Updated
AI Content Cleanup: triage report
Every page decided. Keep, rewrite, merge or cut, on the numbers.
Which of these pages caused the drop, and which are worth keeping?
Sites filled with content published at scale that have since lost traffic.
What actually marks it as slop
Not a hunch about how it reads. Countable tells, measured across every page.
Content effort score. How much of this could only have been written by someone who did the work. Low is the problem. It is what separates a page worth rewriting from one worth deleting.
- Filler vocabulary delve, leverage, robust, seamless, testament, landscape 1,847 uses
- Near-identical openings Pages starting with the same sentence shape 294 pages
- No first-hand experience Nothing only someone who did the work would know 361 pages
- No original numbers No data, measurement or figure of your own 338 pages
- Uniform paragraph length Every paragraph the same size, start to finish 271 pages
- Forced triplets Lists of exactly three, over and over 612 uses
- Em dash density Per thousand words, against a human baseline of 1.2 9.4 per 1k
- Templated heading pattern The same H2 sequence repeated across pages 187 pages
An illustration of the output format, measured against a 412-page example. Your pages, your counts.
- A reason for every decision. A page is not cut because it reads badly. It is cut because it earned nothing and carries none of the things above.
- The effort gap, page by page. Which pages need a paragraph of real experience adding, and which need deleting because there is nothing to add it to.
- What to change, in order. Highest impact first, so your writer or agency has a queue rather than a lecture.
- Something to hand your agency. If they published it, this is the document that ends the argument about whether it was good enough.
Four sample pages, so you know what arrives.
The structure, depth and evidence style below are what you get. Every URL and number on these pages is example data, blacked out the way a client copy would be redacted.
In this example, 90 pages earn removal and 58 fold into stronger pages. The numbers behind every single verdict are in the table, so you can disagree with any of them.
Page 1 of 18. Verdict counts for the whole site, then page level detail behind them.
| URL | Verdict | Clicks, 90d | Effort score | Why |
|---|---|---|---|---|
| /blog/ | cut | 0 | 12 / 100 | Never shown by Google in 90 days, no original numbers, opening repeated on 31 other pages |
| /guides/ | keep | 214 | 78 / 100 | First-hand detail, still earning, one broken source to fix |
| /blog/ | merge | 9 | 41 / 100 | Competes with the guide above for the same query; fold in the two useful sections |
| /blog/ | rewrite | 37 | 29 / 100 | The query deserves a page; this one answers it with filler |
| /blog/ | cut | 1 | 8 / 100 | Template detected across 62 pages, this one adds nothing local |
| /services/ | keep | 388 | 84 / 100 | A money page. Left out of the cleanup entirely |
One row per page, all 412 in the full report, sortable in the spreadsheet copy.
Monthly clicks against Google's published update dates. In this example the fall starts the week an update rolled out, and the pages that fell hardest are the ones the table marks cut and rewrite. The report shows the correlation and says plainly that correlation is not proof.
Built from your own Search Console data, which is why read access is the first thing I ask for.
- Remove the 62 templated pages first. Zero traffic between them, and they are the clearest scaled-content signal on the site.
- Merge the three overlapping guides on . They split one query three ways; the strongest page gets the traffic of all three.
- Rewrite the 12 pages where the query is worth it. Listed with what each rewrite must contain.
Highest impact first, so the cleanup starts paying back before it is finished.
In the report
- Every URL published at scale, with a keep, rewrite, merge or cut decision
- The numbers behind each decision, so you can disagree with any of them
- Your traffic drop correlated against Google's published update dates
- A cannibalisation map showing which pages compete with each other
- AI crawler access check across GPTBot, ClaudeBot, PerplexityBot and Google-Extended
- A content effort score per page, measuring how much of it only someone who did the work could have written
- The countable tells behind that score: filler vocabulary, repeated openings, missing first-hand experience, no original numbers
- The order to work in, highest impact first
Not in the report
- Implementing the changes. That is quoted separately once the size of the job is known.
- Writing the replacement content.
- Link work of any kind.
Saying what a thing does not include is the fastest way to stop you buying the wrong one.
Three things, then I start.
You get asked for these on the confirmation page the moment you pay. Nothing else is needed and there is no call to book.
- Read access to Google Search Console for the domain
- The date you first noticed the drop, if you know it
- Whether anything else changed around then, such as a migration or a redesign
Still not sure it is the right one?
Describe what happened and when. If this report is wrong for you, I would rather say so now than take £395 for it.
Ask first