SEO case studies

SEO Case Studies With the Raw Exports Attached

Most agency case studies are a percentage and a logo. These are the actual Search Console and Bing exports, dated SERP screenshots, and a written methodology you can check. Where a number could not be evidenced, it is not here.

Owned properties, nothing anonymised
6 published studies
AI visibility measured, not asserted

What is in these SEO case studies?

There are 6 SEO case studies here, each on a property I own, published with the raw Google Search Console and Bing Webmaster Tools exports rather than a redrawn chart. Between them they cover organic search, local search, AI citation share and a software directory, over a fixed 16-month window ending August 2026. Where a figure could not be evidenced from an export, it is not on the page.

Every documented property, one table

The aggregate, with the parts that do not add up left out

Six owned properties, one fixed window, one tool per column. The totals row sums only the columns that are genuinely summable. Where a figure does not exist for a property it says so instead of showing a zero.

Search and AI visibility across the six published SEO case studies, 16 months to August 2026, Google Search Console, search type Web.
Property What it is Clicks Impressions Top 31 Copilot citations2
CoimbatoreJunction.in City guide, local search 38,249 3,913,641 113 of 1,000 40,101
ZPlatform.ai AI and SaaS review directory 8,666 1,420,120 102 of 1,000 241,500
VelankanniShrine.in Pilgrimage guide, seasonal spike 42,901 1,342,795 46 of 1,000 16,300
MatchMaker.lk Marketplace, brand as entity 8,250 286,275 120 of 1,000 3,800
SEOTamil.com Tamil SEO education, personal brand 1,374 100,194 222 of 968 336
MakeMoneyOnline.lk Online income education 1,952 71,572 200 of 1,000 not measured
Total 6 properties 101,392 7,134,597 803 of 5,968 302,037 across 5
  1. Top three is counted against the queries Search Console will export, which it caps at 1,000 per property. It is a ratio over the exported set, not a share of everything these properties rank for, and it should not be read as one.
  2. 6 months to August 2026, Bing Webmaster Tools AI Performance, which retains no more than that. The total covers 5 of 6 properties. MakeMoneyOnline.lk has no report, so it is excluded rather than counted as zero.
  3. No traffic estimate from any third-party tool appears in this table. They are unreliable for Tamil, Sinhala and local Indian search, which is most of this portfolio.

Proves: SaaS product

SaaS and Software SEO Case Studies

Software properties, where the query set is product comparisons, alternatives and pricing questions. No commercial metrics are published in this section, and the studies say so.

What this study made me build, and the five predictions behind it

Flagship · Answer-engine visibility at scale

Cited 241,500 times. Clicked 2,500 times. Same platform.

An AI tools directory that Microsoft Copilot cites roughly a hundred times for every organic click it sends, across 1,640 grounding queries, in five languages, with 52% of those citations on commercial-intent searches. One page alone has been cited 108,546 times. This is what it looks like when the answer engine becomes the audience.

ZPlatform.ai · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

241,500

Copilot citations in 6 months

~97:1

citations per organic Bing click, at minimum

1,640

grounding queries cited

52%

of citations on commercial intent

Bing CopilotGoogle AI featuresGoogle organicBing organicMultilingual AI answers

Proves: AI visibility

AI Visibility and Citation Case Studies

Where the answer engine is the audience. Citation counts from the Bing AI Performance report and impression shares from the Search Console Generative AI report, reported separately because they are different measurements.

Small-site AI visibility · Specificity over scale

The smallest property in the portfolio has the highest AI visibility rate

This site earns 1,952 clicks (a fraction of the others) and yet 8.1% of its search impressions come from Google AI features, five times the rate of the largest property. It also holds 200 of 1,000 queries in the top three. The data points to specificity rather than scale, though a single property cannot isolate the variable.

MakeMoneyOnline.lk · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

8.1%

of impressions from AI features, portfolio best

200

of 1,000 queries in the top 3

124

pages appearing in Google AI

4.92%

mobile CTR at position 9.89

Google organicGoogle AI features
Brand as entity · Incumbent capture

Getting recommended by an AI is a different job from getting quoted

87% of this platform’s AI appearances land on one URL, and Copilot cites an average of 2 pages against 35 on my city-guide property. Because the question is “recommend me a site”, not “what is the answer”. It also takes 20% of its Bing impressions from queries aimed at newspaper classified sections that were the category default for generations.

MatchMaker.lk · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

87%

of AI appearances on one URL

2

pages cited, vs 35 on a facts site

~20%

of Bing impressions from rival brands

1 in 5

clicks from the diaspora

Google organicGoogle AI featuresBing organicBing Copilot
AI citation share · Intent over position

Holding 100% of the AI citations for a question three million people ask

A pilgrim guide that is the only source Bing Copilot cites for “velankanni flag hoisting date 2026”, and 84% of citations for the feast itself. It also earns 42,901 clicks at a 3.2% CTR while averaging position 9.1, the clearest proof I have that intent quality, not position, decides what a ranking is worth.

