What 30,000 Citations Taught Us About AI Search

Tl;DR

Generative AI is changing how search works, and it’s more than just a new look. It's changing where search engines find their information. We analyzed ~30,000 sources from thousands of B2B search queries on traditional Google Search and Gemini (with Web Search) and noticed a significant shift.

Right now, Gemini tends to pull information from official company websites (both yours and your competitors'), while it pays less attention to the discussion forums and trade publications that have been a big part of SEO for years. In this blog, we’ll walk you through the data, talk about why this shift is happening, and give you a new playbook to help your B2B company adapt and grow. The main takeaway for marketers and developers is that the best way to build trust and visibility today is to experiment and focus on creating structured, authoritative content.

What’s Changing in AI Search Results—And Why It Matters

For a long time, the main goal of SEO was to get a spot on the first page of Google’s "10 blue links." That time is over. Today, the search results page is a hybrid, mixing those classic links with AI-powered summaries, like Google's AI Overviews. This shift from targeting blue links to influencing AI summaries requires a new discipline: Generative Engine Optimization (GEO). The goal of GEO is no longer just ranking, but achieving trusted AI visibility within the generated answers themselves.

It’s important to remember that this new setup isn't set in stone. Generative models are constantly being updated. The way they source information today is just a snapshot in time. The next big update could change things all over again. Because of this, the most important thing you can do is continuously monitor these changes and be ready to experiment. This blog offers a starting point for that process.

How Did We Collect Data from Google and Gemini for Our AI Search Study?

To be transparent about our process, we designed a study to see how citations differ between Google's traditional search and Gemini's generative answers. The data for this analysis was collected between July 25th-26th, 2025.

Our approach started with framing the experiment around a diverse set of companies. We chose three with different business models and go-to-market motions to see how trends might vary:

• MotherDuck (a developer product)
• Mindtickle (a horizontal B2B SaaS product)
• Prodigal (a vertical FinTech product)

With these companies as our subjects, we then followed a three-step process:

• Question Graph Generation: We created 250 to 350 questions for each of our three target companies to simulate a typical B2B buyer’s research process across different stages: Awareness, Interest, and Consideration.
• Querying the Engines: We sent each question to both Google’s Custom Search API and Gemini’s Generate Content API, with Google Web Search enabled as the grounding tool, to see which sources each would cite.
• Categorizing the Sources: We extracted every citation URL from both result sets. After normalizing them to their apex domain (so blog.example.com became example.com), we mapped each domain to one of 12 distinct categories, such as Self, Competitor, Ecosystem, or Discussion Forum.

While this approach gives us a detailed snapshot, we know this is a small sample size. These findings might not apply to everyone. Our goal isn't to state a new universal rule for SEO, but to highlight major trends and emphasize a key point: every company needs to experiment to see how AI views its own specific corner of the market.

What Changed Between Google and Gemini Search Citations?

Our analysis showed a clear pattern: Gemini seems to trust official company sources more, while Google still pulls from a wider variety of voices across the web.

Content Category	Google Share (%)	Gemini Share (%)	Change (Gemini – Google)
Self (your site)	7.3%	15.3%	▲ +8.0 pts
Competitors	9.6%	22.9%	▲ +13.3 pts
Ecosystem Partners	34.0%	32.7%	▼ -1.3 pts
Discussion Forums	19.9%	3.4%	▼ -16.5 pts
Industry Publications	12.5%	8.0%	▼ -4.5 pts
User-Generated Blogs	10.1%	9.0%	▼ -1.1 pts
Review Sites	1.3%	2.7%	▲ +1.4 pts
News	0.2%	2.1%	▲ +1.9 pts
Videos	0.0%	1.6%	▲ +1.6 pts
Academic & Open Source	5.0%	2.0%	▼ -3.0 pts
Encyclopedic	0.1%	0.3%	▲ +0.2 pts

Here’s a breakdown of the four most important trends we saw.

1. Your Own Website Is More Important Than Ever Gemini cites a company's own website more than twice as often as Google does (jumping from 7.3% to 15.3%). This is a strong hint that the AI model considers your own website, blog, and documentation to be the most reliable source of information about your company and products.

2. Your Competitors Are Getting More Attention This was one of the biggest changes we saw. Gemini cites competitor websites 2.4 times more often than Google (from 9.6% to 22.9%). The AI is actively looking for content that compares you to others to frame its answers. It's no longer enough to rank for your own brand; you need to be part of the competitive conversation the AI is creating.

3. What Happened to Forums? Citations from discussion forums like Reddit and Stack Overflow dropped from nearly 20% on Google to just 3.4% on Gemini. This looks like a huge decline, but there’s another way to think about it. We know that AI models have been trained on the community content, so it’s safe to assume they have internalized that knowledge but are not citing it in the results. This means the value of community engagement is changing. The goal is less about getting a direct link back to your site and more about participating in the conversations that shape the AI’s understanding over the long run.

4. Your Partners Are Still Key Through all this change, one area stayed about the same: the ecosystem. Content from technology partners and integration marketplaces made up about a third of all citations on both platforms. This shows that having a well-documented product ecosystem is still crucial for getting discovered, no matter what the search page looks like.

