What was done, when, how it was measured, and what it does not prove
Case Studies
SEO and web design evidence with the dates, data sources and limits attached—so the work can be assessed without turning correlation into a claim.
A case study here means evidence with its limits attached
A percentage on its own tells you very little. “300% growth” could mean 3 clicks became 12. Before trusting any case-study figure, ask: over what dates, measured from what baseline, from which data source, what else changed in that period, and does the evidence show the work caused it? Every case study on this page answers those questions, and states what it does not prove.
This page is short because I only publish work that can be checked. It also means everything on it can be checked.
Clothing Alterations Nottingham — a WordPress build for a service business
What was built
The brief was a site that let customers find the specific alteration they needed, see what it would cost, and book—rather than a single general page listing everything the business does.
The finished structure has ten individual service pages—dress, wedding dress, bridesmaid, prom, suit, trouser, coat and jacket, and curtain alterations, general tailoring and zip repairs—each built to answer that service’s own questions. Alongside them sit a published pricing page, contact and appointment routes, a guidance blog for questions customers ask before booking, and an accessibility page, which is rare on a small service site and worth a customer knowing about.
Everything is editable by the business in WordPress without developer help for ordinary changes.
The first 50 days of Search Console data
This historical Search Console snapshot covers 29 July–16 September 2026: 50 days, split into two equal 25-day periods.
| Period | Clicks | Impressions | CTR |
|---|---|---|---|
| 29 July – 22 August 2026 | 93 | 2,157 | 4.3% |
| 23 August – 16 September 2026 | 150 | 5,172 | 2.9% |
Source: Google Search Console, web search, clothingalterationsnottingham.co.uk. Two equal 25-day periods.
What this establishes
Across the two periods, recorded impressions increased by 140% and clicks by 61%, while click-through rate fell. Changes in the mix of searches could contribute, but these totals alone do not establish why CTR changed.
What this does not establish
Fifty days of a young property is a baseline, not a result. Nothing here proves the build caused the movement. It covers no seasonal comparison, no year-on-year data and no enquiry figures.
What would turn this into a result
The same comparison repeated over a longer period, set against seasonality, with enquiry data alongside. When that exists it will replace this section.
The rise could reflect Google processing a new site as much as anything done to it. It is published because it is real and dated, not because it proves anything on its own.
Two pages competing for the same searches
On the same site, Search Console showed two pages appearing for the same cost searches—“zip replacement cost”, “zip replacement cost near me” and “zip repair cost”:
| Page | Clicks | Impressions |
|---|---|---|
| Zip Repairs Nottingham (service page) | 10 | 618 |
| Coat Zip Replacement: Cost and Turnaround (guide) | 4 | 357 |
Source: Google Search Console, 29 July – 13 September 2026.
A third zip page—about whether a broken zip can be repaired—never entered the overlap at all. Its queries were about a different problem.
What this shows: overlap was real between two pages, and absent for the third, even though all three are about zips. Shared subject is not shared intent. Deciding which pages actually compete—rather than assuming that similar topics do—is diagnostic work, and it is the difference between consolidating correctly and deleting a page that was doing its own job.
The full reasoning, and the six ways of resolving competing pages, is on Content and On-Page SEO.
Rebuilding a network of near-identical location pages
This one is anonymised. The network, business and service are not named, and nothing here describes a ranking or commercial outcome.
What was found
Two location pages in a network of same-service pages were compared with their town names and customer names masked out. They shared 1,116 of 1,138 words of verbatim text—98.1%—in a single unbroken run of 977 words, from the opening line to the FAQ. Other pages in the set carried sections naming the wrong town and local detail that could not be verified.
What was done
The affected page was rewritten from scratch rather than edited, then measured against its live sibling through four further passes using a four-word match with town names counted. Duplication moved from 28.7% to 22.7% to 19.7% to 12.1%, while the word count barely changed—from 2,381 to 2,335—so the improvement came from rewriting, not cutting.
What it taught, measured
Repeated mandatory wording was the biggest single source of duplication. Pages selling the same service under the same policies must carry some identical phrases—safety statements, policy wording, legal text. The duplication floor for two such pages first calculated at 19–20%. That turned out to be wrong: the pages were stating each mandatory phrase two or three times rather than once. Stating each one once removed around 60 matched words at no cost to the reader, and the real floor was 10–11%.
Tables duplicate harder than prose. A table of the same facts about the same services will always match its sibling, because a table strips out the connecting words that make prose differ. Converting one comparison table to short paragraphs—same facts, different order—took that section from 67 matched words to 14. That is a genuine trade-off: tables are easier to scan and extract, so the choice is not automatically right.
Removing duplication strips the qualifications first—and putting them back adds duplication again. Compressed, distinctive phrasing tends to drop the small words doing the legal and operational work: “generally”, “usually”, “up to”, “subject to”. A page that is 11.7% duplicated can quietly have become less accurate than one that is 20%. Restoring those qualifiers pushed duplication from 11.7% back up to 13.0%, because qualifying language is also the most natural, shared phrasing. The two goals move against each other in both directions, and they cannot be done in the same edit.
One process lesson: duplication must be measured against what is actually published, not against a local draft. A page once measured 11% against a local file and 39% against the live URL, because the live sibling was an older version.
What this establishes: that duplication can be measured precisely, located by section, and reduced by method rather than by guesswork. What it does not establish: any ranking, traffic or commercial effect. A similarity score cannot tell you whether a page is useful, and Google publishes no acceptable percentage.
The wider guidance on location pages, including where Google’s spam policy draws the doorway line, is on Multi-Location SEO.
Why there are no percentage headlines here
Read percentage changes alongside their dates, baseline and limitations. The figures above describe the recorded period; they do not establish that the website work caused the movement.
If you have been shown impressive percentages before and found they did not survive a question, the useful test for any case study, including these, is: over what dates, measured how, compared with what, and what else was happening?
What every future case study will include
As new work is completed, it will be added here only if it can carry:
- a named or explicitly anonymised client, with permission for either
- the exact date range of any figures
- the data source, and ideally a screenshot or export
- what was changed, and when
- a baseline the figures are measured against
- what the evidence does not prove
- anything outside anyone’s control during the period—seasonality, updates, market change
Anything that cannot meet that standard stays off the page, however good the number.
Work completed or results observed?
Completed work
A case study can show what was built, corrected or changed and when. Checking the live work can establish that it was delivered.
Observed results
Search or enquiry figures describe what happened over a period. They need dates and a baseline, and a change alone does not prove which action caused it.
Named example or anonymised example?
A named example lets you inspect the business and, where available, the live work. An anonymised example can still explain a real problem and method, but provides less for an outsider to independently verify. Permission and privacy matter in both cases.
How to assess an SEO case study
Ask for the source, dates, starting point, work completed and relevant limitations. Compare examples with a similar problem to yours, and treat a previous client’s outcome as context rather than a prediction for your website.
Start with your own evidence
If you want the same approach applied to your site, the SEO audit with implementation starts from what your data actually shows. For a new build or rebuild, see Website Design & Development.
Last reviewed: 21 September 2026 by David, founder.