AEO Fundamentals: How to get your site cited by AI in 2026

Search is splitting in two. Half your customers still type queries into Google. The other half ask ChatGPT, Perplexity, Claude, or Gemini. Those AI engines don’t pick their citations the way Google ranks its results, and a site tuned perfectly for one can be invisible in the other. Most contractor sites win neither.

This pillar is the foundation of every other guide Dynamic Promotion will publish about Answer Engine Optimization. It defines the terms, walks through the research, lays out the editorial disciplines that make content AI-citable, and tells you how to measure whether any of it is working. Spokes from this pillar go deeper into specific tactics (schema graphs for trades, llms.txt patterns, freshness automation, hyper-local AEO). Every spoke assumes you read this one first.

What is AEO, and why does it matter now?

AEO stands for Answer Engine Optimization. The job is to structure your content so AI engines pick it up and cite it when users ask questions. Traditional SEO aims for the ten blue links: rank a page high enough that someone clicks through. AEO aims for the answer itself: get your prices, your differentiators, and your real customer quotes into the synthesized recommendation the user actually reads, with your domain credited underneath.

The surface is moving. Users who would have typed “best roofer near me” into Google in 2022 increasingly ask Gemini, ChatGPT, or Claude the same question. Sometimes through chat. Sometimes through voice. Sometimes through whatever AI their phone shipped with. Instead of ten links to compare, the answer that comes back is a paragraph naming two or three businesses and explaining why. If you aren’t in that paragraph, you don’t exist for that query.

There’s a second reason AEO matters in 2026 that’s easier to miss. AI-engine traffic converts at a much higher rate than traditional organic search. By the time a user clicks through from a ChatGPT recommendation, they’ve already read your reviews, your pricing, and your service area in the AI’s summary. They’re closer to booking than the equivalent Google click, where they’re still comparing you against nine other tabs.

The opportunity in service verticals is that almost nobody has done the work yet. Most contractor sites, organization sites, and local-business sites in mid-tier markets ship without LocalBusiness schema, without llms.txt, without question-format headings, without quoted testimonials marked up as Quotation schema. Early movers are walking into an empty room.

How is AEO different from SEO?

The shortest way to say it: SEO ranks pages, AEO selects passages.

Google’s ranking algorithm scores entire pages against a query and orders the top results. A user reads the page that ranked first. Generative AI engines work differently. They pull dozens of candidate pages, extract specific sentences and paragraphs from each, stitch the extracts into a single answer, and credit the sources. The pages that “win” in AEO aren’t the ones that rank well. They’re the ones that contain a quotable, specific, well-attributed passage. A page can sit at position #47 on Google for a query and still get cited heavily by ChatGPT, as long as it has the right kind of extractable content.

That changes how you write:

• Headings carry real weight. AI engines anchor passage selection to your H2s. A question-format heading (“How much does roof replacement cost in Baton Rouge?”) gets selected far more often than a topic-format one (“Pricing”). Users prompt AI in questions, so write your headings as the questions they ask.

• First sentences do most of the work. AI extractors weight the first one or two sentences after each H2 as the most likely answer passage. If your real answer is buried in paragraph three, the extractor never gets there.

• Specificity beats keyword density. “Average roof replacement in Baton Rouge: 2-3 days, $7,500-$18,000 depending on tear-off and decking condition” is extractable. “Fast, affordable roofing” is filler. AI engines route around filler.

• Structured data does different jobs for different consumers. Google uses your LocalBusiness schema to build the knowledge graph behind People-Also-Ask boxes. AI engines use the same schema to verify you’re a real entity with verifiable claims, especially the sameAs array that links to LinkedIn, Facebook, GBP, and Wikidata. Both want the schema. They use it for different things.

The honest version is that the work overlaps roughly 80%. Clean semantic HTML helps Google’s crawler and AI extractors equally. Schema helps both. Page speed helps both. Mobile responsiveness helps both. The foundation is the same. Two different editorial layers go on top, and most agencies have only built one of them.

