AI citations are the short quotations an answer engine attaches to your website when someone asks a question your page should answer. Google AI Overviews, ChatGPT and Perplexity compress the open web into a reply and point at the pages they used. A buyer may read your sentence, or a competitor’s, before scanning a list of blue links.
Contents
Teams notice AI citations after a customer says the overview named a rival and skipped them. Classic rankings still send visits, and CodeWiz already publishes a separate technical SEO guide. This article asks a different question: whether a machine can lift a true, attributable sentence from your pages.
The work behind AI citations is concrete: crawlable HTML, schema that matches the page, and a clear line on who you are, where you are and what you sell. GEO and AEO are the usual names for that work. The check is whether your own sentence shows up in the reply.
What Answer Engines Extract for AI Citations
The unit of AI citations is a passage, not a whole URL. An engine looks for a few sentences that resolve the query and keeps a link so a reader can verify them. A homepage slogan, a hero animation and a PDF behind a button give it little to lift.
What travels is labeled text: a definition under a question heading, a list of inclusions, a comparison table, or a price, city, product name or date written as text. The heading tells the model what the sentence is about.
A strong organic position and steady AI citations are different outcomes. A page can rank and never be quoted when the fact is buried or built only in the browser. Measure the question you care about.
A reply may shorten your sentence or blend it with another source. Publish a sharper number or promise only when the business will stand behind it. The three products buyers mean are easy to tell apart, and none publishes a formula for which passage wins:
| Engine | What you see | What to record |
|---|---|---|
| Google AI Overviews | A generated summary on some searches, with links to supporting pages | Whether your domain is linked, and which sentence the summary echoes |
| ChatGPT | A reply with source links when web search is turned on | Each cited URL, and whether search was enabled for that chat |
| Perplexity | An answer with numbered sources beside the claims | The number next to the claim, and the page that number opens |
Overviews skip many queries. A ChatGPT reply with search turned off is not a quotation of your site. Perplexity numbers its sources, so a missing link is obvious. In each case the sentence has to exist where a fetcher can read it.
Crawlable HTML Versus JavaScript-Only Pages
Crawlable HTML means the answer is in the document the server sends. Fetch the URL and the headings, paragraphs, lists and product name are already there. A JavaScript-only page often sends a shell and fills it after a bundle runs.
Some fetchers never run that bundle. A single-page marketing site, a price painted after a client call, or a help article loaded through an API can look finished while the fetched HTML is almost empty. If a script is the only place the warranty exists, treat it as invisible to those fetchers.
WordPress content, a Laravel or PHP view, a server-rendered Node.js page and a static build put the answer in the first response. Hydration can still run menus. CSS may hide text that is already in the HTML, and an FAQ accordion is fine when the words are in the markup. Words fetched only after a click are a different case: put the answer in the document, and use JavaScript for behavior.
A passage written to be lifted looks like this. The heading is the question; the first paragraph is the answer:
<h2>What is included in the service?</h2> <p>The service includes on-site labor, the replacement part and a written 12-month warranty.</p>Repeat facts in HTML beside images of text. City URLs that swap only the city name have no local fact to quote. State a real constraint: the area, the hours or the part of the offer that applies there.
Schema Markup Behind AI Citations
Schema is a machine-readable copy of facts you already show. Clean schema makes AI citations easier to attribute, and it still has to repeat facts the visitor can read. Google’s structured data policies require markup that matches visible content. A hidden price, review or FAQ is a policy problem.
Four types cover most business pages. Generate JSON-LD from the same fields as the template, so a price change cannot update the visible offer and leave the markup stale.
- FAQPage. Mirror each visible question and answer in FAQPage schema, in the same words. One shared block on every URL is a template, not an answer.
- Organization. Name, URL, logo, city and country, with one brand name in the header, footer, contact page and markup.
- Product. Name and description the product page already prints. Omit price when it is not visible.
- Article. Headline, dates and author. Keep schema dates aligned with the byline, and change the visible date when the advice changes.
Validate JSON-LD against the live copy. Conflicting names, a broken logo URL or an author who does not appear on the site weaken the entity. The paragraph is still what gets quoted.
Answer-First Sections People Can Quote
Answer-first sections give AI citations a sentence that already stands alone. Put the conclusion in the first two or three sentences under the heading, then add conditions, exceptions and steps. A reader who stops there should still understand the point.
Write the heading as the question a buyer asks. “What is included in the service?” can be quoted. “Our approach to excellence” cannot. Use the words customers already use in sales calls and support mail. One fact per sentence is easier to lift than a paragraph that mixes warranty, price and territory.
Lead a list with the rule, and spell out units, currencies and places. “Delivery inside Banja Luka is scheduled within two working days” names a fact. “Fast delivery” does not. Publish only claims the business will keep, and put the company story below the answer.
Entity Clarity: Who, Where and What You Sell
Entity clarity tells AI citations which business the sentence belongs to. Answer engines mix sources. If two firms share a similar name, the page that states its name, city and offer in plain text is easier to attribute. A logo image without a text name does less than one sentence in the footer.
