Portfolio / Case study

Case study · Affiliate website · 10 weeks

Traffic up. Site rebuilt.
Every affiliate click visible.

An affiliate website had lost about 90% of its Google traffic, and nobody could say why. In ten weeks we found the cause and proved it, cut the site down to what earns, rebuilt it, and made every affiliate click measurable.

Affiliate & publishing · Revenue recovery · Search & AI visibility · Analytics & attribution · AI content · Platforms & migrations

Traffic

2.7×

Weekly visits from Google in the first month: 602 → 1,612

Month 2: +67% Google visits on a typical day, on a site 85% smaller

Affiliate tracking

525

Affiliate destinations tracked across 995 pages

Every click tied to the page and the spot that earned it

AI content

430

Data-led articles in 2 weeks, in 2 languages

Every figure computed, every quote checked against its source

Search & structure

Day 1

New articles appear on Google the day they publish

30,369 pages cut to the 4,555 worth keeping

Speed

1.9 s

Phone load time, down from up to 5.4 s

Inside Google’s 2.5 s threshold, on all 5 key page types

Analytics

55%

Of the visits analytics reported were bots, now filtered out

Real engagement: 22.4%, not the 8.9% reported

Client details withheld. Every figure comes from the client’s own data and was re-checked at the source. Traffic growth is measured from the low point: the first week of the engagement against the fourth. Every figure is a record of what happened, not a forecast. Load times were measured on a full copy of the site before launch.

How we got there

Six results,
each with its cause.

2.7×

Traffic grew

What held it back

  • An automatic clean-up job deleted articles two days after publishing, including 670 on Google’s first page
  • 46% of one week’s article visits went to pages that were later deleted
  • A site update had hidden most article pages from Google. We traced it and dated it to within 4 days; the client’s own team removed it in week one

What we did

  • Stopped the deletion: 2 days → 7 days, then switched off for good
  • Asked Google to re-check 22,951 pages it had dropped
  • Clean web addresses built automatically: new articles averaged position 4.8 on Google within 13 days
  • On Google the same day: one article was seen 547 times in search on the day it went live
  • Month 2: Google visits on a typical day 177 → 296, on a site 85% smaller
525

Affiliate revenue, visible

What held it back

  • The main affiliate banner ran on 298 articles, but nobody could tell which page earned a click
  • 55% of the visits analytics reported were bots, not people

What we did

  • Every affiliate link tracked: 525 destinations across 995 pages
  • Each click recorded with its page and its position on the page
  • One dashboard to switch any placement on or off and see clicks by day
  • Fake clicks filtered out: 81 of 98 suspicious clicks were bots
430

AI content, high quality

What held it back

  • Articles regularly went live too late to rank for the searches they targeted
  • One flagship series: 16 of 16 pages were 43–47-word stubs

What we did

  • 430 data-led articles in 2 weeks, in 2 languages
  • About 1,500 pages refreshed or published in a single week
  • Flagship pages rewritten at 1,235–1,475 words, every figure from data
  • Built-in fact checks: no quote or number goes out unless it matches its source
  • 300 section pages audited (4,575 findings), 255 rewritten
Day 1

Search and structure

What held it back

  • 15,587 pages brought 1 visit between them in 25 days
  • 21,179 old addresses forwarded to the home page, which Google treats as dead

What we did

  • Pages offered to Google: 29,277 → 3,255, only the ones worth ranking
  • 25,814 expired and low-value articles archived, all restorable
  • About 1,025 near-identical list pages cut to 40; 1,955 pages Google was being sent twice removed
  • Rules forwarding old addresses: 27,734 → 220
  • The page list Google reads loads 7× faster: 4.5 s → 0.64 s
  • Every link inside the site checked: 1,974, none broken
1.9 s

Site rebuilt

What held it back

  • Phones waited up to 5.4 s for the main content; Google’s threshold is 2.5 s
  • Every page loaded 2,115 colour styles and used 11

What we did

  • One clear structure for every section: category, sub-category, item
  • A new design, with 9 colour themes readers can choose
  • Styling code −76%; database queries on the main listing 815 → 134
  • Phone load time: up to 5.4 s → 1.9 s on all 5 key page types
  • 3,050 pages compared before and after launch: none got worse
331,829

Bot load removed, tracking restored

What held it back

  • One bot made 331,829 requests a day, more than Google’s crawler and three other major bots combined
  • The site’s own security policy was blocking its analytics and ad tracking

What we did

  • 4 security exposures found and reported privately to the owner
  • The heavy bot blocked: 100% of its requests stopped, 0% of Google’s
  • Tracking restored without weakening the security policy: 0 errors on the page

Month by month

Diagnose first.
Then fix what pays.

Month 1 · Diagnose and measure

Find what broke, prove it, measure it.

AreaWhat we didThe number
Data analysis16 months of Google search data gathered and analysed7.35M rows
DiagnosisMain cause of the collapse traced to one site update, proven 5 waysDated within 4 days
RigourOur main finding attacked before we reported it9 attempts to disprove
ContentAn automatic clean-up found deleting articles after 2 days670 page-one articles lost
AnalyticsBots separated from real visitors55% were bots
TrafficWeekly visits from Google, start of month to end602 → 1,612

The cause of every step of the decline, dated, and analytics that tell people from bots.

Month 2 · Rebuild the foundations

Less site, built to be found.

AreaWhat we didThe number
StructureSite cut to what earns; anything removed can be restored30,369 → 4,555 pages
StructureRules forwarding old addresses cleaned up27,734 → 220
TrafficGoogle visits on a typical day, against month 1177 → 296 (+67%)
SpeedThe page list Google reads made 7× faster4.5 s → 0.64 s
ContentPages refreshed or published in one week, 2 languages~1,500 pages
RevenueAffiliate clicks traced to the page that sent them325 links, 184 articles

New articles appear on Google the day they publish. One, live at 09:18, was seen 547 times in search that day.

Month 3 · Grow

The first 10 days, and the plan.

  • Redesign live: phone load 1.9 s on the key pages, 3,050 pages checked, none worse
  • Affiliate dashboard: 525 destinations tracked by page and spot; bot clicks removed
  • 147 new articles; 300 section pages reviewed, 81 cut to half their length
  • 53,509 search phrases in one table, from 4 sources: the map for what to write next
  • Next: title and timing experiments, each with its measure of success set before it starts

Delivered carefully

Fast,
because it is careful.

Delivery

101

Releases shipped in 10 weeks, each one reviewed and with a written way to undo it

290 tracked tasks closed

Data depth

7.35M

Rows of Google search data, 16 months, analysed

431,227 searches sorted by intent to buy

Quality gate

3,050

Pages compared, old against new, before the redesign launched

None got worse

Reporting

Every week

A written report on what changed, what it did and what is still unproven

Every claim labelled proven, likely or unknown

How we work

  1. Diagnose with data

    We find the cause before we change anything. Our main finding survived 9 attempts to disprove it before we reported it.

  2. Test on a full copy of the site

    Every change runs on a complete copy of the live site, real data and all, before it goes anywhere near visitors.

  3. Ship small, reviewed changes

    One change at a time, reviewed, with a written way to undo it. 101 releases in 10 weeks.

  4. Report every week

    What changed, what it did, and what is still unproven. 10 reports in 10 weeks.

Is your site earning
what it should?

Tell us where revenue fell, stalled or never arrived. The first month is built to find out why, in numbers — and nothing on your site changes until the causes are proven.

Prefer to write directly? enable JavaScript to see the address

Talk to an engineer

No sales theater. Tell us where your operation feels slow, repetitive, or difficult — an engineer reads every message.

Your message goes straight to our engineers at our address.