Why this vocabulary exists
For twenty years "search" meant one thing: a ranked list of links you competed to climb. AI answer engines changed the shape of the result. Now a buyer asks a full question — "best payroll software for a 12-person agency" — and the engine reads many sources, writes one answer, and names a short list of brands. You are either in that answer or you are invisible. There is no page two.
That single change rewires the whole vocabulary. The thing you optimize for is no longer a keyword ranking; it's a citation inside a generated answer. The thing you measure is no longer SERP position; it's how often — and how favorably — engines mention you versus your competitors. New problem, new words.
1B+
people reached by Google's AI Overviews (Google, March 2025); AI Mode has since passed 1 billion monthly users — brands are summarized whether they opt in or not
The honest caveat: nobody agrees on the terms yet
AI SEO, GEO, AEO, LLM SEO and AIO are used interchangeably across the industry, and the boundaries between them are genuinely fuzzy. As of early 2026, no consensus academic definition distinguishing these terms has been established. This glossary gives you the most widely-used meaning for each — and flags where the disagreement is real, not invented.
The disciplines — AI SEO, GEO, AEO, LLM SEO
Start here, because this is where the confusion lives. These are overlapping names for the same broad goal — getting cited by AI — viewed from slightly different angles. Most teams doing this work are doing all of them at once. Treat them as one playbook with several labels, not four rival disciplines.
| Term | What it means | Best read as |
|---|---|---|
| AI SEO | The umbrella: optimizing so AI engines find, understand and cite your brand. | The catch-all the rest live under. |
| GEO | Generative Engine Optimization — getting cited and recommended inside generated answers (esp. ChatGPT, Claude, Perplexity). | The most widely-used, research-backed term. |
| AEO | Answer Engine Optimization — targeting AI answer surfaces broadly (e.g. Google AI Overviews, featured snippets). | GEO with a slightly wider, answer-feature focus. |
| LLM SEO | Optimizing specifically for large-language-model assistants. | A plainer synonym for GEO. |
Generative engine optimization (GEO) is the term with an actual paper trail: it comes from a 2023 research paper, "GEO: Generative Engine Optimization," presented at KDD '24. That study is also why the field has any hard evidence at all — more on it below.
Answer engine optimization (AEO) is usually framed a hair wider — optimizing for any AI-powered answer surface, including Google's AI Overviews — while GEO leans toward the chat-style assistants. The distinction is contested and, frankly, not worth losing sleep over. If you want the plainest synonym of all, that's LLM SEO.
The one distinction worth keeping
Don't confuse the work with the scoreboard. GEO, AEO and LLM SEO are the *work* you do. AI visibility and share of voice are the *outcome* you measure. Optimizing without measuring is guessing.
The surfaces — what an "answer engine" actually is
An answer engine is any system that reads multiple sources and synthesizes a single direct answer, instead of returning a list of links to sift through. These are the surfaces your AI SEO work is aimed at:
- **Google AI Overviews** — the AI summary above Google's normal results. Reached 1B+ people by March 2025.
- **Google AI Mode** — a fully conversational search mode powered by Gemini; has surpassed 1 billion monthly users.
- **ChatGPT (and ChatGPT Search)** — OpenAI's assistant, now one of the largest discovery surfaces on the web.
- **Perplexity** — an answer-first engine that leads with citations by design.
- **Claude** — Anthropic's assistant, increasingly used for research and product recommendations.
How AI Overviews actually builds an answer
A common myth is that AI Overviews just rewrites the #1 result. It doesn't. Google uses a "query fan-out" technique — it issues multiple related searches across subtopics and sources at once, then synthesizes one cited answer from across them. That's why a page that never ranks #1 can still get cited: it just has to be the best source for one slice of the fan-out.
How AI reads your site — crawlers, llms.txt, structure
Before an engine can cite you, it has to be able to read you. This is the most technical corner of the glossary, and the one where a single misconfigured file can make you invisible.
AI crawlers
AI crawlers are the bots that fetch your pages on behalf of AI engines. The catch that trips people up: they're not interchangeable, and blocking one rarely does what you think.
| User-agent | Who | What it does |
|---|---|---|
| GPTBot | OpenAI | Crawls to train foundation models. |
| OAI-SearchBot | OpenAI | Surfaces your site inside ChatGPT search. |
| ChatGPT-User | OpenAI | Fetches a page when a user's prompt triggers it. |
| ClaudeBot | Anthropic | Anthropic's primary crawler. |
Blocking GPTBot does NOT remove you from ChatGPT search
GPTBot is the *training* crawler; OAI-SearchBot is what surfaces you inside ChatGPT's answers. They're separate toggles. Block GPTBot if you don't want your content training models — but if you also block OAI-SearchBot, you've removed yourself from ChatGPT's answers entirely. And note: OpenAI says ChatGPT-User is user-triggered, so robots.txt may not apply to it the way it does to a normal crawler.