What does the research actually say?

The most important empirical paper on this topic is the 2024 GEO: Generative Engine Optimization study from researchers at Princeton, Georgia Tech, and the Allen Institute for AI. The team tested nine different content-optimization strategies across thousands of generative-search queries and measured how each one affected the likelihood that a page got cited in the AI’s answer.

The results were lopsided in a useful way. Summarizing the paper’s findings:

Notice what didn’t show up in the high-impact list: keyword density, exact-match keyword targeting, internal linking density, and most of the traditional SEO playbook. Several of those tactics had no measurable effect on AI visibility, and keyword stuffing actively hurt it. The AI engines read keyword-saturated pages as low-quality and pushed them down.

Four tactics did most of the lifting. Each one is an editorial discipline you can apply to a single service page in an afternoon:

Read in plain English: quotations are verbatim text from a named, verifiable source, marked up so AI extractors can pull cleanly with attribution. Statistics are specific, sourced numbers, not “many” or “most” or “industry-leading.” Fluency is prose a human would actually say out loud, with no keyword-optimized acrobatics. Inline citations are hyperlinks to real outside sources placed inside the body of your content, not buried in a footer.

Three of those four are pure writing work. No engineers required. Any contractor who can write a clear paragraph and source a real Google review can compound a 70%+ AEO lift over a site that hasn’t done the work. The fourth, inline citations, is part editorial (knowing what to link to) and part technical (consistent link rendering with rel="cite" per Google’s structured-data guidance).

Practitioner tracking since the paper published points at two more signals worth watching: freshness (recently updated pages keep showing up in citation sets at roughly twice the rate of stale ones in platform-tracking write-ups) and entity verification (sites with populated sameAs arrays in their Organization schema earn more citation confidence than orphaned domains). Neither has GEO-paper-grade evidence behind it yet; treat both as working heuristics.

What signals do AI engines weight when selecting citations?

The GEO study covered editorial tactics. Beyond those, AI engines weight several structural and metadata signals when deciding which passages to extract. Understanding them is how you turn occasional citations into a consistent share.

Entity verification. AI engines build internal confidence scores for whether your business is real. The strongest signal is your Organization (or LocalBusiness subtype) schema with a populated sameAs array linking to LinkedIn, Facebook, Google Business Profile, Wikidata, Crunchbase, and any industry-specific directories you belong to. The full Schema.org reference catalogs every property you can use. A roofing contractor whose LocalBusiness schema includes the GBP URL, the LinkedIn company page, and the BBB profile earns more citation weight than one with a bare schema block.

Freshness signals. Both visible and structured. The visible signal is an “Updated [Month Year]” string near the top of the page that matches the dateModified schema property. The structured signal is the dateModified field itself, kept current. Pages that haven’t been touched in 12 months start dropping in AI citation frequency even when they still rank on Google, because AI engines assume that pricing, hours, and service offerings change.

Speakable specifications. SpeakableSpecification is a Schema.org property that marks specific sentences or DOM elements as the parts voice assistants and AI summarizers should read aloud. It’s a small signal individually, but it adds up. Pages that mark their lede paragraphs and key statistics as speakable get cleaner extractions when an AI is asked a voice-style question.

Server-rendered HTML. Foundational enough that it gets its own section below. The short version: AI crawlers fetch your raw HTML and parse it. They generally do not execute JavaScript. If your hero headline is rendered by client-side React, an AI crawler sees an empty <div>.

Schema completeness. Google’s structured data guide lists every canonical type. For service businesses, the high-impact set is Organization or LocalBusiness, Service (one per offering), FAQPage on every Q&A section, Review sourced from real Google reviews (never invented), BreadcrumbList, and Article or BlogPosting on guides like this one. Validate every page through Google’s Rich Results Test before launch. Schema errors silently downgrade your AI citation eligibility.

