GEO, AEO, LLMO: untangling a vocabulary invented by the people selling it

by | Aug 26, 2026 | GEO

You have been told SEO is dead, replaced by GEO. Then GEO turned out to be incomplete, and you needed AEO. Then somebody proposed LLMO, because after all these are language models. Some write AIO. Others talk about search everywhere optimization.

Five acronyms in three years. No shared definition. And every time, the same promise: this is a new discipline, your current agency does not know how to do it, here is our offer.

The problem is not the vocabulary. Technical vocabulary evolves, that is normal. The problem is that every new acronym makes it possible to resell the same service to a client convinced he is buying something else.

This article does two things. It shows what these terms actually designate, using what the platforms themselves say about them. And it isolates the three things that are genuinely new.

What Google writes, word for word

Start with the source least suspected of indulgence toward traditional SEO: the official Google Search documentation, in the section devoted to generative AI features.

Google defines both acronyms itself, then writes this.

> From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

The next sentence is the one nobody quotes.

> If you’re considering third-party “AEO” or “GEO” advice or services, review our guidance on evaluating third-party SEO advice.

Google explicitly points companies considering the purchase of an AEO or GEO engagement toward its page on evaluating third-party SEO advice. That is the page that teaches you to spot dubious vendors. It is not a neutral referral.

Provenance marking: this is the official documentation of the company that operates the engine in question, not a spokesperson’s remark. The level of evidence is high on what Google’s position is, and nil on what actually works, since Google has an obvious interest in nobody claiming to have mastered its results.

What the literature says about the vocabulary itself

The critical survey of July 2026, which examined forty-five studies published between November 2023 and July 2026, opens on a finding worth taking seriously: terminology, metrics and standards of evidence remain heterogeneous across the entire corpus.

Forty-five academic papers, three years, and no common vocabulary. No shared definition of what is being measured, and no agreement on how to measure it.

When a discipline cannot name its own object after three years, it is generally not because the object is too rich. It is because the object is not stabilized yet, and the names travel faster than the results.

The five acronyms, and what they claim to designate

Acronym	What it claims to designate	What actually sets it apart
SEO	Being found on search engines	The original term, forty years of use
GEO	Being cited in generated answers	Nothing structural. Google files it under SEO
AEO	Being the answer an answer engine selects	Predates generative AI: the term already covered optimized snippets
LLMO	Optimizing for language models	No published operational definition. A purely commercial term
AIO	Optimizing for AI-generated summaries	Describes a display surface, not a method

Not one of those five lines describes a technique the other four would ignore. They describe different display surfaces, reached through the same pipeline.

The three things that genuinely are new

It would be dishonest to conclude that nothing has changed. Three things are authentically different from pre-2023 SEO, and none of the five acronyms names them correctly.

The first: query fan-out. A generative engine does not handle your query as written. It breaks it into several internal queries, retrieves documents for each of them, then synthesizes. A March 2026 analysis covering roughly fifteen thousand original queries, which became 43,233 queries after fan-out, with more than five hundred thousand pages retrieved, documents that mechanism. Provenance: a study published by a tool vendor, with the methodology described. Practical consequence: you are no longer optimizing for a query, you are optimizing for a set of queries you never see.

The second: measurement changes nature. In classic SEO, a position is a number. In generative search, the same question asked twice does not return the same answer. An April 2026 paper puts it without ambiguity: visibility is a distribution, not a point. That is not a methodological detail, it invalidates nearly every commercial report in the sector, all of which present a single measurement as a result.

The third: the shift toward third-party sources. A study conducted across several verticals and several languages establishes a systematic and marked bias toward earned media, meaning authoritative third-party sources, at the expense of the content you publish on your own site. The contrast with Google’s more balanced mix is sharp. If that holds up, the useful work is not on your site.

Those are the three novelties. Notice what they have in common: none of them translates into an on-page service billable by the month.

What the other platforms say, which is to say almost nothing

Anthropic’s public documentation describes three distinct crawlers: ClaudeBot for training, Claude User for user-triggered requests, Claude SearchBot for search. It specifies respect for robots.txt and crawl delay, the absence of CAPTCHA circumvention, and explicitly advises against blocking by IP address.

What it does not contain anywhere: a single content optimization recommendation. The only lever officially offered to publishers is access control. Allow or refuse.

The same holds for the other large players. None publishes a generative optimization guide, none validates the existence of a separate discipline. Google goes further and warns against the people selling one.

Why the acronyms keep multiplying anyway

A survey conducted in March 2026 among 205 American executives and marketing leaders supplies the economic explanation. Provenance: a commissioned survey, relayed by the trade press, self-reported sample.

Fifty-five percent of respondents now have a GEO budget line. Among them, seven in ten devote between eleven and twenty percent of their budget to it. One agency surveyed recommends a pilot budget worth one and a half to two times the existing search budget.

Two details from the same survey are worth all the rest. Not one of the practitioners surveyed disclosed an amount in euros or dollars. And one of the executives quoted sums up the reality this way: if you were spending ten thousand dollars a month on SEO, half of that now covers GEO as well.

This is not a new discipline. It is a reallocation with a new name.

What you do tomorrow morning

Take the next proposal that uses one of these five acronyms and put three questions to it, before you read the price.

Which precise action does this engagement contain that my current SEO does not? If the answer lists schema markup, an llms.txt file and paragraph restructuring, you are paying twice for the same thing under two names.

How do you measure, and how many times? If the answer is once a month, the answer is no, because a single measurement of a random variable tells you nothing.

What share of the work happens off my site? If the answer is zero, the engagement is aimed at the place where the literature says the effect is weakest.

Vocabulary is not neutral. A new acronym lets someone sell new expertise on an object that has not changed much. Google says so in its own documentation, and Google has no interest in defending SEO agencies.

Sources

<strong>LaFactory</strong> measures AI visibility with a published protocol: repeated measurements, paraphrases, control group. No guaranteed placement, ever. Contact us to scope an audit.