01 Case file 01 · all work

The traffic number was lying. I found the truth underneath it.

A US bare-metal hosting company was six months into a traffic collapse everyone read as an SEO failure. It wasn't. I joined at the floor, proved the decline was a structural shift in search — not an execution failure — and rebuilt the measurement and strategy around what was actually happening.

SectorBare-metal hosting · VPS · colocation

SizeMid-market

Lines engagedSearch Systems · AI Visibility · Demand

Duration12 months · Jul 2025–Jun 2026

02 The situation

What the diagnostic found

WHAT THE HEADLINE NUMBER HID — EARLY 2025CLIENT ANONYMIZED · REAL PROPORTIONS

The panic was reading the wrong number. A large share of the old volume was paid traffic about to be switched off and low-intent visitors who bounced on arrival. Underneath, organic had fallen 59% since January and bottomed at 8,900 sessions in July — my first month. And the analytics themselves couldn't be trusted: key events were inflated roughly 13×, and two competing purchase events were double-counting revenue.

Most of the stakeholders read the organic decline as an SEO execution failure. My first job was to prove it was something else — and to fix the instruments before I trusted a single number.

I don't chase the number; I diagnose what's underneath it. Here that meant separating three things the headline had fused together: artificial volume that was leaving anyway, a visitor-quality problem the high session count was masking, and a real structural shift in how search sends traffic. Only one was worth reacting to — and reacting to the wrong one would have cut exactly the budget that was working.

Rankings held. Clicks collapsed. Everyone read that as failure — it was a structural shift, and I named it the Great Decoupling.

Google's AI Overviews were absorbing an estimated 60–70% of informational search clicks across the entire hosting and infrastructure category — the point where ranking position and traffic stop moving together. That reframe changed the whole investment thesis: stop feeding a channel AI was consuming, and shift toward commercial, decision-stage pages AI can't absorb, while competing for citations inside the AI answers themselves.

03 What got installed

Fix the instruments, then build the system that reads them

Reports are disposable. The durable asset I installed is a repeatable growth-intelligence system — it separates what the data shows from what I infer, scores every opportunity the same way, and remembers what it found. In dependency order: trustworthy measurement first, because everything downstream inherits its errors.

INSTALL RECORD — IN DEPENDENCY ORDERSYSTEM & STANDARDS IN CLIENT ACCOUNTS · OWNED BY THE CLIENT

I-01

Analytics integrity rebuild

Fixed the instruments before trusting a reading. A GA4 audit found key events inflated roughly 13× by a broken funnel and two competing purchase events double-counting revenue — so every ROAS figure was unreliable. I set a single clean ecommerce tracking standard: the first trustworthy foundation for attribution in the stack.

Line Search Systems + Demand

I-02

Growth-intelligence system

The analyst I'd hire if I had the budget, made reusable. A drop-in context pack (ICPs, competitors, priority pages, voice) routed into six analysis modes; an evidence engine orchestrating DataForSEO, Google Search Console and Firecrawl with cost discipline; and scored, remembered output — every finding tagged Observed / Comparative / Hypothesized, every opportunity ranked (Impact × Confidence) / Effort, appended to a findings log so nothing gets re-litigated.

Line AI Systems & Agents

I-03

AI-answer citation baseline

Turned "we're invisible to AI" into a ranked build list. Across five commercial clusters the site held a true zero-citation baseline in ChatGPT and AI answers. I mapped exactly who was winning each answer (often disjoint from the Google top 10), the citation-winning content archetype per cluster, and the one on-page gap no competitor had shipped.

Line AI Visibility

I-04

Commercial content architecture

Same rigor, opposite vehicles — because the data said so. Five ICP content plans (Gaming, SaaS, Fintech, Healthcare, Streaming), each matched to how its SERP actually behaves. Streaming had no AI Overview and rewarded a landing-page pillar; SaaS and Fintech were AI-Overview-saturated and needed blog-pillar-first plays. The pivot: away from informational posts AI was eating, toward decision-stage pages it can't absorb.

Line Search Systems + Demand

I-05

Leadership reporting cadence

Kept a declining vanity metric from being misread as failure. A monthly readout that translated the Great Decoupling into plain-English context — so a falling session count was understood as a deliberate shift from vanity volume to commercial intent, and budget stayed pointed at what was working.

Line Demand

The system runs in the client's accounts. The tracking standard, the findings log, and the reporting cadence are theirs to keep — the analyst I'd hire if I had the budget, made reusable and left behind. That's the ownership test every build here has to pass.

04 What it did

The numbers, with their timeframes

Every figure below is drawn from the client's GA4, January 2025 through June 2026. Fixed-width type on this site is reserved for numbers that happened — these happened.

MONTHLY ORGANIC SESSIONS · JAN 2025 – JUN 2026SOURCE: CLIENT GA4 · MARKER = HIRE DATE

+197% off the floor, into a headwind. Organic tripled from its July nadir by September and held near 20K through early 2026, while AI Overviews compressed informational clicks across every content-heavy site in the category. The final months bend back down with the industry — the honest section below says why that's the expected shape, not a reversal.

Organic recovery

26,400▲ +197%

from 8,900 sessions · the July floor

Jul → Sep 2025, against expanding AI Overviews

Engagement rate

57%▲ +16 pp

from 41% pre-hire average

Sustained the rest of the engagement

Session duration

340s▲ +88%

from 172s average

Real evaluation, not bounces

Ecommerce revenue · QoQ

$194K▲ +88%

from $103K the prior quarter

Q3 → Q4 2025; Oct & Nov the two highest months in the dataset

AI-answer citations

5 clustersbaseline set

from a true zero in AI answers

Winners, citation archetype and the one open gap mapped per cluster — a ranked build list

Analytics integrity

1× clean▼ de-inflated

from key events inflated ~13×

Double-counted revenue resolved to a single trustworthy tracking standard

THE HONEST LINE — WHAT THE DATA DOES AND DOESN'T SAY

The later organic decline is industry-wide, not a strategy reversal. AI-Overview expansion is compressing informational traffic for every content-heavy hosting site. The tell that it's structural: through the same window, the things I controlled — engagement rate, session duration, revenue-per-purchase, bounce rate, analytics integrity — all improved or held. That's the signature of a deliberate shift from volume to intent. Two more caveats I'd flag before anyone else does: the old peak was partly inflated by paid traffic that ended mid-2025, so the fair comparison is smaller and higher-quality; and the biggest bet is still maturing — four AI-resistant commercial ICP pages only entered indexation near the end of the engagement, so their organic payoff sits after the window this data covers.

05 Next step

Is your traffic number telling you the truth?

If your organic is moving in ways nobody can quite explain, or you suspect AI search is quietly taking your traffic, that's exactly the work. I fix the instruments first, then tell you whether it's your team or the market — and I can prove which. The diagnostic is where it starts.

Get in touch

CAPACITY — [N] BUILD SLOTS OPEN · [QUARTER]