Here’s the uncomfortable truth nobody on LinkedIn wants to admit out loud: roughly 90% of the new content hitting the web in 2026 is at least partially machine-written. The flood happened. We’re already swimming in it.

And the question for any serious marketing leader has shifted. It’s no longer „which AI tool should I use?”. It’s the question Róth puts on page one of Signal Over Noise: will your brand be the signal — or just more of the noise?

„AI didn’t commoditize expertise. It commoditized fluency. Visible human judgment is now worth more than it has been in two decades.” — Miklós Róth, Signal Over Noise (Ch. 4)

That single sentence is, honestly, worth the price of the book. It reframes the entire panic. The thing AI got really good at — sounding smart — is exactly the thing customers have learned to distrust first.

Plenty of „AI marketing books” pretend to be for everyone. This one isn’t. It’s surgically aimed. To save you time, here’s how the audience clusters break down — and which intent each chapter actually serves:

It’s a simple loop. The cheaper content gets to produce, the harder it is for any one piece to matter. Output goes up. Attention stays flat. Trust drops. So you produce more, hoping volume rescues you. It doesn’t. You’ve just built a louder version of yourself for an audience that stopped listening months ago.

Brands that scaled blindly in 2024 and 2025 are quietly bleeding now. Organic CTR — down. Branded search interest — flat or declining. Average session depth — collapsing. Signal Over Noise opens with this autopsy and refuses to flinch.

Note from Róth: these aren’t vendor numbers. The book footnotes each one with primary sources — Pew, Gartner, Edelman, internal SERP audits. That alone separates it from 80% of the genre.

SICT is the spine of the book. Pronounced „sick-tea” — yes, Róth jokes about it on page 12. Four pillars. None optional. Skip one and the whole organism collapses, which is exactly what’s happening to most „AI-first” content operations right now.

What I appreciate about the framework — and I’ve stress-tested it on five client engagements since the book dropped — is that it’s diagnostic. You can map any failing content team to one missing pillar and know, almost immediately, where to intervene. Most agencies are over-invested in Structure (they call it „technical SEO”) and dangerously under-invested in Cohesion. Sound familiar?

The blue link is dying. Slowly, then all at once. For 25 years, „discoverability” meant ranking for a query. In 2026, increasingly, it means being the cited source inside an AI summary. Different game. Different rules.

Róth devotes nearly a fifth of the book to the mechanics of Answer Engine Optimization (AEO) and its broader cousin, Generative Engine Optimization (GEO). The chapter on Perplexity’s citation behaviour alone is sharper than anything I’ve read on the platform — and I’ve read most of it.

Three things he hammers on, repeatedly:

• Be the entity, not the page. Knowledge graphs index things, not URLs. If your brand isn’t a recognized entity, you’re invisible to half the new search surface.
• Write in machine-parsable beats. Short factual sentences near the top. Then context. Then nuance. The inverse of how most blog posts are structured.
• Cite or be uncited. Pages that quote primary sources well get cited back at roughly 4× the rate of pages that don’t. Reciprocity is real, even with bots.

This is the chapter I keep sending to clients. Generative AI has trained consumers — fast, and largely unconsciously — to recognize „the AI sound.” You’ve felt it too. That smooth, slightly empty cadence. The bullet lists where every item is suspiciously the same length. The em-dashes everywhere (yes, I know, look at me).

Customers have built an implicit ladder. From most trusted to least:

1. Direct personal experience
2. Trusted human in their network
3. Named expert with verifiable track record
4. Branded human-signed content with thick context
5. Branded content (no signature)
6. Anonymous content with clear authorial voice
7. Anonymous content with the AI-pattern signature ← actively reduces trust

Notice that last line. It doesn’t just fail to help — it actively damages the brand attaching its name to it. That’s the part most agencies haven’t internalized yet. Publishing more sloppy AI content is worse than publishing less.

If you came in looking for a prompt engineering for marketers book, fair warning: Róth treats prompting as the kindergarten level. Useful, necessary, but not where the leverage lives. The ladder he draws looks like this:

The Human-in-the-Loop matrix on page 142 is a keeper. Two axes: failure cost on one, judgment structure on the other. Four quadrants — automate, supervise, co-create, reserve. Print it. Stick it above your desk. It will save someone on your team from approving a tone-deaf campaign in Q3.

This sounds dry. It isn’t. The argument Róth makes — and I’ve never seen it framed quite this way before — is that data hygiene has graduated from an IT problem into a creative one. The reason: personalization is now the highest-leverage move in marketing, but the gap between what your data can support and what your AI tools can technically execute is precisely where „creepy” lives.

Send a personalized email referencing something the customer never knowingly shared with you, and you’ve just trained them to distrust your brand. Forever. That’s not a tech failure. That’s a creative judgment failure dressed up in a CRM.

EU AI Act. GDPR. ePrivacy 2.0. Most marketers groan at the words. Róth reframes them as offensive infrastructure. Companies that build proper governance in early — disclosure flows, model documentation, opt-in audit trails — capture market share when underprepared competitors have to retreat from regulated channels entirely.

And here’s the part that genuinely surprised me: small teams are now structurally advantaged. Production capability has decoupled from headcount. A sharp three-person agency running SICT can out-execute a 50-person legacy shop trapped in functional silos and politics. That’s not a slogan. I’ve watched it happen, twice, this year alone.

„The story of marketing in 2026 isn’t ‘big AI eats small marketers.’ It’s ‘small AI-fluent operators eat slow incumbents.’ If you’re under 10 people, this is the most exciting moment in two decades.” — Signal Over Noise, Ch. 11

I want to be fair, because most reviews of this kind are pure cheerleading. Signal Over Noise isn’t for everyone.

• If you’re a complete beginner who doesn’t know what a content brief is — start somewhere gentler. This book assumes you already operate.
• If you want a step-by-step „do this in ChatGPT” recipe book — go elsewhere. Róth deliberately stays substrate-agnostic. He’ll tell you how to think, not which prompt to paste.
• If you need fresh, recent vendor screenshots — print books age fast. The frameworks will hold; the tool examples will date by 2027.

Everyone else: yes. Buy it. Mark it up.