VelankanniShrine.in · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

100%

Copilot citation share on one query

16.3K

Bing Copilot citations in 6 months

3.2%

Google CTR at position 9.1

86

Distinct pages cited by AI

Google organicBing organicBing CopilotGoogle translated results
Local SEO · Answer-shaped content at scale

When the clicks fall and the AI citations quadruple

A city guide with 3.9M Google impressions, 60,500 generative-AI appearances and 40,101 Bing Copilot citations, and one page pulling 396,968 impressions at a 0.09% click rate while being the single most-quoted page on the site. The strongest signal I have that page format shapes which channel a page wins, though with one property I cannot isolate the variable.

CoimbatoreJunction.in · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

3.9M

Google search impressions

60,500

Google AI feature appearances

40,101

Bing Copilot citations in 180 days

0.09%

CTR on the most-cited page

Google organicGoogle AI featuresBing organicBing Copilot

Proves: Organic search

Organic Search and Personal Brand Case Studies

Ranking first, and what ranking first turned out not to buy. The clearest evidence in the portfolio that organic position and citation share are two separate jobs.

Personal branding · Search Everywhere Optimization

Owning “SEO Tamil” across Google, AI Overviews and Bing

How a deliberately small 26-page Tamil-language site took #1 in Google for “SEO Tamil”, got Alston Antony named inside the AI Overview, and earned 336 Bing Copilot citations, the personal-branding play behind Search Everywhere Optimization.

SEOTamil.com · 16 months · Apr 2025 – Aug 2026

Read the case study

Headline results

#1

Google for “SEO Tamil”

222

Queries in the top 3

5,260

Google AI feature impressions

336

Bing Copilot citations

Google organicGoogle AI OverviewBing organicBing Copilot

The same evidence, read against one question

SEO Case Studies by Type

The studies above are named after the properties, because that is what they are. These pages take the same exports and answer one question with them, and each states what its evidence does not show.

In preparation

The rest are being written up

Every property below is being documented to the same standard, in the order the exports justify it. Each one gets its own screenshots and methodology rather than a summary paragraph.

Ask about a specific result

The standard

What has to be true to publish one here

  • The raw platform export is on the page, not a redrawn version of it
  • Every screenshot is dated and shows the property, filter and window
  • The methodology says how each figure was derived, including the thresholds
  • Surfaces I did not measure are marked as not measured
  • No anonymised client with invented percentages

This is a deliberately awkward bar. It is also the reason there are 6 case studies here rather than a page of logos.

Methodology

How to Read an SEO Case Study, Including These Ones

Four checks. They are the ones that separate a case study from a results announcement, and applying them to this page is fair game.

01

Find the window before you read the number

A percentage with no window attached cannot be checked and usually cannot be repeated. Every figure in this section names its window, and where two windows differ (Search Console retains 16 months, the Bing AI report retains 180 days) any ratio between them is published as a floor rather than a rate.

02

Check which report each figure came from

Clicks, impressions and position come from Search Console. Citation counts come from the Bing AI Performance report. AI feature impressions come from the Search Console Generative AI report. They are three different measurements and adding them together produces a number that means nothing.

03

Read the section on what was not measured

Every study here has one. ChatGPT, Perplexity, Claude and Gemini expose no per-site citation report, so presence on those surfaces is not measured and is not claimed. A case study with no stated scope limit is making a bigger claim than its data supports.

04

Look for the part that went wrong

Each study states the hypothesis I started with and where it failed. On the software property the traffic bet failed outright. On the city guide clicks fell while citations quadrupled. A case study with no correction in it has either been edited into a brochure or was never analysed.

Questions

What People Ask About These Case Studies

Are these client case studies or your own properties?
All six are properties I own and run. That is the reason the full Search Console and Bing exports can be published rather than a chart with the client name removed. It is also the honest limit: I chose these subjects, so nothing here proves the method survives a brief somebody else wrote. Client work exists and its numbers are not mine to publish.

What should a credible SEO case study contain?
The raw platform export rather than a redraw of it, a dated screenshot showing the property and filter, a stated window for every figure, the method used to derive any calculated number, and an explicit list of what was not measured. If a study reports a percentage with no window and no export, there is nothing in it a reader can verify.

Why are there only six?
Because the standard above is deliberately awkward. I run more properties than this, and the ones without a full write-up are listed further down this page with their dated metrics rather than promoted into studies they cannot evidence yet. Six studies with exports beats twenty with logos.

Do you have an ecommerce SEO case study?
No. There is no ecommerce property in this portfolio, so there is no ecommerce case study, and I would rather say that than assemble one out of adjacent evidence. The same applies to B2B lead generation: no cost per acquisition or pipeline figures are published anywhere in this section because none of these properties generate them.

Is there anything here for a small business site?
The most transferable finding is on the smallest property. MakeMoneyOnline.lk has 71,572 impressions, the least in this portfolio, and draws 8.1% of them from Google AI features, the highest rate here by roughly five times. One property cannot isolate that variable, so treat it as a lead worth testing rather than a proven rule, but it runs against the assumption that AI visibility needs scale.