Citation reciprocity. Pages that link out to authoritative sources get cited by AI engines more often than pages that don’t. That sounds backward. You might assume that linking out gives away credibility. The data goes the other way. A page with five outbound links to peer-reviewed studies, government sources, or industry publications signals that the content is grounded. A page with zero outbound links reads to AI engines like an echo chamber.

Where most AEO results actually come from

A counterintuitive split that practitioners across the AEO space keep landing on: roughly 30% of your AEO results come from work you do on your own site. The other 70% comes from earned media, third-party citations, and entity-graph reinforcement off your site. Treat the exact ratio as directional rather than measured; the imbalance is the point.

That’s the opposite of how traditional SEO breaks down. Classic SEO is closer to 50/50: half on-page (content, schema, page speed, internal links) and half off-page (backlinks from authoritative domains). AEO leans harder on off-page validation because AI engines treat third-party confirmation as a credibility signal. Your Organization schema can claim you’re a roofing expert, but the AI wants to see your name on industry publications, your team quoted in trade journals, your business profiled by local news outlets, your reviews populating Google Maps, and your team’s LinkedIn profiles populated with relevant credentials.

The 30% you do on your own site is everything this pillar covers. Do it first because everything else depends on it. Once the on-site foundation is in place, the highest-leverage next investments live off your domain:

• Real, attributable quoted testimonials from real customers, posted on platforms AI engines crawl (Google Reviews, BBB, industry-specific directories), not only on your own site
• Listings in industry directories, association memberships, and any data source AI engines treat as authoritative for your vertical
• Author and team-member entity profiles: LinkedIn pages, industry publication bylines, conference talks, podcast guest spots
• Citations from third-party publications: local newspaper features, trade journal articles, blog posts from adjacent businesses
• A Google Business Profile that you maintain like it’s your homepage: photos, posts, Q&A monitoring, review responses

Most of that work involves no engineering at all: editorial work, relationship work, and discipline over time. DP’s AEO consultation playbook covers the off-site half in depth, including the monthly rhythm we use for citation acquisition and the templates we use to scaffold real-customer review collection. This pillar focuses on the 30% on-site foundation because every other AEO investment compounds against it. If your on-site signals are weak, no amount of off-site work will fully cash in.

HTML-first content is the foundation

The single most important AEO rule, and the one most often broken: critical content has to live in the server-rendered HTML response, not in client-rendered JavaScript.

There are a lot of these crawlers now. OpenAI’s GPTBot, Anthropic’s Claude-Web, PerplexityBot, Google’s various crawlers, and a long tail of smaller ones. They all do roughly the same thing: fetch your page and parse the raw HTML they get back. They generally do not execute JavaScript. They don’t wait for React to hydrate. They don’t trigger an Intersection Observer to lazy-load content into view. If your hero headline gets injected by client-side code after DOMContentLoaded, the crawler sees an empty <div> where your value proposition should be.

The patterns that disappear from AI engines:

• Hero text rendered by a client-side framework after initial HTML response
• Text baked into hero background images instead of real <h1> and <p> elements
• Accordions and tabs that only render content on user interaction (the content needs to be in the DOM, just hidden via CSS)
• iframes containing critical text (crawlers don’t follow iframe boundaries)
• Content gated behind authentication (obvious, but worth flagging)
• Content that fetches on scroll position
• Single-page-app routing where each “page” is JS-rendered without a meaningful HTML fallback

The defensible architecture is static-first with progressive enhancement. All critical content ships in the initial HTML response. JavaScript layers on top for interactivity but is never required for visibility. That’s why DP standardizes on Astro for every build. Every page ships as plain HTML with all content visible on first paint, before a single byte of JavaScript runs.

The quick diagnostic: open your page, right-click, View Source. Look at the actual response from the server, not the DevTools-rendered DOM. Search for your hero headline as plain text. If you can find it, AI crawlers can find it. If View Source shows a <div id="root"> with nothing inside, you’re invisible to AEO no matter how much other work you’ve done.

