There is no magic file and no magic schema. Google says so directly. GEO optimization is the discipline of making a company legible, verifiable and consistently described so retrieval systems can find, trust and reuse it.
That framing matters, because a large share of the GEO advice currently circulating sells tags and files as shortcuts. Start from what the platforms actually document.
Start with what Google states
From Google's own AI features documentation:
There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.
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.
Eligibility requires only that a page is indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. The controls that apply to snippets apply identically to AI features: nosnippet, data-nosnippet, max-snippet, noindex.
Google's recommended practices are unglamorous and worth following: allow crawling in robots.txt, make content findable through internal links, maintain good page experience, keep important content in text form, support text with quality images and video, ensure structured data matches visible text, and keep Merchant Center and Business Profile current.
If that reads like competent technical SEO, that is the point. GEO does not replace SEO. It extends it toward a second surface.
Crawler access: the most expensive mistake
Before any content work, check that retrieval crawlers can reach you. Many sites blanket-block AI agents, then wonder why they are absent from AI answers.
| Vendor | Agent | Purpose |
|---|---|---|
| OpenAI | GPTBot |
Model training |
| OpenAI | OAI-SearchBot |
Surfaces sites in ChatGPT search |
| OpenAI | ChatGPT-User |
User-triggered fetch |
| Anthropic | ClaudeBot |
Training data |
| Anthropic | Claude-SearchBot |
Indexes content to improve search quality |
| Perplexity | PerplexityBot |
Surfaces and links sites in Perplexity, not training |
Googlebot |
Search index, gates AI Overviews and AI Mode eligibility | |
Google-Extended |
Opt-out for AI training and grounding in some Google systems, does not affect Search |
Separate the training decision from the retrieval decision. Blocking GPTBot is a defensible content-licensing position. Blocking OAI-SearchBot is self-removal from ChatGPT search. These are not the same choice and should never be made with one line.
What the evidence supports
The term GEO comes from Aggarwal et al., KDD 2024 (arXiv:2311.09735), which introduced GEO-bench and tested which content changes shifted visibility in generative responses. The abstract reports GEO "can boost visibility by up to 40% in generative engine responses", and that "the efficacy of these strategies varies across domains".
Reported as directional ranking rather than precise figures, the tested tactics ordered by effect:
- Adding quotations from relevant or expert sources — the largest single gain
- Adding statistics and data points — large
- Fluency optimization, rewriting for readability — large
- Citing sources — meaningful
- Technical terminology — moderate
- Easy-to-understand language — modest
- Authoritative tone — small
- Unique words — negligible
- Keyword stuffing — negative effect
Two honest caveats. The study predates AI Mode and the current model generation. And domain variance was explicit in the findings, so tactics do not port between industries unexamined.
The practical read: synthesis systems favour material that is quotable, sourced, specific and readable. That is a writing standard more than a technical one.
How retrieval actually differs by platform
Treating the assistants as interchangeable is the second most common error.
Google AI Mode issues "multiple related searches concurrently across subtopics and multiple data sources and then brings those results together", and "makes a plan, conduct[s] searches to find information and adjust[s] the plan based on what it finds". It draws on the Knowledge Graph and shopping data alongside web content. This is query fan-out, and it means you are competing for subtopics you never targeted directly.
ChatGPT search "rewrites your query into one or more targeted queries" sent to search partners; OpenAI's help documentation names Bing and Shopify among them without stating exclusivity, and states any website or publisher can choose to appear.
Gemini grounding decides to search, runs generated queries, synthesises, and returns inline citation annotations with character offsets.
Claude web search decides autonomously whether to search, may search several times in one turn, and always returns citations with source URL, title and cited text.
Perplexity retrieves in real time and every answer carries clickable citations. Its backend ranking is undocumented. Its own documentation notes Perplexity-User generally ignores robots.txt because the fetch is user-initiated.
Copilot is generally understood to inherit Bing's index. Microsoft has not published a current architecture statement we could verify, so we describe it that way rather than asserting it.
What does not work
llms.txt. No major engine documents consuming it. It is a September 2024 proposal by Jeremy Howard of Answer.AI, published at llmstxt.org, adopted by some documentation tooling. Ship it if you like. Do not expect visibility from it.
FAQPage schema as a visibility tactic. Per Google's documentation, FAQ rich results stopped appearing on 7 May 2026, with the search appearance, rich result report and Rich Results Test support dropped in June 2026, and Search Console API support removed in August 2026. FAQPage remains legitimate machine-readable semantics for other consumers. Anyone selling it as a Google visibility tactic is working from a 2023 playbook. HowTo rich results were likewise removed in September 2023.
Submitting your site to ChatGPT. There is no such submission mechanism. Anyone offering it is selling nothing.
Publishing AI content at volume without review. Covered in the AI adoption roadmap: two documented sites publishing roughly 1,000 generated posts each were algorithmically deindexed with no manual action. Small sample, clear enough signal.
Keyword stuffing. Measurably negative in the GEO paper, and the one tactic the evidence actively rules out.
A note on the vocabulary
GEO, AEO, LLMO, AI SEO, GAIO and AI Search Optimization are all in circulation with no consensus. Profound, a vendor, argues publicly that GEO is a poor term because it collides with geo-targeting, and pushes AEO. Others use both interchangeably. We use GEO because it is the term from the original academic paper, and the label matters less than the discipline. If a vendor's pitch depends on their preferred acronym, that is a signal.
Where this fits
Read the methodology for the operational sequence, GEO ROI for measuring it, the AI adoption roadmap for phasing, the glossary for definitions, and the FAQ for common questions. Our AI Visibility service and Search Dominance service are the two halves of this work.