How AI Decides Which Sources to Cite

Imagine a potential customer asking an AI for a service provider in your industry. Does your company appear in the response? If not, you’re missing out on a growing channel. According to a G2 survey from April 2026, 51% of B2B software buyers already start their research with AI more often than with Google. AI systems don’t make recommendations at random. They follow a clear logic when selecting their sources, and this logic differs fundamentally from the way Google ranks results.

This article explains how this logic works, why traditional SEO strategies are only of limited help in this context, and what criteria AI uses to decide which companies to recommend. The technical terms for this are LLMO (Large Language Model Optimization) and GEO Optimization (Generative Engine Optimization). If you’d like to know which specific measures will boost your AI visibility, we recommend our guide to AI visibility for businesses.

Can AI search find your business? Here’s how to test it

Before you invest in LLMO or GEO optimization, you should know where you stand. With a simple self-assessment, you can find out how your AI visibility is doing in just five minutes.

The 5-Minute Self-Test

Open three AI platforms and ask the same question on each one. Phrase it the way a potential customer would search for it:

Replace "Web Agency Hamburg" with your own industry and location. Test at least three variations: a general industry search, one with a specific specialization, and one that includes your location. For each test, note whether your company is mentioned by name, whether the information displayed is correct, and which competitors are recommended.

What the results reveal

There are three possible scenarios: Your company is mentioned and described accurately (ideal scenario); it doesn’t appear, but competitors do (urgent need for action); or the AI doesn’t mention any specific companies at all (still the norm in many niches—those who act now will be first to the market).

Repeat the test regularly. AI search results are not static and change as new sources are added or existing information is updated. According to a Seer Interactive study , 50% of all citations on Perplexity already come from content published in 2025 or later. For Google AI Overviews, this figure is 44%. This means that new content is displacing old sources more quickly than in traditional Google rankings. To learn more about how you can systematically measure your AI visibility, check out our article on measuring AI visibility.

Here’s how AI selects its sources

To build AI visibility in a targeted way, you need to understand how AI search selects its sources. AI systems don’t just make up their answers on the spot. They use a two-step process called Retrieval-Augmented Generation (RAG).

How the RAG Process Works

First, AI systems search through large amounts of web content and databases for relevant information (retrieval); then, they formulate a coherent response based on the sources they have found (generation). In a study on RAG quality (2025), Google Research demonstrated that the error rate of AI responses rises from 10.2% to 66.1% if the retrieval phase does not provide sufficiently high-quality sources. The quality of the sources found therefore directly determines the quality of the AI response.

How AI Identifies Trustworthy Sources

The key difference from traditional Google search: Instead of displaying ten blue links, the AI selects the most reliable sources, summarizes their content, and identifies the company that most frequently and consistently matches the answer to the question asked. In doing so, it evaluates each potential source based on several factors. The Seer Interactive study on Content Recency and AI Visibility (2025) shows that 65% of all AI bot accesses are to content from the past year, and only 6% are to content that is older than six years. Recency is thus one of the strongest filtering mechanisms.

In addition to timeliness, consistency across sources plays a key role. If three independent websites describe your company as an expert in a specific field, the AI will weigh that significantly more heavily than a single self-promotional statement on your own website. An Ahrefs study of 75,000 brands (2025) backs this up with concrete figures: Branded web mentions (mentions of your company on third-party sites) correlate with AI citations at a Spearman’s correlation coefficient of 0.66, while traditional backlinks only reach 0.22. This means that mentions on third-party sites are three times as important for AI visibility as traditional SEO link-building measures. Brands in the top quartile of this study received an average of 169 AI citations, compared to just 14 for brands in the bottom quartile.

Why Each AI Platform Prefers Different Sources

The three major AI platforms pursue fundamentally different strategies when it comes to source selection. An analysis of 680 million AI citations by Profound (2025) highlights these differences:

• ChatGPT favors authoritative sources of knowledge. Wikipedia accounts for 7.8% of all citations, making it the most frequently cited single domain. The system relies on established, editorially vetted sources of information and gives greater weight to institutional media than to user-generated content.
• Perplexity relies more heavily on community knowledge and real-time sources. Reddit accounts for 6.6% of all citations and is the most-cited single domain. The system weights user reviews, ratings, and current discussions significantly higher than other platforms.
• Google AI Overviews use their own search results as a basis and draw on a broad mix of sources. Reddit (2.2%) and YouTube (1.9%) top the list of most-cited domains, but the distribution is more even than on other platforms.

These differences have direct consequences. Those who want to be visible on just one platform can optimize their efforts specifically for that platform. Those aiming for cross-platform AI visibility must maintain a presence on multiple levels. For a detailed comparison of the platforms, we recommend our separate article on AI search platform comparisons.

What distinguishes AI search from traditional Google ranking?

Many companies assume that good Google rankings automatically lead to AI visibility. The data paints a different picture. The key difference: Google is all about position (whoever ranks first gets the most clicks), while AI search is all about citation (the AI selects the most trustworthy source, regardless of its Google ranking). The following table highlights the key differences:

Why Ranking #1 on Google Is No Guarantee of AI Visibility

Only 38% of the pages cited in Google AI Overviews come from the top 10 organic search results, according to a 2026 Ahrefs study. In mid-2025, that figure was still 76%. Google now breaks down search queries into several sub-queries and draws on separate sources for each one. As a result, pages that rank 30th or 50th in the original search can still appear as sources in the AI response. AI evaluates content in a fundamentally different way than the Google algorithm. While Google weights links, load time, and technical signals, AI checks whether a piece of content answers a question precisely and whether the information is corroborated by other sources. A blog article that is technically perfectly optimized but lacks depth of content is ignored by AI. AI systems have been trained on billions of texts and are familiar with content that appears on page 5 of Google’s results or isn’t indexed at all—including expert forums, scientific papers, and industry directories—which may be more relevant for AI citations than for Google rankings.