One bonus: HTML-first content is also what WCAG 2.2 requires for screen reader users. The same architecture that makes you AI-discoverable makes you accessible to assistive technology. Same work, two wins.

The two patterns that move the needle: quotations and specific data

The GEO study put quotations at +41% and statistics at +33%. Those are the two biggest editorial moves you can make. Here’s how to do them without breaking the integrity that makes them work.

Quotations

Every service-detail page should carry at least three verbatim quotations from real customers. In priority order, source them from:

1. Google Reviews of your business — verbatim, attributed by first name + last initial + city
2. Direct testimonials from named customers — written or verbal, with explicit permission to publish
3. Industry-publication quotes that mention your business or team members — secondary, but useful for credibility

Mark them up with Schema.org’s Quotation type. When you can, include an isBasedOn link back to the verifiable source URL (the Google Review page, the LinkedIn post, the publication). The visible HTML uses a <blockquote> with a nested <cite> for attribution:

<blockquote cite="https://g.page/r/CXXX...">
  <p>"They replaced our roof in two days flat, came in under
  their estimate, and the crew cleaned up better than we
  did the day we moved in."</p>
  <cite>— Sarah M., Baton Rouge · Google Review · 5★</cite>
</blockquote>

Three rules protect the system:

• Never invent testimonials. Fabricated quotes break the verification heuristic AI engines rely on, expose you to legal trouble, and get caught at scale through source cross-referencing.
• Never edit a customer’s words. You can shorten with an ellipsis (...) and add a clarifying word in square brackets ([their roofer]). You can’t rewrite what they actually said.
• Always link to the verifiable source. If a quote came from a Google Review, link to the review URL. If it came from a personal email, you need written permission to publish and a privacy-respecting attribution (first name + last initial is the standard).

Specific data

Replace generic marketing language with sourced numbers. Examples:

The right-column versions are longer, and that’s fine. AI engines pull substantive sentences and skip generic ones; a longer sentence that carries real information beats a short one that carries none.

The discipline that anchors everything: every number on your site has to be defensible. If you say “247 roofs installed,” that number should match what your job-management software can produce on demand. If you say “average 8 years’ roofing experience,” you should be able to walk through the team roster that proves it. Specific data that isn’t verifiable is worse than generic data, because it’s a claim a regulator, a competitor, or an attorney can pull apart.

Anti-marketing language discipline

This is the discipline most agencies skip. It’s also the easiest one: delete every adjective that doesn’t carry a verifiable claim.

The disqualified list:

• “Industry-leading”
• “Best-in-class”
• “World-class”
• “Premium”
• “Trusted”
• “Award-winning” (unless you can name the award)
• “Leading provider of…”
• “We pride ourselves on…”
• “Committed to excellence”
• “Customer-first”
• “Cutting-edge”
• “State-of-the-art”
• “Comprehensive solutions”
• “Innovative approach”

These phrases are background noise. AI engines treat them as wallpaper. They don’t raise your citation weight and they don’t lower it, but they occupy space that could have carried a substantive claim. Every “industry-leading” you delete is room for “GAF Master Elite certified, top 2% of US contractors by sales volume,” which actually moves the needle.

The test for any descriptive phrase: can you verify the claim independently? If yes, it stays. If no, delete it.

The reason this is uncomfortable on existing content is obvious. You’re either going to find proof for the claims you’ve been making for years, or you’re going to delete them. Either way, the site that comes out the other side is the one AI engines will cite.

How do you measure AEO success?

Most AEO measurement tooling is either expensive (paid SaaS at $99-499 per month) or noisy (rank trackers that don’t actually capture AI citations). The good news: a free manual system is both more reliable and more useful than most paid alternatives, at least for the first 90 days of an AEO effort.

The free system has three parts.

