10 Lessons From Tracking 50,000 AI Citations Across 6 Engines

For 90 days I tracked how six major AI search engines cite sources when they answer B2B software queries. The engines were ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, and Microsoft Copilot. The dataset covers more than 50,000 AI-generated responses across cybersecurity, identity, developer tools, and marketing SaaS.

The analysis ran on GrackerAI, which monitors AI visibility across ten engines and reports citation frequency and share of voice per engine rather than blending everything into one number. That per-engine breakdown turned out to matter more than any other methodological choice, because the single biggest finding is that treating AI search as one channel is the core mistake most companies are making.

Here is what the data showed, and where the current industry playbook gets it wrong.

---

Lesson 1: Each Engine Has Its Own Citation Personality

The biggest surprise was how differently each engine behaves. Perplexity cites aggressively, often pulling 8 to 12 sources per response with inline links. ChatGPT is selective, typically citing 2 to 4 sources and favoring well-known brands. Claude rarely provides direct URLs but references brands by name with high contextual accuracy. Gemini leans heavily on Google's index and tends to cite sources that already rank well in traditional search.

The practical implication is that a single "AI visibility score" is misleading. A brand can have 60% visibility on Perplexity and 5% on ChatGPT for the same query. If you are optimizing for "AI search" as one channel, you are optimizing for nothing in particular. The rest of these lessons keep circling back to this point, because it is the one that reorders everything else.

Lesson 2: Listicles Get Cited 3 to 4x More Than Thought Leadership for "Best X" Queries

When someone asks "best SIEM tools for mid-market" or "top passwordless auth platforms," every engine overwhelmingly pulls from listicle-format content. Articles titled "7 Best X for Y (2026)" get cited at roughly 3 to 4 times the rate of in-depth analysis pieces for these query patterns.

But this only holds for "best" and "top" queries. For "how does X work" or "X vs Y" queries, listicles almost never appear. The content format has to match the query intent, not just follow the SEO playbook. Producing brilliant thought leadership and expecting it to win "best tool" queries is a category error, since those queries want a structured list, and the engines reward exactly that.

Lesson 3: Freshness Is a Harder Signal Than Most People Realize

Content published or updated within the last six months gets cited significantly more than older content, even when the older piece is more thorough. I watched a 2023 "definitive guide" to zero trust lose citations to a thinner 2025 article that had been recently updated.

The year in the title matters too. Articles with "(2026)" in the title showed higher citation rates in 2026 responses than identical content without a year marker. The engines appear to use the title date as a freshness proxy. The takeaway is uncomfortable for anyone sitting on a great evergreen asset: depth does not protect you from decay. Recency is doing real work in the citation decision, and a lightly updated recent piece can beat a superior older one.

Lesson 4: Your Competitors' Content Strategy Is Visible If You Look

One of the most actionable findings is that you can reverse-engineer why a competitor gets cited and you do not, simply by examining the exact prompts where they appear. In the identity and auth category, I tracked prompts like "best SSO solution for B2B SaaS" and found that vendors with dedicated comparison pages ("X vs Okta," "X vs Auth0") appeared in AI responses at nearly double the rate of vendors without them.

This is not speculation. Looking at the specific prompts and responses where competitors show up, and which sources the engine pulled from, makes the gap concrete. The gap between "we think we are competitive" and "the data shows we are invisible" was stark for most brands I tracked. Comparison pages, in particular, are a repeatable and underused lever: buyers ask comparative questions constantly, and the vendors who publish the comparison are the ones who get cited in the answer.

Lesson 5: Domain Authority Still Matters, but Differently

In traditional SEO, domain authority is a ranking factor. In AI search, it functions more like a trust filter. The engines do not appear to rank sources by authority, but they do seem to apply a credibility threshold. Content from established domains with existing backlink profiles and review-site presence gets cited more reliably than identical content on new or low-authority domains.

The workaround is to earn the trust signal from third parties rather than trying to manufacture it on your own domain. A mention in a G2 comparison, a reference in analyst coverage, or a link from a well-regarded industry blog creates the credibility signal the engines rely on. This is why authority-building and AI visibility are connected: the same earned coverage that has always signaled credibility now feeds the citation decision directly.

Lesson 6: Structured Content Earns More Citations Than Prose

The engines extract information more easily from structured content: clear H2 and H3 headings, comparison tables, numbered lists, FAQ sections with direct answers, and definition blocks. A 3,000-word essay with no subheadings gets cited less than a 1,500-word article with clear structure covering the same information.

