Traditional SEO optimizes for a ranked list of blue links. AEO (answer engine optimization) optimizes for your content being extracted and read aloud as a direct answer. GEO (generative engine optimization) optimizes for your content being synthesized and cited inside a generated response with a source link attached. They overlap heavily, but they are not the same target, and conflating them is why most 'GEO strategy' content right now is either recycled SEO advice with a new acronym stapled on, or actual technical guidance buried under buzzword marketing.
I run both playbooks on my own site — schema.org Person and FAQPage markup, an llms.txt file, and an AI-crawler-allowed robots.txt — because I build retrieval and agent systems for a living (LangGraph, RAG pipelines, FastAPI backends) and wanted to understand the mechanics from the implementation side, not from a listicle. This post is what I actually found once I separated 'what Google says' from 'what the other engines do.'
The three things are not interchangeable
Traditional SEO targets Google's ten blue links and ranking factors: backlinks, crawlability, keyword relevance, page experience. AEO targets being the single extracted answer in a featured snippet or voice response — structure and directness matter more than backlink count. GEO targets being cited as a source inside an LLM-generated synthesis — which sources an LLM chooses to quote depends on different signals than what makes Google rank a page. A page can win at SEO (rank #1) and lose at GEO (never get cited by ChatGPT) at the same time, because the retrieval and selection mechanisms are different systems entirely.
The reason this matters commercially: AI Overviews now appear on a large share of Google searches, and they measurably suppress click-through to the sites they summarize — a user gets the answer without visiting anyone. Meanwhile ChatGPT accounts for the overwhelming majority of AI-driven referral traffic across the other assistants combined. So the two halves of this problem — surviving inside Google's AI Overviews, and getting picked up by ChatGPT/Perplexity/Claude — are worth solving separately, because the traffic math behind each is different.
What Google's own guidance actually says
Google's stance, straight from their AI features optimization documentation, is blunt: there is no special markup for AI Overviews or AI Mode. The ranking systems are the same core systems used for classic Search. If your content ranks well under normal Google SEO — helpful, original, demonstrates real experience, satisfies E-E-A-T — it is already eligible to be surfaced and summarized in an AI Overview. Google explicitly does not want you writing content 'for AI' as a separate exercise from writing content for people.
Two things Google specifically warns against: don't fragment your content into disconnected chunks or bullet fragments purely to make it 'easier for AI to extract' — it hurts readability for humans and doesn't buy you anything with their systems. And don't treat E-E-A-T as a checkbox; it's the same trust signal set (author credentials, first-hand experience, citations, accuracy) that has driven classic rankings for years. If your GEO plan for Google is 'add some FAQ schema and hope,' you've misread the actual guidance — for Google specifically, the lever is still: write genuinely useful, well-sourced content for a human reader.
What actually changes for ChatGPT, Perplexity, and Claude
This is where GEO stops being a rebrand of SEO and becomes a genuinely different discipline, because these engines don't share Google's index or ranking pipeline. They retrieve from their own crawled corpus (or live web search) and select passages to quote based on extractability and perceived authority — not PageRank.
- Structure for extraction: answer-first paragraphs, direct definitional statements up top, comparison tables, and numbered lists get lifted into responses far more easily than narrative prose that buries the answer three paragraphs in.
- llms.txt: a machine-readable index of your site's key pages in plain Markdown, sitting at /llms.txt. Not every engine consumes it yet, but it costs nothing to publish and is the emerging convention for telling an LLM crawler what on your site is worth reading.
- robots.txt AI-bot allowlisting: GPTBot, PerplexityBot, ClaudeBot, and similar crawlers are distinct user-agents. If your robots.txt blocks them (many default configs do, or block everything except Googlebot), you are invisible to those engines regardless of content quality. Explicitly allowing them is a prerequisite, not an optimization.
- FAQPage and Person/Organization schema: structured data doesn't directly control Google's AI Overviews per their own guidance, but it does give other engines a clean, unambiguous data source to pull entity facts and Q&A pairs from when synthesizing an answer.
- Third-party citation weight: these engines lean hard on being able to attribute a claim to a source. Content that reads as a source of record — data, named claims, a real author — gets quoted; vague marketing copy gets paraphrased or ignored.
On my own site this looks like: a Person schema block with verified sameAs links to my professional profiles, FAQPage schema wrapping the actual FAQ section, an llms.txt at the root listing my key pages in plain Markdown, and a robots.txt that explicitly allows GPTBot, PerplexityBot, ClaudeBot, and Google-Extended rather than silently blocking them. I'm not claiming a traffic or ranking lift from this — I haven't measured one, and anyone citing hard numbers from a single-site implementation this recent is guessing. What I can say is that the implementation is mechanically correct: the bots can reach the content, and the content is in a shape they can parse and quote.
What the Princeton GEO study actually measured
The most-cited empirical data point in this space is Princeton's GEO research, which tested what content-level changes actually move the needle on being cited by generative engines. The results cut against a lot of the 'AI-bait content' advice floating around:
- Citing sources: +40% visibility in AI-generated responses
- Adding statistics: +37%
- Including direct quotations: +30%
- Writing in an authoritative tone: +25%
- Keyword stuffing: −10% (it actively hurts AI citation)
The pattern across all five findings is consistent: the levers that help are the same things that make content genuinely more credible and information-dense — cite your sources, back claims with numbers, quote real people, write like you know what you're talking about. The one thing that hurts is the one classic black-hat SEO tactic. That's a strong signal that 'optimizing for AI' and 'writing good, well-sourced content' are converging, not diverging.
Don't build a parallel 'AI-bait' content track
Google explicitly classifies mass-produced, low-value content aimed at gaming any ranking or citation system — including AI systems — as scaled content abuse, and it's a manual-action-eligible violation. The same page needs to serve a human reader and an LLM parser. If your GEO plan involves publishing a separate thin layer of 'AI-optimized' content bolted onto your real site, you're building the exact pattern Google's spam policies target.
Frequently asked
Do I need llms.txt to rank in Google?
No. llms.txt has nothing to do with Google's ranking or AI Overviews — Google's own guidance doesn't reference it. It's a convention aimed at LLM crawlers like GPTBot and ClaudeBot. Publish it for the other engines; don't expect it to move a single Google metric.
Does GEO replace SEO?
No. They're additive. Traditional SEO fundamentals — crawlability, page experience, backlinks, E-E-A-T — still determine whether Google indexes and ranks you at all, including whether you're eligible to appear in an AI Overview. GEO is a separate layer on top, aimed at engines that don't share Google's pipeline.
Will AI Overviews kill my traffic?
For queries where the AI Overview fully answers the question, yes, expect fewer clicks — that's the documented trend, and commercial-intent queries triggering AI Overviews have grown sharply year over year. The mitigation isn't a markup trick; it's ranking for queries specific and deep enough that a two-sentence summary can't fully substitute for your page.
Which matters more right now, AEO or GEO?
Depends on where your traffic actually comes from. If it's overwhelmingly Google, the AEO/AI Overview surface matters more, and the fix is just better core SEO. If a meaningful share of your referral traffic already comes from ChatGPT or Perplexity, the crawler-access and structure work described above is the higher-leverage lever, since almost none of it happens by default.
If you're implementing this for a team
I do this kind of technical GEO/AEO implementation — schema, llms.txt, crawler allowlisting, content restructuring — as part of the AI engineering work I take on. Happy to talk through it if you're setting it up for your own site.