Home › Case studies › Case study library
SEO case studies: real numbers, stated limits
Four documented studies — three first-party studies from GSA and Gear properties plus one anonymised US local-SEO market snapshot. Each one shows what was measured, what changed, what stayed broken, and what the result does not prove. No invented clients, no “grew traffic 400%” screenshots.
Search visibility, sitewide audits and US local SEO · evidence dates printed on every study
What this library is
A case study is only worth reading if you can check it. So each study here answers four things before it shows a single number: what was measured, on which property, over which window, and what it cannot prove. Then it shows the evidence.
That order matters. Most agency case studies open with the win and bury the conditions — which makes them impossible to compare and easy to fake. We do the reverse. Where a study reports something unflattering, it stays in.
What you will find here: four studies covering search visibility, technical audits and a US local-SEO market snapshot.
What you will not find: invented client logos, screenshot charts with no source, or a claim that a change caused revenue. None of these studies prove rankings, traffic, leads or sales for your site — they prove what happened on theirs.
Why we publish our own results first
An agency can run any method once on a client site and call it a success. It is much harder to run the same method on your own property, publish the number, and keep publishing it when the number is bad.
Three studies below are our own properties. That is deliberate. When your own Search Console shows impressions falling by half because one comparison page slipped off page one, your own audit shows sixty-six orphan pages concentrated in a single location silo, and your Gear property has a perfect technical score and no visitors at all — those are the studies worth showing a prospective client, because they demonstrate the judgement you are actually buying. The Phoenix study answers a different question: what a US operator can safely infer from market context before treating demand as proof of selection.
When a case study means something — and when it does not
A before-and-after number is evidence only if the property was measured the same way, on the same scale, across two comparable windows, with nothing else changing in between.
- Meaningful: same property, same measurement method, two clean windows, the change isolated to one thing.
- Weaker but usable: a shift in what you are being shown for, where the cause cannot be isolated — you can report the movement, not the cause.
- Not evidence: an overlapping window, a redesign mid-comparison, a single spot-check dressed up as a trend, or a score used to describe a device it was not measured on.
Every study below labels which kind of evidence it is. The studies report observations and limits rather than turning a single score or snapshot into a promise.
The five studies
Each row gives the headline, the reason it is worth reading, and the problem we did not manage to close. Follow any title for the full readout.
Five fix passes, one error left, no score games
We pointed Semrush Site Audit at our own site and published the before and after. Broken pages went to zero, site health moved from 75% to 94%, and the error count fell from 641 to 1. The remaining 491 warnings are left in the log because not every warning is a growth decision.
What we did not resolve: The page count also fell from 491 to 324 because the second scan followed the current live set. That is a scope change, not a claim that 167 pages were fixed.
More clicks from fewer impressions
We compared two adjacent 90-day Search Console windows on our own domain. Clicks and CTR rose while impressions and position fell. Read together, that usually means the site is being shown for fewer but more relevant queries — not that it got penalised, and not that one edit did it.
What we did not resolve: The domain and site structure changed after March 2026, so this is direction, not a controlled experiment.
A clean technical baseline, with the messy part left in
A read-only audit of every published page and post on our own site. Critical faults and non-blocking warnings are kept apart, so the fix list is ordered by consequence rather than by warning count — which is the exact mistake most audit tools encourage.
What we did not resolve: Orphan candidates are not automatically defects. 57 of the 66 sit in one location silo, which is a linking question rather than a fault.
Zero errors is not the same as working
A read-only audit of a newer sub-property with a small published surface. Every technical check passed — and it still had no search rows. We publish that because a “100% clean audit” is often just a site nobody can find, and dressing that up as a win is how audits lose credibility.
What we did not resolve: Zero search rows can mean an early property, a configuration issue, or a coverage problem. The audit cannot tell you which.
Phoenix local SEO: demand is not selection
Phoenix, Arizona home-service market
US market snapshot, not a ranking
Customer demand and operator intent are different
A dated snapshot across Phoenix plumbers, pest control and movers. Strong customer-service demand does not automatically prove demand for local SEO services or an easy path into the local results. Owners need to separate the customer, operator and selection questions.
What we did not resolve: The snapshot does not measure calls, enquiries, sales or revenue and does not establish organic ranking difficulty.
What these studies have in common
- Critical is separated from cosmetic. A blocking fault and a missing meta description do not belong in the same list or the same sentence.
- Read-only first. Every audit here observes the site. Nothing is changed on a property until the measurement that justifies it exists.
- Zero is reported as zero. A clean audit that produced no visibility is a finding, not a victory.
- Limits are printed, not appended. The caveat sits next to the number it qualifies, so the two cannot be read apart.
How to read a case study before you trust it
Whoever wrote it, run these four checks in order. Any study that fails one is marketing, not evidence.
- Is the property named? If you cannot see the domain, the numbers cannot be checked or compared.
- Are the windows stated with dates? “Over 90 days” is not a window. Two sets of dates are.
- Is the method consistent between the periods? A change of tool, device or threshold invalidates the comparison.
- Does it say what it cannot prove? If nothing is listed, nothing was checked.
Frequently asked questions
What is an SEO case study?
An SEO case study documents a real measurement on a real property: what was wrong, what was changed, what the numbers were before and after, and what the result does not prove. It is evidence, not a sales brochure. The ones on this page all belong to us or to a client who agreed to be named, and each one states its measurement window and its limits.
Why would an agency publish its own bad numbers?
Because a case study that only shows wins tells you nothing about how the agency behaves when something goes wrong. Our Search Console study shows impressions and average position falling. Our Gear study shows a technically perfect site with zero search visibility. We publish those because that is the honest version, and because spotting those exact problems is what you are paying an audit for.
When is a case study result actually meaningful?
When the same property was measured the same way across two comparable windows, and nothing else changed in between. Most published SEO case studies fail that test: the site was redesigned, the content changed, a migration happened, or the two windows overlap. Where those caveats apply to ours, they are printed on the study itself.
Do these case studies show client results?
Three studies are first-party evidence from GSA or Gear properties, and one is an anonymised US market snapshot. We label the evidence boundary on each page and do not present market context as a client engagement or result.
Can I use these to judge whether an SEO audit is worth buying?
Yes, and that is the point. Read how we separate critical faults from warnings, how we handle a clean audit that produced no visibility, and how we state what a measurement cannot prove. If you want the same treatment on your own Search Console data, that is the Revenue-Leak Audit.
Related GSA destinations
Where to go next, depending on what you came for:
- Case studies overview — the two narrative studies and the live numbers behind this library.
- SEO audit report template — the structure we write every readout into.
- Free website SEO audit — run the diagnostic on your own site.
- The Revenue-Leak Audit — the paid readout built on your own Search Console data.
- India hub — the India-specific service and pricing.
Want this measured on your own site?
Start with a free website audit, or run the Revenue-Leak Audit on your own Search Console data and get a ranked, page-by-page fix list — the same discipline these studies are built on.
Dated guidance. No ranking, traffic or revenue guarantee.