SEO vs GEO vs AEO: What Actually Changes (and What Doesn't)

Note on verification: I could not find a primary, dated record of who first used the term "SEO", so I have left that origin out. The claim that Jason Barnard coined "answer engine optimization" in 2017 comes from Kalicube, which is Barnard's own company, and the independent 2018 article it points to does not actually credit him with the coinage. I report both below. Keyword volumes behind this post are Google Keyword Planner buckets, not exact counts.

---

Where SEO, GEO and AEO actually came from

GEO is the only one of the three with a paper you can open.

"GEO: Generative Engine Optimization" was posted to arXiv on 16 November 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, and was accepted to KDD 2024. It introduced both the term and GEO-bench, a benchmark for testing it.

The headline result, a visibility boost of up to 40%, was measured on the authors' own benchmark, not on ChatGPT or Perplexity as your buyers use them today.

AEO is older and murkier. Kalicube's methodology page states the term was coined by Jason Barnard in 2017 and offers, as its canonical third-party record, a Search Engine Watch article by Rebecca Sentance from 7 February 2018.

I opened that article. It quotes Barnard on voice search turning search engines into "answer engines", then says the strategy "has come to be known as AEO", without crediting anyone with the coinage. So the independent record confirms the term was circulating by early 2018. It does not confirm the attribution.

SEO has no clean origin story I could verify, so I am not going to invent one.

That is the honest state of the taxonomy: one academic paper, one contested attribution, and one term so old nobody agrees where it started. I wrote more about why the labels blur into each other in what "AI search optimization" actually means.

SEO vs GEO vs AEO in one table

	SEO	AEO	GEO
First solid record	Predates the other two, no primary source I could verify	Term in use by Feb 2018 (Search Engine Watch); 2017 coinage claimed by Kalicube	arXiv paper, 16 Nov 2023 (Aggarwal et al.)
Optimises for	A ranked list of links	One direct answer: featured snippet, voice result, knowledge panel	A generated answer that cites sources
Where the buyer sees you	Your page, after a click	Google's answer box or an assistant reading aloud	Inside ChatGPT, Perplexity, Gemini, Claude, AI Overviews
Unit of success	Position and clicks	Being the one answer	Being named or cited, some percentage of the time
Stable enough to rank?	Yes	Mostly	No
Access controlled by	Googlebot, Bingbot, robots.txt	The same crawlers	A separate bot per engine: OAI-SearchBot, PerplexityBot, and others
What the work looks like	Content, technical, links	Content, technical, links, structured answers	Content, technical, links, structured answers

Look at the last row. That is the point of this whole post.

What doesn't change: the tactics

Google's own documentation is blunt about it. Its AI features page, last updated 10 December 2025, says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

Google has an obvious interest in that answer, so I do not take it as the final word. But outside data points the same way.

RESONEO's July 2026 study of 1,249 ChatGPT answers found that one retrieved URL in three already sits on Google's first page, against one in twenty for Bing's. Classic ranking is still feeding a large share of what the model reads.

Wikipedia's entry on generative engine optimization notes that as of early 2026 there was still no consensus definition separating GEO, AEO and LLMO, and quotes researcher Nikhil Lai describing them as "significantly, but not fundamentally, different from SEO".

So: write clearly, structure your pages so a machine can lift a clean answer, earn mentions on sources the models trust, keep the site crawlable. None of that is new. Anyone selling you a separate tactical playbook is selling you SEO with a fresh invoice.

What does change: measurement

This is the part that genuinely breaks.

SparkToro published research on 28 January 2026, run by Rand Fishkin across 600 volunteers and 2,961 runs against ChatGPT, Claude and Google's AI, and found less than a 1 in 100 chance that two answers list the same brands. Ask the same question twice and you get two different worlds.

There is no position 1 to hold. There is only the share of answers you appear in, across enough repeated runs to mean something.

That changes the unit of reporting from rank to frequency, and it changes what a report has to prove. A single screenshot of ChatGPT naming you is not evidence. Eighty runs of a buyer question, scored, is. That is why our service is built around repeated scans rather than a rank tracker.