Are you building a tool out of this method?
Yes, and it is not finished. A SaaS SEO tool that connects to Search Console, reads the product from its own documentation and changelog, and issues five prioritised actions a week scored on trials and pipeline rather than on audit severity. It is in private build with no launch date, no public pricing and no settled name, so there is no product page to send you to. There is a page in this section that sets out the bet, the six parts and the five predictions that would prove it wrong, and it is deliberately not counted as one of the studies because it has no export behind it yet.

Which of these results transfer, and which do not?
The structural findings appear to transfer: answer the question in the first line, use real HTML tables for anything numeric, one comprehensive page per question, and plain sentences that survive machine translation. The absolute numbers do not transfer at all. None of these properties ran a holdout, so every finding is correlation I can evidence rather than causation I can prove.
In private build Not a case study

The SaaS SEO tool these case studies turned into

The software property above took 241,500 Copilot citations against roughly 2,500 organic clicks, and the traffic hypothesis I opened it with failed outright. No tool I was paying for measured the surface that was actually carrying it, and none of them knew what the product did. So I started building one that scores a week of work by trials and pipeline rather than by audit severity.

It is in private build with no name, no pricing and no launch date, which is exactly why it is not one of the studies above. What is on its page instead is the bet and five predictions, each with the measurement that would prove it wrong.

Read the thesis and the predictions

What is on that page

  • The bet, and where it is still unproven
  • Six parts, each tied to a dated failure of my own
  • The scoring model, written out
  • Five falsifiable predictions
  • What I am not claiming yet

If you want this done on your site

Which of these results maps to which service

Also running

The Other 7 Properties I Own and Rank

Live, earning and structurally measurable, but without a full write-up yet. The figures below are the same ones a case study would open with: Ahrefs metrics from a single dated pull on 11 August 2026, refreshed together, and platform figures that are verifiable on a public page of the property itself. Traffic estimates are deliberately not published. They are unreliable for Tamil, Sinhala and local Indian search, which is most of this portfolio.

Advice.lk

Sri Lanka's biggest advice and guides platform, seven years in.

Content platform

Practical guides for Sri Lankan readers across finance, career, technology and everyday life. The oldest continuously growing authority site I own, and the one that taught me what compounding actually looks like.

  • 46 Domain Rating Ahrefs
  • 2,672 Referring domains Ahrefs
  • 129 Ranked keywords Ahrefs
  • 26 Keywords in Google top 3 Ahrefs
Sri LankaAuthority SiteContent Strategy

DigitalMarketingTamil.com

Ranked #1 across Google, Bing and ChatGPT for Tamil digital marketing.

AI search visibility

My longest-running GEO and AI-visibility test bed, paired with the Digital Marketing Tamil YouTube channel. Tamil is a language most SEO tools do not index properly, which makes it a harsh place to prove a method works.

  • #1 ChatGPT, Google & Bing Platform
  • 5.03K YouTube subscribers Platform
  • 229,346 YouTube views Platform
GEOAI VisibilityDigital MarketingTamil

AIToolsMarketer.com

AI marketing tools scored on a published rubric, not a vendor brief.

AI review platform

Independent reviews of AI software for marketers across seven categories. Every tool is scored on five named dimensions, pricing is taken from the vendor, and commercial relationships are disclosed on the review itself.

  • 55 Tools reviewed Platform
  • 7 Categories Platform
  • 39 Domain Rating Ahrefs
AI ToolsReviewsMethodology

EveryAny.One

Software research with a "last verified" date on every single record.

Software discovery

A comparison platform for business software. Where a fact is not published by the vendor it says "not disclosed" instead of estimating, and every profile carries the date it was last checked. Placement is not for sale.

  • 102 Software profiles Platform
  • 13 Categories Platform
  • 9 Comparison guides Platform
SaaSComparisonAEOData Freshness

DigitalMarketingMind.com

The file comes first. No article wrapped around it, no email gate.

Resource library

Marketing templates, checklists and audit workbooks I actually use in client and employer work: planning docs, competitor analysis, channel audits, proposals and reporting. Free to reuse commercially.

  • 34 Downloadable files Platform
  • 22 Excel workbooks Platform
  • 25 PDF guides Platform
TemplatesFree ResourcesDigital Marketing

AITamil.in

AI explained in Tamil, including the parts AI tools get wrong in Tamil.

Tamil AI education

AI literacy for Tamil speakers, from fundamentals to LLMs, assistant comparisons and reusable prompt patterns. It covers Tamil output quality and the specific failure modes of English-optimised models on Tamil text.

  • 43 Sections published Platform
  • 16 Prompt templates Platform
AI TamilEducationPrompt Engineering

DigitalMarketingCourse.lk

Sri Lanka's free digital marketing learning map, not a course pitch.

Education resource

A reviewed index of 150+ ways to learn digital marketing as a Sri Lankan: free certifications, paid courses worth the money, local institutes, tools, books and salary data. It recommends free options over anything I sell.

  • 150+ Resources indexed Platform
  • 40+ Free certifications Platform
  • 15+ Sri Lankan institutes Platform
Sri LankaEducationFree Resources