AI search optimization is the work of making a brand findable, quotable, and trusted by systems that answer questions instead of listing links. It travels under several names: Answer Engine Optimization (AEO), Generative Engine Optimization (GEO), and AI Optimization (AIO). The labels differ, the problem does not. When an assistant answers your customer’s question directly, the only visibility that counts is whether it cites you.
This page covers what actually changed, what earns a citation, and how I measure it. It is the same work I apply for iGaming operators and affiliates and for B2B SaaS companies, where a single AI answer can absorb the entire research phase of a buying decision.
What Actually Changed
Classic SEO competes for a position in a list. AI search competes for inclusion in an answer. Those are not the same contest. A model retrieves a set of candidate sources, then synthesizes a response from a subset of them. Ranking first does not guarantee you are in that subset, and pages that rank modestly are cited every day.
That splits visibility into two separate problems. The first is retrieval: can the system find and parse your page at all. The second is citation: once retrieved, does it trust your page enough to quote it. Most sites that are invisible in AI answers are failing the second test, not the first, which is why publishing more content rarely fixes it.
What Earns a Citation
| Signal | Why it matters | In practice |
|---|---|---|
| Entity clarity | A model has to know who you are and what you are credible on | Consistent brand entity, Knowledge Panel completeness, corroboration off site |
| Structured data | Machine readable meaning. Google says no AI specific schema is required, but clear structured meaning still aids extraction | Schema for reviews, FAQs, organizations, and products |
| Extractable content | Models quote what they can lift cleanly | Direct answers near the top, self contained passages, clear headings |
| Heading and query alignment | Retrieval matches your headings against the question asked | Headings phrased as the question a user actually types |
| Off site corroboration | Models weight consensus across independent sources | Mentions and consistent facts on third party sites |
What the Research Actually Shows
Most AEO advice is assertion. There is now actual research. The GEO study (Aggarwal et al., presented at ACM SIGKDD 2024) built a benchmark of 10,000 queries across 25 domains and tested nine content changes to measure which ones genuinely increase visibility inside AI generated answers.
| Method | Measured effect |
|---|---|
| Quotation addition | The strongest single lever. Adding relevant quotations produced roughly a 41% relative improvement in visibility. |
| Statistics addition | Adding concrete figures produced gains in the 30 to 40% range. |
| Citing sources | Gains in the 30 to 40% range, and it compounds when combined with the above. |
| Keyword stuffing | Failed outright. It performed worse than doing nothing, around 10% below baseline on Perplexity. |
Two conclusions follow. The levers that work are the ones that make a passage more quotable and more verifiable: quotations, numbers, and citations. That is not a trick. It is what an answer engine needs before it will risk repeating you. And the habits carried over from classic SEO, keyword density above all, do not merely fail here, they actively hurt. The study also found that effectiveness varies by domain, which is precisely why a generic AEO checklist underperforms a strategy built around your vertical.
Brand Mentions Are Doing the Heavy Lifting
The lever moving AI visibility fastest right now is not a link. It is a mention. Language models build their picture of a brand from how the wider web describes it, and that description is assembled from ordinary text, not from anchor tags. A brand named repeatedly and consistently across reviews, roundups, comparison posts, forums, and industry coverage becomes part of the model’s answer to the category question, whether or not a single one of those mentions carries a link.
That inverts a familiar habit. Classic link building treats the mention as the delivery mechanism and the link as the prize. In AI search the mention is the asset. Co-occurrence does the work: if your brand keeps appearing next to the phrases your buyers actually search, the model learns the association and reaches for you when the question comes up. Repetition across independent sources reads as consensus, and consensus is what a model needs before it will name you in an answer.
In practice this reframes digital PR. The targets shift from domains that pass authority to the places a model is likely to have read, and success stops being a link acquired and becomes a brand described accurately in the sources that shape the category. It also explains a pattern that keeps showing up in audits: sites with modest link profiles that get cited constantly, because the web talks about them, and sites with strong link profiles that are invisible in AI answers, because nobody does.
The Technical Floor Nobody Checks
Before any of the above matters, a model has to be able to reach the page and be allowed to quote it. Two requirements get missed constantly, and both are cheap to fix.
- Crawler access – OpenAI runs separate crawlers for separate jobs. OAI-SearchBot is the one that surfaces and cites sites inside ChatGPT’s search features. GPTBot is for training the foundation models. Blocking GPTBot is a defensible business decision. Blocking OAI-SearchBot means opting out of being cited in ChatGPT at all, and plenty of sites have done that by accident while trying to keep their content out of training data.
