with the queries close to page one.