Why AI Traffic Converts More Effectively

At the same time, the data shows that visitors who arrive at a website via AI recommendations have an above-average willingness to purchase. Similarweb (2026) estimates the conversion rate for AI referral traffic at 7.1%. That’s nearly as high as for paid search engine advertising (7.8%) and significantly higher than organic Google traffic. The reason: Visitors who arrive at your website via an AI response have already asked a specific question and received a targeted recommendation. Unlike a traditional Google click, where the user first compares several results, the AI has already made a preselection and identified your company as a good fit. This pre-qualification by the AI makes the traffic more valuable than almost any other organic source.

How AI Systems Compile Information

To understand why AI recommends certain companies and not others, it helps to look at how AI systems compile information from various sources. The foundation is always the RAG process: The AI searches the web, collects relevant text passages, and uses them to construct a response. However, the path from the search query to the final recommendation is more complex than a simple match.

How AI Recognizes Businesses as Entities

AI systems work internally with clearly defined concepts such as companies, people, products, or places. When a user searches for “web agency Hamburg,” the AI doesn’t simply look for pages that contain those words. Instead, it identifies concepts that match the query and then checks what information is available from various sources for each concept. The more independent sources that describe a company—and the more consistent those descriptions are—the more likely it is to be included in the response. This mechanism explains why consistent NAP data (Name, Address, Phone) across business directories, Google Business, and the company’s own website is so important for AI visibility. If the information doesn’t match, the AI cannot uniquely identify the company and will instead select one with clearer data.

How Citation-Worthiness Determines Visibility

The GEO study by Princeton, Georgia Tech, and IIT Delhi (2024), published at the ACM KDD Conference, tested nine different optimization methods and measured how much each method increased visibility in AI-generated responses. The results show that GEO-optimized content achieves 30 to 115% higher visibility in AI responses. Adding concrete statistics increased the citation rate by 40%, while including expert quotes increased it by as much as 115% in certain categories. This means: It’s not just about whether your content exists, but whether it’s in a format that AI classifies as quotable. A paragraph that begins with a clear statement and backs it up with numbers or facts is more likely to be included in an AI response than a long text that merely circles around the topic. AI looks for text passages that can serve as standalone answers without requiring additional context. In our article on technical website optimization for AI, we explain in detail which structural measures make your content quotable.

What does this mean for your AI strategy?

How AI Weights Its Own Content and Third-Party Sources

The Ahrefs study of 75,000 brands shows that AI systems draw on both their own content and third-party sources, but weight them differently. In the study, 84% of all AI citations came from earned media—that is, mentions on third-party sites. YouTube mentions showed the strongest single correlation with AI visibility at 0.74, followed by general web mentions (0.66) and branded anchor text (0.53). Traditional backlinks were significantly lower at 0.22.

For businesses, this means that their own website is and will remain the central source of information, because that’s where the content is that AI systems can cite. At the same time, AI needs external validation to build trust in this content. An industry listing on Clutch with genuine customer reviews, a guest article in a trade magazine, or a mention in a Reddit thread provides the AI with the independent context it needs to consider your company as a recommendation. We describe the specific steps you can take to strengthen your presence on third-party sites in our GEO guide.

LLMO as an Ongoing Process

Based on our experience with over 60 projects for small and medium-sized businesses at Evelan, we know that the combination of consistent company data, citable content, and a broad presence on third-party sites determines whether a company appears in AI responses. LLMO is not a one-time project, but an ongoing process. AI systems regularly update their selection of sources, and competitors who invest early secure positions that are harder to capture later on. The good news: Many LLMO and GEO measures also strengthen your Google ranking. Structured data, citable content, and mentions on third-party sites are signals that benefit both areas. You can find a complete overview of all these measures in our guide to AI visibility for businesses.

Related Evelan Articles

• GEO: How to Make Yourself Visible in AI Searches
• How AI Is Changing Brand Identities: Opportunities, Risks, and Practical Applications
• AI Content and Google Rankings: What the Data Really Shows
• SEO Basics: How to Make Your Website Visible
• Google Algorithm Updates and Your Ranking

Sources

• G2: B2B Software Buyer Behavior Report (2026)
• Ahrefs: 38% of AI Overview Citations Pull From The Top 10 (2026)
• Ahrefs: Top Brand Visibility Factors in AI Search, 75k Brands Studied (2025)
• Similarweb: Gen AI Stats 2026, AI Visibility Trends and Insights (2026)
• Profound: AI Platform Citation Patterns, 680M Citations Analyzed (2025)
• Google Research: Deeper Insights into RAG, The Role of Sufficient Context (2025)
• Seer Interactive: AI Brand Visibility and Content Recency (2025)
• Princeton, Georgia Tech, IIT Delhi: GEO, Generative Engine Optimization (ACM KDD 2024)
• TechCrunch: ChatGPT reaches 900M weekly active users (2026)
• BrightEdge: AI Overview Citations Now 54% from Organic Rankings (2026)