Three things probably happened to you this month. A certification course in generative engine optimisation turned up in your feed often enough that you read the curriculum. Someone you follow changed their title to GEO Specialist. And a client or a recruiter asked, slightly nervously, whether you “do GEO” yet.

It is a career question, not a strategy question: a new skill you have to go and learn, or the thing you already do with a new name on it?

GEO is not a new discipline. It is SEO with the weights moved: getting retrieved and quoted matters more, holding a blue-link position matters slightly less, and about four working practices are genuinely new. The rest of the syllabus on any GEO course is SEO you already know. Treating it as a separate specialism is mostly how agencies justify a second line item for work an SEO retainer should already cover.

Will GEO be a thing, then? As a set of practices, already. As a job title held by people who cannot do SEO, probably not. As a billable service category, certainly, because that part never depended on the underlying claim holding up.

Where the term came from, and why there are three of them

GEO has a real academic origin, which is more than most marketing acronyms manage. The paper is “GEO: Generative Engine Optimization” by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, posted to arXiv in November 2023 and accepted to KDD 2024. It sets out a black-box optimisation framework for content visibility inside generative engine responses, introduces a benchmark the authors built called GEO-bench, and reports visibility gains of up to 40%.

That 40% now appears in sales decks with the context stripped off. It measures visibility inside generative responses, on a benchmark the authors built, under their own definition of visibility. Not traffic, not revenue, not a forecast for your site.

Then the neighbours. AEO, answer engine optimisation, predates GEO and came out of the featured-snippet and voice era. AIO means AI optimisation to some people and Google’s AI Overviews to others, so any sentence containing it is ambiguous. Vendors use all three interchangeably and inconsistently, and two agencies will sell you materially different work under the same word. Hence a free diagnostic: ask what GEO means to them, which surfaces they claim to influence, and what they would do on Monday that a technical SEO engagement would not.

What Google actually says, in its own words

This is the load-bearing fact, and almost nobody selling GEO quotes it. Google’s Search Central documentation on AI features states: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” It continues: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.” The same page confirms that sites appearing in AI features are included in overall search traffic in Search Console, reported in the Performance report within the “Web” search type.

Read it precisely, because it is narrower and more useful than either camp admits. Google makes two claims: no special files, and no special optimisations beyond ordinary search guidance. It does not say that how you write makes no difference to whether a passage gets quoted; a page can be perfectly crawlable and still written so nothing can be lifted out of it. And Google speaks only for Google’s surfaces. Even so, it is the largest AI answer surface by query volume by a wide margin, and its documentation is the closest thing to a published specification anyone has. When the biggest player says in writing that there is no secret file, the burden of proof sits with whoever sells you one.

A related detail that catches people out in client meetings: Google-Extended, the robots.txt token governing training and grounding in Gemini apps and Vertex AI, “does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.” It is not an AI Overviews opt-out. For that you are pointed at nosnippet, max-snippet or noindex, all of which degrade your ordinary search appearance too.

The part that carries over, which is most of it

If you already do SEO competently, here is what does not change. Not “changes a bit”. Does not change.

• Crawlability and indexability. Nothing gets retrieved that cannot be fetched and parsed. Render-blocking, soft 404s and orphans are still the first three days of the job.
• Information architecture and internal linking. Retrieval discovers pages through the same graph. A page four clicks deep with one internal link is still weak.
• Page quality and first-hand experience. Google’s helpful-content guidance already asks whether content shows expertise from having actually used the thing. It needed no update for AI.
• Entity clarity. Company, product and people named in full and consistently, accurate Organization data, no leaning on pronouns. Now the difference between being resolved as an entity and being ambiguous.
• Structured data where it is actually supported. Emphasis on actually. Google’s supported-features gallery is the only list that matters, and it moves: the FAQ rich result stopped appearing in Google Search on 7 May 2026 and the documentation was removed that June. Anyone still selling FAQPage markup as an AEO tactic is selling a feature that was switched off.
• Links. Still a signal, still worth earning, still not the whole story, exactly as in 2019.