Use one name in the title, header, footer, contact page and Organization schema. Say the city, the country and the offer in concrete nouns, such as “WordPress maintenance for online stores”. A one-city workshop should name that city. A product should name itself in text on the pages you hope will be quoted. If buyers ask in Bosnian, Croatian or Serbian, that sentence has to exist in that language on a URL you control.
How to Check AI Citations by Hand
Keep a simple log of AI citations: date, question, engine, cited URL and whether the sentence exists in your HTML. No single dashboard settles this for every engine. A spreadsheet and an hour of careful queries beat a score from a tool you cannot audit.
Write about ten buyer questions, branded and unbranded. Run each in Google, in ChatGPT with search on, and in Perplexity. Record “not shown” when an overview is absent, and “no sources” when ChatGPT did not search.
- Log the question, date and engine before you read the answer.
- If your domain appears, confirm the matching sentence is in your HTML.
- If a competitor is cited, save the URL and the heading, and study the shape of the answer.
- Search the raw HTML for the fact. A fact that exists only after JavaScript is the gap.
- Read robots.txt for blocks on Googlebot and on documented fetchers such as GPTBot and PerplexityBot.
A public engine quotes what it can fetch, so a login wall needs a public paragraph you will stand behind. Repeat the list monthly and after template or robots.txt changes. Engines rewrite answers and drop sources without notice.
Common Mistakes That Keep You Out
Several quiet mistakes suppress AI citations on pages that otherwise look finished. Most of them are visible in the HTML or in robots.txt, which is why a manual check beats another pass of slogan writing.
- The answer lives only in client-side JavaScript, a modal or a PDF.
- FAQ, Product or review schema describes something the visitor cannot see.
- The same FAQ block is injected on every URL.
- The city, the offer and the name never appear as text, only as a slogan or a logo.
- Key facts sit in images, sliders or video, with no HTML equivalent.
- robots.txt or a firewall blocks fetchers you meant to allow.
- Article dates are missing, or schema dates contradict the page.
- Repeated phrases leave no calm sentence that can be quoted.
When You Need This Work
Commission this when absent AI citations are already showing up in sales calls, or when buyers ask an engine before they scan organic results. It fits service firms, local businesses, stores and publishers that can stand behind a specific sentence.
While the site cannot be crawled or indexed, start with a technical SEO pass, then make key pages quotable. A private chatbot is a separate job: RAG quotes your files for your users. Public answer engines quote the public web. Our AI/ML integration work covers that internal system.
How We Work
The last step repeats the same questions and records whether AI citations moved. The earlier steps make that comparison possible. We keep the loop short so you see changed pages, not only a strategy document.
- Questions. We collect prompts buyers, support and sales already hear, and pick the URLs that should answer them.
- Fetch audit. We compare raw HTML with the rendered page, review FAQ, Organization, Product and Article schema, and read robots.txt.
- Answer-first plan. We outline headings, opening sentences, lists and the entity facts (name, place, offer) each priority URL must state.
- Implementation. We change templates and content in WordPress, Laravel/PHP or Node.js so the answer is in the HTML and the schema matches it.
- Manual check. We run the question list in Google AI Overviews, ChatGPT and Perplexity, log cited URLs and adjust the pages the log flags.
Why CodeWiz
- A full-stack agency. CodeWiz was founded in 2023 by Dragan Plavšić in Banja Luka. The work covers WordPress, Laravel/PHP, Node.js, APIs, DevOps and Cloudflare.
- The surrounding practice. Performance, security, SEO and e-commerce are part of the same work. The live site scores 100/100/100/100 on PageSpeed.
- Both sides of retrieval. AI work includes RAG and local models, for public pages and for private retrieval on your own data.
- How we choose tools. Technology serves the business. Solutions are tailored, open-source comes first, and pricing stays fair.
- Talk to us directly. Email hello@codewizdev.com, or message on Viber or WhatsApp at +387 66 852 373.
Frequently Asked Questions
Do AI citations replace ordinary rankings?
Rankings, maps and on-site search still send people to your pages. Track clicks and quotations in separate logs, because a change on one surface can leave the other untouched.
Does FAQ schema guarantee we will be quoted?
Publishing FAQ schema does not by itself produce AI citations. The questions have to be visible, specific and true. Markup copied from a plugin demo, or repeated on every URL, gives an engine nothing unique to attribute to you.
Can a JavaScript application be quoted?
It can, when the server sends the answer in the HTML. On React, Vue or a headless front end, render or prebuild those pages and keep interaction in JavaScript.
How often should we repeat the manual check?
Monthly fits most sites, plus a run after template or robots.txt changes. Keep the question list stable. Engines rewrite answers without notice.
Is this the same project as technical SEO or a private chatbot?
Technical SEO gets a site crawled and indexed. A private RAG chatbot quotes your documents for your users. This work makes public pages easy to attribute in Google AI Overviews, ChatGPT and Perplexity. Scope the three tasks separately.
Bring Your Questions to CodeWiz
Bring us the questions you want to own and we will trace how AI citations currently treat your pages. Contact CodeWiz or look through the services we implement, from page templates to schema that matches them.