For a small site, AEO and GEO mostly change how clearly each page answers its main question and how accurately its structured data describes that page. AEO (Answer Engine Optimization) means organizing a site so answer systems can identify and present direct answers. GEO (Generative Engine Optimization) means improving the likelihood that generative systems retrieve, summarize and cite a site.
Start with the page's main question. Answer it in plain text near the top, then add JSON-LD that matches what the page visibly shows, using the most specific schema.org type that fits. Validate the markup, publish, and inspect the live URL.
Valid markup does not guarantee a rich result, a ranking, an AI citation or traffic. What it does is describe your page and the entities on it in a form machines can read consistently, and that description is worth getting right for every system that reads it.
What changes on your site
Six jobs cover most small sites. First, pick one primary question for each important page and answer it in plain language near the top. Second, use headings that describe the questions or tasks your readers have, with short sections, lists and tables where they help. Third, add accurate JSON-LD for the page, the organization behind it and its place in the site.
The other three are upkeep. Keep names, dates, prices and policies consistent across the site. Make pages crawlable, indexable and linked from other pages, with the important information in text rather than only in JavaScript. Then validate the markup and watch Search Console and Bing Webmaster Tools after publishing.
None of this is new, and that is useful. The work for AEO and GEO sits close to good search work, so a page that already answers its question clearly, carries true markup and loads without tricks is most of the way there.
What AEO and GEO mean
AEO is an industry term for making content easier for systems that answer questions directly to identify, extract and present. It can include traditional search features, voice assistants, AI Overviews and other answer interfaces. No single technical specification defines it, so treat it as an umbrella for the practices above.
GEO was formalized in the paper GEO: Generative Engine Optimization by Aggarwal and colleagues, first published in 2023 and presented at KDD 2024. The paper describes generative engines as systems that synthesize information from multiple sources and produce answers with inline citations. It proposes GEO as a black-box approach for increasing how visible a source is inside those answers.
In practice the two overlap. AEO leans toward being the page that supplies a direct answer, and GEO leans toward being a source a generated answer draws on and names. Both start with a page that states its answer clearly, so one piece of work serves both terms.
What the GEO paper measured
The paper built GEO-Bench, a set of 10,000 queries across diverse domains. In its experimental settings, tested content changes increased measured visibility by up to 40%. Adding citations, quotations and statistics produced significant increases in those tests, results varied by domain, and testing on Perplexity showed improvements up to 37% in that experiment.
Read those numbers with their limits beside them. The study measured visibility inside generated responses, not rankings, sales, traffic or guaranteed citations. It tested particular engines, prompts, domains and content changes, and generative systems are proprietary and change over time.
The paper also does not show that structured data alone causes AI visibility. Its tactics concern the content itself. If you add a quotation, a statistic or a citation to a page, add it because it is accurate and useful to the reader, and keep the markup in agreement with what the page says.
What structured data can and cannot do
Structured data is machine-readable information embedded in a page. JSON-LD is a JSON-based linked-data format that W3C standardized in 2020, and Google says it recommends JSON-LD when practical because it is generally easier to implement and maintain (Google structured data introduction). Structured data is the semantic layer; headings, text and lists are the visible layer, and the two should agree.
The word "correct" can mean four separate things, and none implies the next. The JSON can be syntactically valid. The types and properties can be valid schema.org vocabulary. The markup can meet the requirements of one specific Google feature. And Google can actually display something for the page. Passing the first three does not mean the fourth happens, and Google's general structured data guidelines say markup does not guarantee a rich result.
Two features show why current status matters. Google stopped showing FAQ rich results in Search in May 2026 and later removed its FAQ documentation (Google Search updates). FAQPage is still a valid schema.org type, so add it only when it describes a visible question-and-answer section, not to chase a result. Google's August 2023 announcement deprecated HowTo rich results, and a 2025 Google post on simplifying search results covers other retired appearances.
Finally, never mark up what readers cannot see. Hidden content, invented facts and self-serving reviews break Google's guidelines and mislead every other system that reads the page.
Put the answer on the page first
The visible page does most of the work. State the page's primary question in a heading or the opening lines, then answer it in a short paragraph that makes sense when quoted alone. Add the detail below it: steps, a list of options, a definition, an example. Descriptive headings let a reader and a machine both see where each answer lives.