How AI Citation Trends Differ by Industry: Developer Tools vs Horizontal SaaS vs Vertical FinTech

When we looked closer at our three different types of companies, we saw that these big trends affect them in different ways. You'll need to tailor your strategy to your specific market.

• For Vertical SaaS (Prodigal): In a regulated field like finance, Google tends to lean heavily on industry publications (39.1% of citations). Gemini does the opposite, pulling almost half of its sources directly from competitor websites (47.6%). It seems like Gemini is relying heavily on vendor websites as the source of truth rather than on industry publications.

  

• For Developer Tools (MotherDuck): Developer products have always relied on communities like Reddit and Hacker News. While Google still reflects that (19% of citations), Gemini drops that number to just 3.4%. Instead, it focuses on official documentation from the company itself and its partners. For content teams, this means a single, high-quality guide on your own website is now more valuable than dozens of scattered forum comments. As mentioned above, these forums are still important but may not serve the same purpose as before.

  

• For Horizontal SaaS (Mindtickle): These companies see a drop in citations from "thought leadership" blogs (from 15.0% down to 2.3%). That visibility has shifted to competitor websites and structured review sites like G2 and TrustRadius. It seems Gemini trusts aggregated reviews more than individual blog posts.

  

How to Increase Your AI Search Visibility: The Generative Engine Optimization (GEO) Playbook

So, what should you do with this information? A seasoned SEO professional would tell you that creating high-quality, trustworthy content is nothing new. It’s simply an evolution of principles like E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness), and they’d be right. What this data shows, however, is that the importance of different signals is being re-weighted dramatically, and the pace of change is picking up.

It's also important to acknowledge that executing this playbook requires a real investment in content, especially for creating technical documentation and multimodal content. We suggest taking it one step at a time, starting with the areas that will have the biggest impact on your specific market, whether that’s creating competitive comparisons or deepening your technical guides.

The playbook itself may seem familiar, but the need for quick, data-driven decisions is more urgent than ever. Keep in mind that this analysis only looks at Google's ecosystem; other AI engines like ChatGPT might behave differently. The core task is to experiment, measure, and adapt.

1. Own the conversation about your competitors.

• Action: Create detailed "Alternative to X" and "Product A vs. Product B" comparison pages. Don't be afraid to mention competitors by name.
• Tech Tip: Use Product, Review, and FAQ schema markup on these pages. This gives search models structured data, making it easier for them to understand features and pricing, which increases the odds that they'll use your content in an AI-generated answer.

2. Invest in your own technical content.

• Action: Put significant effort into your public-facing documentation. This means in-depth architectural guides, easy-to-follow tutorials, and clear instructions for your integrations.
• Why it works: This type of content establishes your company as a technical authority for AI models, especially when it’s structured as a topic cluster with pillar pages and subtopics, as outlined in our Topical Authority Playbook. It also directly feeds the Self and Ecosystem categories that both Google and Gemini value.

3. Change how you think about community forums.

• Action: Stay active in forums. Your goal is now twofold: continue building brand awareness and, just as importantly, contribute high-quality information that can help train future AI models.
• Example: When you answer a question on Reddit, give a complete and accurate response. Then, use that same question as inspiration for a more detailed, official guide on your own website. This way, you help train the model while also creating an asset that could earn a direct citation.

4. Use third-party review sites to your advantage.

• Action: Keep your company profiles on sites like G2, Capterra, and TrustRadius up to date. Encourage your customers to leave detailed reviews that mention specific features.
• Why it works: Gemini’s preference for these sites shows that it can use structured reviews to answer questions like, "Is Product A good for large companies?"

5. Get ready for more video content.

• Action: Create more videos, such as demos, webinars, and talks. Posting them on YouTube is crucial, but you should also include the full transcribed text on your website alongside the video.
• The opportunity: While the numbers are still small, Gemini is already citing video sources (1.6%), whereas traditional search never did. Creating and making video content accessible to the AI is a great way to get ahead of the curve.

Looking Ahead: What to Watch Out For

To wrap up, let's touch on a couple of bigger trends and risks that will shape the future of AI search.

• Negative SEO is changing. The fact that competitor content is cited so much more often creates a new risk. A rival could publish a misleading comparison page that, if structured well, Gemini might use as a source and present as fact. Creating your own fair and detailed comparisons is no longer just good marketing; it's a necessary defense.
• The data sources are in flux. The relationship between AI companies and publishers is changing quickly. Some sites, like Reddit, are making deals to license their content for AI training. Others are blocking AI crawlers to protect their content. These "data wars" will affect what information future AI models can learn from and cite. This uncertainty makes it even more important to build authority on the web properties you control.

Conclusion: It's All About Experimentation

The data from our snapshot in July 2025 shows that search is in a period of major transition. Right now, Google's AI models seem to prefer structured, official content from company websites, while the direct influence of community forums has faded. A future version of this study will examine how other AI search engines, such as ChatGPT and Perplexity, are behaving.

But if there's one thing to take away from this blog, it's that these are the rules for today, not forever. The only constant in this new era is change. The winning strategy generative engine optimization isn't to follow a rigid playbook, but to build a company culture of constant, data-driven experimentation. Success no longer comes from mastering one algorithm, but from adapting to a whole ecosystem of smart, fast-moving systems. The B2B teams that do well will be the ones who treat their content strategy like a living hypothesis, something they are always testing, measuring, and improving.