It also changes which pages you build. In our own study on 28 August 2026, 40 buying-intent prompts run on ChatGPT and Google, across 490 cited URLs, comparison pages led "X vs Y" questions at 56.7%, pricing pages led cost questions at 53.4%, and product pages took 37.7% of ChatGPT's citations. That is our data, self-reported, on our query set, and you should read it as a direction rather than a constant.

What does change: prompt research replaces keyword research

Keyword research assumes a short query typed into a box. Prompt research assumes a sentence, usually with context attached: budget, industry, team size, a competitor already in mind.

"crm software" and "which crm should a 12 person agency use if we already run hubspot for marketing" are not answered by the same content.

The work is to collect the questions your buyers actually ask, from sales calls rather than from a keyword tool, then run them repeatedly and watch who gets named.

What does change: you manage access engine by engine

In SEO, robots.txt was one conversation with two crawlers. Now every engine has its own, and the rules differ.

OpenAI's bot documentation says OAI-SearchBot exists "to surface websites in search results in ChatGPT's search features", while GPTBot is the crawler whose content may be used to train models. Block the wrong one and you either lose visibility or lose nothing, depending on which you picked.

Perplexity's documentation says PerplexityBot is "designed to surface and link websites in search results on Perplexity", and that it is not used to crawl content for foundation models.

Then there is the plumbing. Olivier de Segonzac of RESONEO reported in Search Engine Land on 17 August 2026 that ChatGPT's robot does not execute JavaScript, that pages over 4 MB are rejected outright with an HTTP 400, and that in instant mode the model sees roughly 200 characters anchored on your H1.

A client-side rendered page with a vague H1 is invisible to that pipeline. No amount of content strategy fixes it.

GEO vs AEO: a distinction that stopped holding

When people ask me about geo vs aeo, they usually want to know whether to buy two things.

You should not. AEO was built for a world of one extracted answer, and GEO for a world of synthesised answers with citations. The aeo vs geo split made sense in 2018, when the target was a featured snippet. Today the same structured, quotable page serves both, and the engines have converged on reading pages the same way.

Treat them as one practice with three labels, and spend the argument time on measurement instead.

What I would do with this

Keep your SEO team and your SEO budget. Add three things: a set of real buyer prompts, a repeated measurement habit, and a crawler access check across the engines that matter to you.

If you want that done as a one-off diagnosis first, that is what our AI visibility audit covers.

If you had to defend your AI visibility number to your CFO next quarter, what would you show them, a screenshot or a sample size?

Sources

• GEO: Generative Engine Optimization, arXiv:2311.09735, first posted Nov 16, 2023, accepted to KDD 2024 — Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande — https://arxiv.org/abs/2311.09735
• The rise of Answer Engine Optimization: Why voice search matters, Feb 7, 2018 — Search Engine Watch (Rebecca Sentance)
• Answer Engine Optimization (AEO) methodology page, accessed Sep 17, 2026 — Kalicube
• AI features and your website, last updated Dec 10, 2025 — Google Search Central — https://developers.google.com/search/docs/appearance/ai-features
• What ChatGPT pulls, what it shows, what it cites, July 2026, 1,249 ChatGPT answers (vendor research) — RESONEO — https://think.resoneo.com/chatgpt-retrieval/
• Inside ChatGPT's retrieval stack: The index, cache, and pages it actually reads, Aug 17, 2026 — Search Engine Land (Olivier de Segonzac) — https://searchengineland.com/chatgpt-retrieval-stack-index-cache-pages-485036
• New research: AIs are highly inconsistent when recommending brands or products, Jan 28, 2026, 600 volunteers and 2,961 runs (vendor research) — SparkToro (Rand Fishkin) — https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
• Bots documentation (OAI-SearchBot, GPTBot, ChatGPT-User), accessed Sep 17, 2026 — OpenAI
• Bots documentation (PerplexityBot, Perplexity-User), accessed Sep 17, 2026 — Perplexity — https://docs.perplexity.ai/guides/bots
• Generative engine optimization, accessed Sep 17, 2026 — Wikipedia
• Which Content Formats Win the Most AI Citations for B2B SaaS?, 40 prompts and 490 cited URLs, Aug 28, 2026 (our own data, self-reported) — GeoCited — /blog/which-content-formats-win-ai-citations-b2b-saas