I rewrote a listicle that had lost 91% of its impressions. Ten keywords are back in the top 10.
A listicle that had quietly lost 91% of its search impressions came back with a rewrite built around a visible test method and a fact-check.
TL;DR
- A listicle that had fallen out of the top 10 for every tracked keyword lost 91% of its impressions over five months while untouched.
- Competitor audit ruled out the obvious fixes: the page already led the SERP on links and length. The real gaps were a missing test method, no named tester, and a title that didn't match the H1.
- The rewrite restructured the proof (a method block, hands-on testing of every tool, a fact-check) and capped length instead of adding more.
- Ten keywords returned to the top 10 within five weeks, with estimated traffic up roughly 40x on those terms.
0 → 10
Keywords in the top 10, of 10 tracked
13 → 522
Est. monthly organic traffic on those keywords, about 40×
Not ranking → #4
"best Keyword A" (3.6K searches/mo)
#15 → #5
"best Keyword B" (1.6K searches/mo)
This is the same AI writing SaaS behind two earlier case studies, on a keyword cannibalization fix and an AEO citation outreach channel, anonymized the same way here.
Between March and August 2026, a "best Keyword A" listicle on this client's site went from 2,195 US clicks a month to 14. Impressions fell from 48,712 to 4,198, and average position slid from 4.5 to 24.5. Nobody had touched the page in that time. It just kept losing ground.
I led the research and the rewrite. The page went live again on October 3. By November 4, ten keywords with a combined 9.9K monthly searches ranked in the top 10, and Ahrefs estimated their traffic at 522 visits a month, up from 13. Here's how it went and what I'd treat with caution.
Reading the SERP first
My first move was to look at what Google was showing. On September 8 I pulled the results for "best Keyword B." Position 1 was an AI Overview. Video, Reddit, Quora and forum threads filled most of the top 15. The ten article pages sat between positions 2 and 22, and ours was at 21.
I audited those ten pages against ours for content, structure, entities, keyword usage and backlinks. I expected links or length to explain the gap. They didn't.
- Our page had 42 referring domains at DR 40+. The page at position 2 had 15, and five competing pages had none.
- At 4,902 words, ours was the longest page in the set.
- The title said 11 tools, the H1 said 12, and the body reviewed 12. The title also targeted "Keyword D" (KD 81) instead of "best Keyword B" (KD 10, 11K US searches).
- Five of the ten competing pages had a "how we tested" section. Ours had none.
- The H1 claimed 41 tools were tested, yet the page showed no named tester, no detector screenshots and no method.
The page didn't need more links or more words. It needed visible proof.
The rewrite
I wrote the content brief and directed the writer through the rewrite. The brief capped the article at 6,000 words, allowed no new tools and no extra tables, so the explainer sections had to shrink to make room for the test results.
- Verdict and top picks first. The definition, free vs. paid, and tips sections moved below the list and shrank to 250 to 400 words each.
- One H2 per tool, with a "best for" label written into each heading. Competitors got narrow technical niches such as long documents, multilingual text and free tier limits.
- The list went from 12 tools to 11, and the title tag, H1 and body now all say 11.
- One comparison chart replaced 18 separate tables.
- A single method block with test conditions, linked samples, a results table with a baseline row and numbered limitations. The last limitation is an ownership disclosure: the site publishes the comparison and sells one of the tools in it, with a note on what was done to keep that honest.
- A rebuilt FAQ of seven answer-first questions, and a closing verdict at the end.
- A spec fact-check against vendor sites. Three claims on the old page were wrong, covering a tool count, a language count, and an "unlimited words" claim the page's own pricing table contradicted.
| Element | Before | After |
|---|---|---|
| Tools listed | 12 | 11 |
| Tool headings | H3 under one parent | 11 H2s with "best for" labels |
| Comparison tables | 18 separate | 1 consolidated |
| Title / H1 | 11 / 12 | Both 11 |
| Test method | None | Method block, limitations, disclosure |
| FAQ | 8 questions | 7, answer-first |
| Length | 4,902 words | 6,000 word cap |
Information gain
Most listicles in this space say the same things about the same tools. I didn't want ours to be one more copy, so I had the writer test every competing tool hands-on. Each of the 11 entries carries a first-hand result, and the page says something the other ranking pages don't.
The results
Seven keywords entered the top 10 from outside the top 100, and three moved up by 10 to 15 positions. Two of them, "best Keyword A" and "best Keyword B," account for 349 of the 522 estimated monthly visits.
| Keyword | Vol. | Oct 1 | Nov 4 | Est. traffic |
|---|---|---|---|---|
| best Keyword A | 3.6K | Not ranking | #4 | 256 |
| best Keyword B | 1.6K | #15 | #5 | 93 |
| Keyword C | 2.6K | Not ranking | #8 | 91 |
| best Keyword E | 450 | #21 | #6 | 22 |
| what is the best Keyword B | 600 | #20 | #7 | 21 |
| best Keyword F | 300 | Not ranking | #6 | 15 |
| best Keyword G | 350 | Not ranking | #7 | 14 |
| best Keyword H | 400 | Not ranking | #9 | 9 |
| best Keyword I | 30 | Not ranking | #8 | 1 |
| sites like Competitor A | — | Not ranking | #9 | 0 |
Source: Ahrefs Organic Keywords for the refreshed URL, Oct 1 vs. Nov 4, 2026.
What I'd treat with caution
These are Ahrefs estimates for ten keywords over four and a half weeks. They aren't Search Console clicks. This page has also swung a lot since May, so I'm calling the results early and will keep tracking them.
What this case illustrates
- 01
Read the SERP before touching the page
The competitor audit showed which gaps mattered (no method, no tester, a mismatched title) and ruled out the ones that might have cost weeks of the wrong work: links and length.
- 02
Proof structure carries weight
A listicle that sells one of its own entries needs a visible method, stated limits and a plain ownership disclosure. The pages ahead of ours already had a method block, and hands-on testing of every competitor gave ours something new to say.
Disclaimer: All ideas and data are mine, Claude helped organize it and publish.