llms.txt
llms.txt is a proposed standard — a single Markdown file at your domain root (`/llms.txt`) that hands LLMs a curated, clean map of your most important content. It was published by Jeremy Howard in September 2024. The format is simple: an H1 with your site name, a blockquote summary, optional prose, then H2 sections of `name` links, plus an "Optional" section that can be dropped in shorter contexts.
llms.txt is a proposal, not robots.txt
Treat llms.txt as an emerging convention with growing — but partial and voluntary — adoption. It is not an official, universally-honored web standard, and not every AI company reads it. It's low-effort to publish and a tidy way to point models at your best pages, but don't mistake it for a guarantee or a ranking factor.
Extractable content & structured data
Even with the door open, the model has to be able to *lift* you. Answer-first writing, clear definitions, schema.org markup and citable statistics make a passage easy to quote verbatim into an answer. Content that buries the answer ten paragraphs down rarely gets pulled.
How you measure it — visibility, share of voice, citations
This is the bucket that replaces rankings and clicks. Five terms cover almost everything you'll need.
- **AI visibility** — whether, how often, and how prominently AI engines mention, cite and recommend you. The headline scoreboard. (Closely related: **AI search visibility**, the same idea framed around search-style answer surfaces.)
- **Share of voice** — your slice of all brand mentions versus competitors across a fixed prompt set. The number that tells you whether you're winning the category.
- **Brand mentions** — how often you're named at all, with or without a link. The raw count under visibility.
- **Citation tracking** — monitoring whether answers actually link to your content, and whether they credit the right page.
- **Sentiment** — whether a mention reads positive, neutral or critical. "The cheap one" and "the best for teams" are not the same win.
Two flavors of share of voice
Entity-based SOV measures how often you're *named* as a recommendation. Citation-based SOV measures how often your *content* is cited as a source. They answer different questions — "is the model recommending me?" vs "is the model reading me?" — and a healthy AI SEO program watches both.
One catch makes measurement non-optional rather than one-and-done: AI answers drift. One study tracking 2,500 prompts across Google AI Mode and ChatGPT found that 40–60% of cited sources changed month to month. A single reading is noise — the signal is the trend line, which is why visibility has to be tracked on a schedule.
40–60%
of cited sources in AI answers change month to month (tracking 2,500 prompts across Google AI Mode + ChatGPT) — why you track the trend, not a snapshot
There's also a higher-confidence signal most tools skip: your own server logs. When an AI crawler reads you or an AI answer sends a referral click, that's ground truth, not a synthetic sample. SourceWatch measures both sides — the mentions and share of voice across ChatGPT, Perplexity, Gemini and Claude, *and* the real AI-crawler and AI-referral traffic actually hitting your site.
Want to see where your brand stands across the AI engines right now? Run a free AI SEO audit — it checks whether AI engines can read, recognize and cite your site in about 15 seconds.
Run a free AI SEO auditWhat actually works (and what doesn't)
The field has more confident folklore than evidence, so anchor on the one peer-reviewed source that measured it. The 2023 GEO research paper built a benchmark of 10,000 queries (GEO-bench) and tested which content changes actually moved a source's visibility in generated answers.
- **Add citations, quotations and statistics.** These were the top-performing methods — lifting visibility in generative engines by up to ~40%. Make your content the thing a model can quote with confidence.
- **Don't keyword-stuff.** The classic SEO reflex showed "little to no" improvement in generative responses. Optimizing for AI is not optimizing for a 2015 Google ranking.
- **Be a recognized entity.** Consistent schema, profiles and off-site mentions help engines know who you are before they'll recommend you.
- **Open the door, then point the way.** Allow the AI crawlers you want, and consider an llms.txt file as a curated map to your best pages.
- **Measure, change one thing, re-measure.** Because answers drift, AI SEO is a loop, not a launch.
Five myths this glossary exists to kill
1) "GEO, AEO and SEO are separate, settled disciplines" — they overlap and have no agreed boundaries. 2) "Optimize for AI like you optimize for Google rankings" — keyword stuffing barely moves AI answers; citations and stats do. 3) "llms.txt is an adopted standard like robots.txt" — it's a 2024 proposal with partial adoption. 4) "Blocking GPTBot removes me from ChatGPT" — no, that's OAI-SearchBot. 5) "AI Overviews just rewrites the #1 result" — it fans out across many sub-queries and synthesizes.