01 Insights

Fewer posts. Each one carries something I had to do the work to know.

Field notes from inside the practice: AI-search experiments with the raw numbers, the case for treating search as one full-stack system, and teardowns of the systems I install for clients. Written in the first person because you should know exactly whose judgment you're reading.

02 The bar
FIG. 01 — WHAT EVERY POST HAS TO CLEARNO EXCEPTIONS · THE FILTER IS THE MOAT
01

An artifact, not an article

A framework, a dataset, or a build teardown you can use without hiring me. If it only works as an ad for the practice, it doesn't ship.

02

Original data or a real build, every time

A number from an install, an experiment run across live AI engines, or the architecture of a system that actually runs. No think-pieces assembled from other people's think-pieces.

03

A stated point of view

At least one line a competitor would be unwilling to publish. Balanced coverage is what you write when you have nothing to defend.

04

Names the negative space

What doesn't work gets said out loud — the vendor tactics, the metrics that mislead, the places the automation quietly degrades. Trust is built by naming failure, not hiding it.

This is why there are three posts here and not thirty. The ceiling is two a month, and most months it's one. A post that couldn't exist without doing the work is expensive to write and hard to copy — which is exactly the point.
03 Latest
FEATURED · MOST RECENTOriginal data · Methodology you can steal
AI & SearchOriginal data

I tracked AI answers like a pollster. Here's the methodology — and the scoring sheet.

I ran 40 buying-intent prompts across ChatGPT, Perplexity and Google's AI Overviews, on repeat, with fixed sampling and published confidence intervals. Who gets cited, how little the engines agree with each other, and a reusable sheet you can run against your own category this afternoon.

Read the study →
FIG. — CITATION OVERLAP ACROSS ENGINES (ILLUSTRATIVE)
Three overlapping circles representing which brands each AI engine cites for the same set of buying-intent prompts. The circles for ChatGPT, Perplexity and AI Overviews overlap only slightly in the center, illustrating that the engines largely cite different sources for the same questions.CHATGPTPERPLEXITYAI OVERVIEWSSMALLSHAREDCORE
The engines mostly cite different sources for the same question. Real overlap figures are in the post.
04 All posts

Three clusters, one bar

Everything lives in one of four categories. The list is reverse-chronological; filter it if you're only here for one thing.

INSIGHTS LEDGER — NEWEST FIRSTLAUNCH SET · 1 LIVE · 2 QUEUED
01

I tracked AI answers like a pollster

A prompt-tracking methodology with fixed sampling and published confidence intervals, run live across ChatGPT, Perplexity and AI Overviews — plus the scoring sheet, free to steal.

AI & SearchOriginal data
[ DATE ]
Read →
02

The honest GEO buyer's guide

Most "AI-visibility" vendors are buying mentions, not building them. A checklist a marketing director can run against any vendor's deliverables before signing — where the mentions came from, and whether they survive a source-quality audit.

AI & SearchBuyer's guide
[ SOON ]
In progress
03

Teardown: the GSC reporting engine I install

The full build — architecture diagram, honest running costs and limits, and sample output — of the Search Console reporting engine that ends the weekly "what's working?" meeting. Built, not advised.

Systems & AgentsBuild teardown
[ SOON ]
In progress
04

Next in the queue — the full-stack search thesis, and the context-layer install

Published when they clear the bar, not to fill a calendar. The list stays short on purpose.

[ SOON ]
In progress
No posts in this category yet — it's on the bench, not abandoned. Try another filter.
05 Who writes this
[ FOUNDER
PHOTO ]
The practice, in one person

Marvin Viachica

I run Searchline Partners — a senior search and demand generation practice for B2B companies. I've spent years leading search and demand for US companies in-house, and I build the systems I write about: reporting engines, AI-visibility audits, content architectures, installed in the client's own stack.

The writing here is the same judgment you'd hire. No ghostwriters, no content team — if my name is on it, I did the work and I'll defend the number.

More about the practice →  ·  LinkedIn
06 Get the next one

One email when there's something worth your time.

No drip sequence, no gated PDF, no "10x your pipeline" nonsense. When a post ships — the ones with the real numbers and the stated opinion — I send it once. That's the whole newsletter. Unsubscribe in one click, and I won't take it personally.

No spam · No sharing your address · ~1–2 emails a month, max