Monthly prompt audits. Pick 5 to 10 commercially important queries. These are the questions your real customers would type into ChatGPT or Perplexity. For a Baton Rouge roofer that might be “Who’s the best roofer in Baton Rouge?”, “What does roof replacement cost in Baton Rouge?”, “How do I find a storm-damage roofer in the Watson area?” Run each query monthly across:

• ChatGPT (with web browsing enabled)
• Perplexity
• Claude (with web search)
• Google AI Overviews (it shows up sometimes)
• Gemini

Log which sites get cited in each answer. Track over time: are you appearing, are you appearing more often, are you being mentioned by name even when not linked?

GA4 referral filtering. AI engines tag their outbound referrals. In GA4, filter by Source and look for:

• chatgpt.com
• perplexity.ai
• claude.ai
• gemini.google.com
• you.com

Visits from these sources are AEO traffic, meaning users who clicked through because an AI cited or recommended you. The volume is small but the conversion rate is high. AEO referrals typically convert at multiples of organic search traffic.

Brand-mention monitoring. A free Google Alert on your business name catches blog posts, news articles, and forum threads that may not link back to you. Some AI engines extract mentions even without hyperlinks, anchored to the name property in your Organization schema, so unlinked mentions still contribute to your AEO footprint.

Paid tools (Profound, Goodie, Otterly, and others) automate the prompt-audit work at $99 to $499 per month. They become cost-effective once you’re tracking 50+ queries or watching competitors at scale. For the first year, the manual rhythm is usually enough and has the side benefit of making you read the AI answers yourself, which is where the intuition actually develops.

Speakable markup helps measurement too. The SpeakableSpecification properties on your WebPage node give voice assistants and AI summarizers a structured hint about which parts of the page hold the most extractable content. The same hint that helps a screen reader prioritize content makes it easier for an AI extractor to land on the passage that answers the user’s question.

For sites with a real publishing cadence (regular blog posts, frequent service updates), IndexNow integration is the highest-leverage technical investment for AEO. IndexNow is a protocol backed by Microsoft, Yandex, and several smaller search engines that lets your site push notifications to search and AI engines the moment a page is published or updated. Without IndexNow, new content has to wait on the crawler’s schedule, which can be days or weeks. With IndexNow, new content can appear in AI engine answers within hours. DP ships IndexNow at the Grid (Tier 4) baseline and offers it as an add-on at lower tiers.

From this pillar to your AEO playbook

This pillar covered the foundation: what AEO is, why it matters in 2026, the research that grounds the methodology, the high-impact patterns (quotations, specific data, HTML-first structure, anti-marketing language), and how to measure whether the work is paying off. The hardest part isn’t knowing the rules. It’s applying them consistently across every page, then auditing the results month after month and adjusting based on what you see.

If you’re going to implement AEO on your own business yourself, the right order is foundation first (audit your current HTML, fix any client-side-rendered critical content, deploy LocalBusiness schema, add llms.txt), then editorial discipline (rewrite your service pages with question-format H2s, replace generic claims with specific data, source three real Google Review quotes per page), then measurement (set up the manual prompt audit, configure GA4 AEO filtering, start tracking). Expect 60 to 90 days before AEO results compound noticeably. The lift is real, but it builds. It doesn’t appear overnight.

If you’d rather hand the technical layer to someone else while you focus on the editorial work, that’s specifically what Dynamic Promotion exists to do. Every site DP builds ships with the AEO baseline described here: HTML-first Astro architecture, full schema coverage, and llms.txt. The Maintenance retainer adds monthly performance checks, and IndexNow, JSON SSOT, and AI-citation monitoring are available as the business grows. Future deep-dives will branch off this pillar into LocalBusiness schema patterns, hyper-local AEO for multi-city operations, and retrofit audits for existing sites.

The mandate that drives every page on this site is the same one we sell to every client: built for search, ready for AI. The agencies that figure that out in 2026 are going to own the next decade of service-business lead generation. The ones that don’t will keep optimizing for a search surface that’s shrinking faster than they realize.