This is a content practice, not a published ranking rule. Use question-style headings where your reader is really asking that question, and plain topic headings where they are not. The goal is that a person landing on the page can find the answer in seconds, and that a system pulling out one passage gets something accurate.
Keep the page useful without scripts too. If your answer only appears after JavaScript runs or behind a login, a crawler may never see it, and markup describing it would then describe something unseen. Check that the page is not blocked by robots.txt, noindex, a login or a hosting control.
A two-person team planning a content update can run this as a short routine. Write each page's question on a note, answer it in a sentence, and decide which schema.org type describes the page. Anything that cannot be answered visibly does not belong in the markup either.

Describe the page with JSON-LD
The example below is one fictional page for Example Bakery, using only example.com and example.org addresses. It describes a BlogPosting and the Organization that publishes it, connected by @id, inside a single @graph. Replace every name, URL, date, image and description with facts that appear on your real page.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "BlogPosting",
"@id": "https://example.com/blog/sourdough-answers/#article",
"url": "https://example.com/blog/sourdough-answers/",
"headline": "How Example Bakery Answers Common Sourdough Questions",
"description": "A fictional guide to common questions about sourdough bread.",
"image": ["https://example.com/images/sourdough-16x9.jpg"],
"datePublished": "2026-05-04T09:00:00-05:00",
"dateModified": "2026-05-18T09:00:00-05:00",
"author": { "@id": "https://example.com/#organization" },
"publisher": { "@id": "https://example.com/#organization" },
"mainEntityOfPage": {
"@type": "WebPage",
"@id": "https://example.com/blog/sourdough-answers/"
}
},
{
"@type": "Organization",
"@id": "https://example.com/#organization",
"name": "Example Bakery",
"url": "https://example.com/",
"logo": "https://example.com/images/example-bakery-logo.png",
"description": "A fictional bakery publishing information about bread.",
"sameAs": ["https://example.org/profiles/example-bakery"]
}
]
}
</script>BlogPosting is a more specific type within schema.org's Article, and Google's Article structured data page covers both. Google says Article has no required properties and asks you to use the applicable recommended ones: headline, image, datePublished, dateModified and author. Properties such as url, description and mainEntityOfPage are valid schema.org and help describe the page, though they are not Google's recommendations for this feature.
Google's Organization structured data page likewise lists no required properties. It recommends name, url, logo and sameAs where they apply. List real profiles of the same organization in sameAs; the single address above is a placeholder.
The breadcrumb trail goes in its own block, because it describes where this page sits in the site rather than the page itself.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BreadcrumbList",
"@id": "https://example.com/blog/sourdough-answers/#breadcrumb",
"itemListElement": [
{
"@type": "ListItem",
"position": 1,
"name": "Home",
"item": "https://example.com/"
},
{
"@type": "ListItem",
"position": 2,
"name": "Blog",
"item": "https://example.com/blog/"
},
{
"@type": "ListItem",
"position": 3,
"name": "How Example Bakery Answers Common Sourdough Questions"
}
]
}
</script>Google's Breadcrumb structured data page requires itemListElement, and each ListItem needs position and name. The item URL is recommended for every entry except the last. Match the trail to the breadcrumbs your readers see, and add it only when the page has a real place in a site hierarchy.
To build blocks like these without typing JSON by hand, the article schema generator produces Article-type markup from the details you enter.
What Google says about AI features
Google says AI Overviews and AI Mode have no additional technical requirements and no special AI schema, and that existing search fundamentals still apply: crawlability, text content, internal links and structured data that matches visible content (Google Search Central, AI features and your website, last updated December 2025). The same page notes there is no guarantee that a page is crawled, indexed or served.
Google also says an llms.txt file is not needed for Google Search (Google Search updates, 2026).
So, keep doing the fundamentals and keep your markup accurate. Do not invent an AI-only schema type; use the schema.org types that describe the page. And do not remove accurate markup just because one provider says it asks for nothing special; other engines and tools read the same page, and good markup costs little to maintain.
Bing and AI systems beyond Google
Bing offers a view of AI activity that Google's documentation does not. The AI Performance report in Bing Webmaster Tools shows citation activity, cited pages and grounding queries, which are grouped phrases describing the topic behind content cited in an AI answer. It reports trends, samples its data and covers supported Microsoft and partner experiences, and it separates citations from traffic.