The four practices that are genuinely new

1. Writing passages that answer a question completely on their own. Retrieval chunks pages and evaluates fragments, so the unit of craft moves down a level. A definition has to name its own subject rather than say “it”. A statistic needs its unit, date and attribution in one sentence. A claim has to survive being read with no heading above it. The test costs nothing: cut any eighty-word window out of a page, paste it into a blank document, and see whether it still answers something and still says who said it. Most B2B copy fails, and fixing it is the highest-return editorial change available right now.

2. Being cited and mentioned, not only linked. A nofollowed mention in a trade title, a quote in a podcast transcript, a recommendation in a forum thread: all of it puts your brand into text that retrieval reads, and none of it appears on a link-building report. The skill is press and expert placement aimed at being the quoted source, which means tracking unlinked mentions as a primary metric.

3. Tracking your brand’s presence inside assistant answers. There is no Search Console for ChatGPT, so you build a panel: a fixed set of buyer-shaped prompts, frozen wording, run on a schedule, recording whether you were mentioned, whether you were linked, who else appeared, and what was said about you. The skill is not running prompts. It is methodological discipline, mostly resisting the urge to improve the prompt list and thereby destroy your own baseline.

4. Treating crawler access as a commercial decision. New as a category of work, not just a technique. Different bots from the same vendor do different jobs: training, search indexing, and fetching a page because a user asked about it right now. Block the wrong one and you vanish from answers your buyers are reading. Block the right one and nothing changes. So you need the tokens, you need to verify bots by IP rather than user agent because user agent strings are trivially spoofed, and you need to put the trade to a finance director in revenue terms. That last part is a negotiation, not a config change.

Notice what they have in common. Two are writing skills, one is a press skill, one is measurement plus commercial judgement. Not one is a new algorithm, a new file format, or a tool you have to buy.

The claims that do not survive scrutiny

llms.txt and its relatives. Be fair to the proposal before dismissing the pitch. Jeremy Howard proposed llms.txt in 2024, thousands of sites publish one, doc platforms generate them automatically, and the AI labs publish llms.txt files for their own developer docs. All true, and none of it is the claim being sold to you: publishing a file is not evidence that any assistant fetches yours and uses it to decide what to say about you. It is not a ratified standard at the IETF, the W3C or anywhere else, and Google’s documentation says you do not need AI text files. It costs almost nothing, so ship one if your developer docs benefit. Do not pay four figures for it or build a reporting line around it.

Keyword stuffing in a prompt costume. Targeting “prompts” as though they were keywords, one page per long-tail phrasing. Google’s spam policies define scaled content abuse as generating many pages “for the primary purpose of manipulating search rankings and not helping users”, with examples including “using generative AI tools or other similar tools to generate many pages without adding value for users” and “creating many pages where the content makes little or no sense to a reader but contains search keywords”. Renaming a tactic does not move it out of the definition, and that is a spam policy, not a ranking suggestion.

Anyone promising placement inside an assistant’s answer. There is no ranked list of ten positions to buy into. Output varies with the user, the session, the phrasing, the model version and whatever was retrieved in the moment. A vendor guaranteeing the top spot in ChatGPT is either showing you a screenshot from their own session or reporting a number they generated, with no independent source to check it against. Ask how the guarantee is verified, and by whom.

What to actually learn, and what it plugs into

What I would tell a mid-level SEO to study, with the reason and the competence it extends. None of it needs a certificate.

An order, if you want one: passage-level editing first, because it pays immediately and is entirely in your control. Then the measurement panel, so a baseline exists before anyone asks you to prove something. Then logs and bot verification. Press work is a longer game.

Should you put GEO on your CV or in your service list?

Yes, and carefully, for slightly cynical reasons.

Freelancer or agency: put the term in your service list, because buyers search for it and you cannot capture demand for a word you refuse to say. Describe it as a component, not a discipline. Something like: “SEO, including optimisation for AI answer surfaces (AI Overviews, AI Mode, ChatGPT, Perplexity): passage-level rewrites, entity and structured-data cleanup, mention building, and monthly assistant-visibility reporting.” That contains the search term, tells a buyer exactly what they get, and implies no lever nobody has. What I would not write is “GEO, the new SEO”, because the first client who reads Search Central will ask about it. Nor “certified GEO specialist”: there is no accrediting body and no recognised examination, so the certificate is a receipt for a course, and hiring managers who know the field read it as a signal in the wrong direction.

