Commodity Content Is Dead: Google Just Drew the Line in Milan

TL;DR: Google stopped being polite. At Search Central Live in Milan, it split content into two buckets: commodity and non-commodity. Commodity content is anything the open web already said in a thousand interchangeable ways. Non-commodity content carries experience, original measurement, and operator scar tissue. Google is not primarily punishing "AI writing." It is punishing content with no lived advantage. If a model can generate a cleaner version of your page in one second, AI Overviews will answer upstream and your site becomes optional.

I am James, CEO of Mercury Technology Solutions. Hong Kong — August 2026

Stop producing content the entire internet can already produce.

Google finally said the quiet part in public.

At Search Central Live in Milan, the framing was blunt enough that the old content-farm cope stopped working. Content now sits in two categories:

1. Commodity content

2. Non-commodity content

That is not a style guide update. That is a survival filter.

For sites that lived on "rewrite what already ranks," this is not a temporary ranking wobble. It is the beginning of traffic going to zero with a clean conscience from Google's side.

Google Is Not Hunting AI. Google Is Hunting Emptiness

The industry keeps misreading the war.

The lazy narrative says: Google is cracking down on AI content.

Wrong target.

Google is cracking down on content with no experiential surplus.

If your article is a polished rearrangement of public consensus, the model layer can synthesize a better version instantly. Once that happens, AI Overviews has no reason to send the user to you. The answer is complete before the click.

This is the new fork:

• Cited by AI because you add something the model cannot cheaply invent

• Replaced by AI because you are just another compression of the same commons

In client tests over recent cycles, one split keeps repeating with ugly consistency:

Pages with first-hand testing, proprietary data, or operator-specific failure modes get selected. Pages with generic explainers get summarized and skipped.

Not "sometimes." As a pattern.

The deciding variable is no longer word count, heading hygiene, or synonym coverage. It is whether the page contains non-replicable residue from reality.

Commodity Content, Defined Without Mercy

Commodity content is not "bad writing."

Commodity content is content whose information advantage is approximately zero after the public web and foundation models already exist.

Examples:

• "What is CRM?" primers with no implementation scar tissue

• "10 tips for better SEO" assembled from other tip lists

• Tool roundups with no benchmark protocol

• Industry definitions rewritten for the 400th time

• "Ultimate guides" that never leave the realm of safe consensus

These pages used to survive because search was a document-retrieval game. Rank the best-formatted summary, harvest the click, monetize the session.

AI Search breaks that economy.

When the SERP can generate the summary itself, the summary page loses its reason to exist.

Your competitor is no longer another blogger. Your competitor is the answer engine's synthesis layer.

If that layer can outperform you on clarity, breadth, and speed, you are not competing. You are volunteering to be source mulch.

Non-Commodity Content Is Not "Longer." It Is Costlier to Fake

Non-commodity does not mean purple prose. It does not mean more emojis. It does not mean "add a personal story paragraph" as decoration.

It means the page carries costs that lazy producers will not pay:

• you ran the test

• you paid for the failure

• you measured the delta

• you held the operational constraint

• you can show the path-dependent judgment call

In GEO / LLM SEO terms, non-commodity content is high citation utility under uncertainty.

Models and answer engines need claims they can lean on when generic consensus is not enough:

• what broke in production

• what the benchmark actually measured

• what changed after a policy shift

• what the local market constraint does to the textbook playbook

• what numbers moved after a specific implementation

That is why first-hand tests and exclusive data are no longer "nice editorial flavor." They are selection fuel.

Without them, you are asking AI to cite a mirror.

Search Console Is Opening the Black Box on Purpose

There is a second signal people are underweighting.

Google is exposing more AI-related controls and click data inside Search Console.

That is not random product clutter.

It is a strategic message:

AI search is no longer a mysterious side effect. It is a surface you are expected to operate against.

When a platform starts giving measurement, it is telling serious operators the game is legible enough to optimize.

Black boxes get mythology. Instrumented surfaces get process.