- Snippet eligibility – Google states that a page must be indexed and eligible to be shown with a snippet in order to appear in AI Overviews and AI Mode. That means
nosnippet,max-snippet:0, anddata-nosnippetdo not simply trim your preview. They remove you from AI answers entirely.
These are the cheapest wins available in AI search, and they are invisible to any content audit that never looks at robots.txt or the snippet directives.
What Google Says, and What That Rules Out
Google’s own documentation is unusually blunt: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. It goes further, stating that you do not need to create machine readable files, AI text files, or markup, and that there is “no special schema.org structured data that you need to add”.
That rules out a good deal of what is currently being sold. An llms.txt file is not used by Google. There is no AI specific schema type that unlocks citation. Anyone charging you for either is charging for a file nobody reads.
Google’s guidance and the GEO research get read as contradictory, and they are not. Google is saying there is no separate switch to flip. The research is saying that how you construct a passage changes whether a model can lift it. Both are true at once. AI search is not a new discipline bolted onto SEO. It is doing the fundamentals well enough that a machine can extract, verify, and attribute what you wrote.
How I Measure AI Visibility
Most AEO advice is untested opinion, published by people who have never checked whether a single page of theirs gets cited. The work I do is measured.
- Citation audits – A defined query set is run across ChatGPT, Perplexity, and Google AI Overviews, logging for each page whether it was retrieved, cited, or ignored.
- Retrieval versus citation – A page can be retrieved constantly and never quoted. Separating the two tells you whether you have a discovery problem or a trust problem, and the fixes are completely different.
- Heading and query alignment – How closely your headings match the questions being asked is measurable, and below a certain threshold citation rates drop off sharply.
- Schema coverage – Structured data lifts citation rate independently of everything else, so it is measured on its own rather than bundled into a general technical score.
- All or nothing patterns – Pages tend to be cited almost always or almost never, rather than sometimes. Knowing which bucket a page sits in is far more useful than an average.
The output is a list of the pages that never get cited, with the reason for each. That is actionable in a way a generic AEO checklist is not.
What This Means for iGaming
Gambling queries are exactly the type AI answers absorb: which casino to use, whether a bonus is worth taking, whether an operator is licensed in a given market. Because these sit squarely in YMYL territory, AI systems are conservative about who they are willing to quote. Trust and entity signals matter more here, not less, which means the sites with genuine authority gain ground while thin affiliate content quietly disappears from the answer entirely. See iGaming SEO.
What This Means for B2B SaaS
SaaS buyers now run their research through assistants. Comparison, alternatives, and “best tool for X” queries get summarized into a shortlist before a human ever reaches your site. Being the source that shortlist is built from is the new category entry point, and it is decided long before anyone clicks. See B2B SaaS SEO.
What This Is Not
It is not a replacement for SEO, and anyone selling AEO as a separate discipline with its own secret tricks is selling you a rebrand. The foundations are the same ones that have always earned rankings: a clear entity, structured meaning, extractable content, and real topical authority. What is genuinely new is the surfaces and the measurement.
Frequently Asked Questions
Is AEO different from SEO?
It is an extension of it, not a replacement. The same entity, structure, and authority signals do the work. What changes is that you are optimizing for inclusion in an answer rather than a position in a list, and that you can no longer assume a good ranking means visibility.
AEO, GEO, or AIO: which term is right?
They describe the same work. Answer Engine Optimization, Generative Engine Optimization, and AI Optimization are competing labels for making content retrievable and quotable by AI systems. The terminology will settle eventually. The underlying problem will not change.
Can you actually measure whether AI cites my site?
Yes. A fixed query set is run across the major AI surfaces and each result is logged as retrieved, cited, or absent. Repeating that over time shows whether changes moved the needle, which turns AEO from a matter of opinion into something you can report on.
How long does it take?
Structured data and content extractability can shift citation behaviour within weeks. Entity and authority signals move on the same slower timeline as traditional SEO, because they depend on how the wider web describes you, not on what you publish.
Do I need an llms.txt file or special AI schema?
No. Google states plainly that you do not need to create machine readable files, AI text files, or markup, and that no special schema.org structured data is required to appear in its AI features. If someone is selling you an llms.txt implementation as an AI visibility service, they are selling you a file that Google does not read.
Sources
- Aggarwal et al., GEO: Generative Engine Optimization, ACM SIGKDD 2024.
- Google Search Central, AI features and your website.
- OpenAI, Crawlers and user agents.
Find Out If AI Is Citing You
Most companies have never checked. The SEO Growth Audit includes an AI visibility pass, so you learn which of your pages get cited, which get ignored, and why.