In-house: do not change your job title. Change one slide in your quarterly review: the pages you rewrote for extraction and what happened to them, the mention-building work and where it landed, a dated baseline from your prompt panel. A title gets you interviews at other companies; that slide gets you budget at this one. Put the word in your CV’s skills section if applicant tracking filters worry you. That is a keyword decision, not a claim about your identity.

When an interviewer asks whether you do GEO, the answer that lands is not “yes, I’m certified”. It is “yes, and here are the three things I changed, the surfaces I changed them for, and how I measured whether it worked.”

How you will know in two years whether it became its own discipline

Specific signals rather than vibes, in descending order of how much each would change my mind.

• A ratified technical standard that assistants honour. The IETF has an AI Preferences working group chartered to “standardize building blocks that allow for the expression of preferences about how content is collected and processed for Artificial Intelligence (AI) model development, deployment, and use”, with two Internet-Drafts in progress: draft-ietf-aipref-vocab and draft-ietf-aipref-attach. If those become RFCs and major assistants implement them, there is a technical surface with compliance work attached, and that work is not ordinary SEO.
• Controls that do something other than remove you. Every documented mechanism today is subtractive: block the crawler, suppress the snippet, deindex. The day a vendor documents a control that changes how you are represented rather than whether you are, there is a discipline in it.
• Reporting you can be held accountable to. Google announced generative AI performance reports for Search Console on the Search Central blog in June 2026, so open your own property rather than trusting any article about it, this one included. If per-surface impression and citation data matures, measurement stops being a hand-maintained prompt panel and becomes an account you log into.
• The labour market, read carefully. In-house roles where AI answer visibility is the whole job, at companies that do not sell GEO services. Agency ads for GEO specialists tell you about agency positioning, not about a discipline.
• Divergence of practice. Today the right technical answer for an assistant and the right answer for Google Search agree almost everywhere. The day they contradict each other and you have to choose, you have two disciplines whether anyone likes it or not.
• The anti-signal. More certifications, more conference tracks, more acronyms, and none of the five above. That is an industry marketing itself, which is not a field forming.

My own bet, offered as a bet: in two years this looks like technical SEO or international SEO. A named specialism with its own job titles, conference track and tooling, sitting inside SEO rather than replacing it, practised by people who can also fix a canonical tag. Technical SEO acquired every trapping of a separate discipline and never stopped being SEO, and the people who read it as permission to stop learning the rest of the craft did not do well.

So: not a new skill to learn. A few new practices to add, a couple of vendor claims to refuse, and one honest conversation with whoever is quoting you for it.

If you want the work rather than a course, that is what we do. AB7 Solutions runs SEO, AEO, GEO and AIO as one programme rather than a retainer with a GEO surcharge bolted on: passage-level rewrites of the pages that matter, entity and structured-data cleanup against the features Google actually supports rather than deprecated ones, crawler and bot-access decisions with the commercial reasoning written down, mention building aimed at making you the quoted source, and a dated visibility baseline. The reason to trust us here is the reason this article names what does not work: if your existing SEO already covers most of it, we will say so and scope only the gap. Call +1 321 341 7733, or email ab@ab7solutions.com or director@ab7solutions.com. More at www.ab7solutions.com.

Sources: Google Search Central, AI features and your website (quoted directly), Spam policies for Google web search, Google common crawlers (Google-Extended), Structured data markup that Google Search supports (FAQ rich result withdrawn 7 May 2026), and Search generative AI performance reports in Search Console (June 2026); Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, “GEO: Generative Engine Optimization” (arXiv, November 2023; KDD 2024; up to 40% visibility gain on the authors’ own GEO-bench); Jeremy Howard, the llms.txt proposal (a proposal, not a ratified standard); IETF AI Preferences (aipref) working group charter and its Internet-Drafts.