If you still treat AI Overviews as weather — something that "happens to traffic" — you are already late. Google is handing out cockpit data. Use it.

The Three Moves That Matter Now

Forget the 37-point content checklist. Run these three.

1. Kill pure information pack content

If the page exists only to restate known facts more cleanly than the last guy, stop production.

AI already won that format.

Your "complete beginner guide" is not an asset. It is a future zero-click casualty with a CMS ID.

Replace information packs with decision assets:

• what we tested

• what failed

• what we would repeat

• what only looks smart in decks

• what changed the metric

2. Answer in the first 60–100 words with machine-usable precision

AIO does not need your brand journey paragraph before the point.

If a query has a direct answer, put the direct answer immediately:

• plain language

• specific claim

• bounded context

• no throat-clearing

This is not "writing for robots" in the old keyword sense. It is raising citation probability by reducing extraction friction.

Think of the opening as an answer packet, not an introduction.

If your first screen is vibe, biography, and suspense, you are optimizing for human patience in a machine-selected environment. Bad trade.

3. Publish scars and test residue

The content Google is hungry for is the content most teams avoid because it is operationally expensive:

• the migration that destroyed rankings for 11 days

• the schema change that improved entity consistency but not clicks

• the sales script that increased demo quality and lowered volume

• the local market constraint that makes US playbooks fail in Hong Kong

• the benchmark table nobody else has because nobody else ran the protocol

This is non-commodity.

Not because it sounds authentic. Because it is hard to counterfeit at scale without doing the work.

Experience is not a tone. Experience is an audit trail.

The Strategic Reframe

Most teams are still asking:

"How do we protect organic traffic from AI?"

Wrong question.

The better question:

How do we become the source AI prefers when it has to justify a non-trivial answer?

Those produce opposite content systems.

Traffic-protection mode creates more pages, more synonyms, more defensive publishing. Citation-utility mode creates fewer pages with higher irreplaceability.

In an AI Overview world, volume without residue is a liability. Every commodity page is another opportunity for the system to answer without you.

So stop treating AI as a thief of clicks. Start treating AI as the largest distribution layer that decides whether your entity is source material or background noise.

What This Means for GEO Teams

If you are doing LLM SEO / GAIO / SEVO seriously, update the acceptance test before publish:

1. Can a frontier model generate 80% of this without our proprietary inputs?

If yes, reject.

1. Where is the first-hand evidence?

Test, dataset, operator decision, customer constraint, measured before/after. No evidence, no ship.

1. Is the answer extractable in the first screen?

If the useful claim starts after 250 words of context, rewrite.

1. Does this page make an AI system more correct by citing it?

If citation does not improve answer quality, you built decoration.

1. Are we measuring AI surface outcomes, not only classic sessions?

Impressions, AI-referred behavior, branded lift, assisted conversions. Old dashboards lie by omission.

This is how content strategy stops being editorial self-esteem and becomes systems work.

The Uncomfortable Part

A lot of "content engines" were never engines. They were rearrangement factories with editorial calendars.

That model required one assumption to stay true:

Search would continue sending users to the best-formatted public summary.

That assumption is dead.

Google drew the line in Milan with institutional clarity:

• commodity content is interchangeable

• non-commodity content is differentiated by experience and original signal

• AI surfaces will complete the commodity answer path without a click

So yes, natural traffic will feel different. It should.

If your graph is full of pages anyone can regenerate, the decline is not mysterious. It is the market clearing.

If your graph is full of tested judgments, local constraints, unique datasets, and operator-level specifics, AI stops being the thing that bypasses you and becomes the thing that routes through you.

Bottom Line

Do not ask whether content was written by AI.

Ask whether the content contains anything AI cannot cheaply recreate from the commons.

• Stop shipping worldwide-generic explainers.

• Put the answer up front.

• Write the pits you fell into and the measurements you actually ran.

• Optimize to be cited, not merely ranked.

The creators who win the next cycle will not be the ones who fear AI Overviews.

They will be the ones who made themselves expensive to replace.

Mercury Technology Solutions: Accelerate Digitality.