That makes it a useful observation tool. If a page you changed starts appearing among cited pages, you have a pattern to follow up on. It is not proof that one change caused it, and it says nothing about rankings.
For ChatGPT, Claude and Perplexity, treat accurate schema as part of a well-described page rather than a lever for citations, and watch each system on its own: look at where your pages are cited, test your own questions, and let observed behavior guide your next change. The guide on whether AI systems use schema markup takes that question on directly, and the guide on which schema types to add first for AI search covers priorities.
llms.txt and schema do different jobs
The llms.txt file is a community proposal for a Markdown file at /llms.txt that summarizes a site and links to its important pages. Structured data describes the entities and facts on a single page inside that page's HTML. They answer different questions, so one does not replace the other, and neither replaces crawlable HTML, internal links, robots directives or a sitemap.
Google says llms.txt is not needed for Google Search. If you want to try it for other consumers, treat it as optional and low-risk, and keep the summary consistent with your pages. The llms.txt generator can draft one from your page list.
Add JSON-LD on your platform
Check what your site already outputs before adding anything. Many themes, SEO plugins and apps print their own schema, and a second block describing the same page can conflict with the first. View the rendered source of one page, find existing application/ld+json blocks, and edit those where you can.
On WordPress you can add JSON-LD through a theme, plugin, custom fields or a block or template integration; the official Metadata API covers how metadata is stored and retrieved, and our WordPress schema guide walks through the options. On Shopify, the Liquid structured_data filter can output schema.org data for product and article objects, so inspect the rendered page before adding more.
Webflow has a Schema markup field in page settings, and its help article, updated January 2026, describes manual entry, Webflow AI generation, publishing, validation and fields pulled from a Collection. In Next.js, render a script element with type application/ld+json in layout.js or page.js, and escape < characters when serializing values, as the Next.js JSON-LD guide describes.
For Wix, Squarespace and HubSpot, follow each platform's current official SEO documentation for structured data and code injection, then validate the rendered page. On a hand-built site, place the script in the page HTML and keep it in step with the visible content. The general steps are in our guide on how to add schema markup.
Validate and monitor after publishing
Run the same steps for every new or changed page. Paste the code or the live URL into Google's Rich Results Test and the Schema Markup Validator, or use the Schema Nova validator and schema checker. Each reports at a different level: syntax, schema.org vocabulary, or the requirements of one Google feature, so note which one you are reading.
Then inspect the live URL in Search Console with URL Inspection, and allow time for recrawling. Submit a sitemap when that fits your site. For Bing and Microsoft's AI experiences, check Bing Webmaster Tools AI Performance once the site has enough data.
A short list helps keep the whole routine in one place:
- Name the primary question and answer it in visible text.
- Choose the most specific type that matches the visible content.
- Add only properties that are true, and remove anything you cannot see on the page.
- Validate, publish, inspect the live URL and review trends over time.
- Update dates, prices, availability and business details whenever they change.

Treat the trends you watch as observation. A rise in citations, impressions or clicks after a change is worth noting, but it does not prove the change caused it.
Questions people ask
What do AEO and GEO mean?
AEO is Answer Engine Optimization, an industry term for making content easy for answer systems to identify and present. GEO is Generative Engine Optimization, a term formalized in a 2023 paper for improving visibility in generated answers. Neither has a single technical specification.
Is structured data required for Google AI Overviews?
Google says AI Overviews and AI Mode have no additional technical requirements and no special AI schema (Google Search Central, last updated December 2025). It also says structured data should match visible content. Accurate markup still describes your page for other systems that read it.
Does JSON-LD guarantee AI citations?
No. Valid markup does not guarantee a rich result, a ranking, a citation or traffic. For ChatGPT, Claude and Perplexity, watch where your pages are cited and judge each system separately.
What should a small site add first?
Start with Article or BlogPosting on editorial pages, Organization where the organization is described, and BreadcrumbList where the site has a real hierarchy. Add only properties the page visibly supports. The guide on which types to add first goes further.
Does llms.txt replace schema markup?
No. An llms.txt file is a proposed site summary, while schema markup describes the content of each page. Google says llms.txt is not needed for Google Search, and the two serve different purposes.