This makes sense mechanically. Retrieval-augmented generation systems chunk content by sections. If your content has clear section boundaries with descriptive headings, the retrieval step pulls cleaner chunks, which leads to more accurate citations. The lesson is not "write less," it is "structure everything." A long piece with strong internal structure does fine. A long piece that is one undifferentiated wall of prose gives the retrieval step nothing clean to grab.

Lesson 7: Your "About" and "Company" Pages Get Cited More Than You Expect

A surprising finding: company overview pages, "about" pages, and product landing pages with clear positioning statements get cited for brand-related queries. When someone asks "what does [Brand] do?" or "tell me about [Brand]," the engines pull heavily from these pages.

Companies that treat their about page as an afterthought are losing citations on every branded query. A clear one-liner, a structured product description, and explicit category positioning on your homepage and about page directly shape how the engines describe you. This is a low-effort, high-return fix that most companies overlook because they assume nobody reads the about page. The engines read it, and they repeat what it says.

Lesson 8: Perplexity and ChatGPT Cite Completely Different Sources for the Same Query

I ran identical prompts through all six engines at once. For "best passwordless authentication platform," Perplexity cited developer docs, GitHub READMEs, and technical blog posts. ChatGPT cited marketing landing pages, review sites, and well-known brand names. Claude referenced the brands by name without URLs but demonstrated awareness of technical differentiation.

The implication reinforces Lesson 1: you cannot optimize for AI search as a monolith. Your developer docs matter for Perplexity. Your marketing pages matter for ChatGPT. Your brand's overall web presence matters for Claude. A per-engine strategy is the only strategy that works, because the same query pointed at two engines pulls from two entirely different parts of your web presence.

Lesson 9: Reddit and Community Mentions Are an Underrated Citation Source

The engines, especially Perplexity and ChatGPT, pull from Reddit threads, community forums, and discussion platforms more than most marketers realize. A genuine Reddit thread comparing several auth providers with real user opinions gets cited for comparison queries at a rate comparable to published blog content.

This means your community presence, the earned mentions in subreddits, on Hacker News, and in developer forum threads, is part of your AI visibility whether you manage it or not. You cannot fabricate this credibly, and you should not try, but you can participate genuinely in the communities where your buyers already talk, and you can make sure the honest picture of your product is present where those conversations happen.

Lesson 10: AI Visibility Compounds Like SEO, but Faster

The most important finding is that AI visibility improvements compound. Once an engine starts citing your brand for a particular query cluster, it tends to keep citing you as long as the underlying content stays fresh and relevant. Early movers in a query category build a citation advantage that is difficult for latecomers to displace.

I watched one cybersecurity vendor go from 7% visibility to 47% in 90 days by systematically addressing the content gaps and prompt clusters where competitors appeared and they did not. The method was simple: identify the prompts where you are invisible, create or improve content targeting those prompts, and monitor until citations appear. That loop, monitor, find the gaps, close them, and track, is the whole game.

The window for establishing AI citation authority in most B2B categories is still open, but it is closing. Companies that build their AI visibility foundation now will hold a structural advantage for years, because the compounding works against whoever arrives late.

---

What the Data Adds Up To

Step back from the ten lessons and a single argument emerges. AI search is not one channel with one playbook. It is many engines with distinct citation behaviors, rewarding different content formats for different query intents, filtered by authority and freshness, and pulling from parts of your web presence you may not even think of as marketing. The companies winning at this are not the ones with the highest single visibility score. They are the ones running a per-engine, per-query-intent strategy and closing the specific gaps the data reveals.

That is also why measurement has to come first. You cannot close gaps you cannot see, and a blended score hides exactly the per-engine, per-prompt detail that tells you what to fix. The vendor that went from 7% to 47% did not guess. It measured where it was invisible and acted on it.

If you want to see how the AI engines currently perceive your brand, GrackerAI offers a free AI visibility score. But the deeper point stands regardless of tooling: the brands that treat AI visibility as a measurable, per-engine discipline, rather than a single number to chase, are the ones building citation authority that lasts.

---

Methodology

• Timeframe: 90 days (March to May 2026)
• Engines tracked: ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Microsoft Copilot
• Total responses analyzed: 50,000+
• Categories: cybersecurity, identity and auth, developer tools, marketing SaaS
• Prompts tracked: 500+ unique prompts across categories
• Tooling: GrackerAI for visibility monitoring, citation tracking, and competitive analysis

---

Related reading:

• Why GEO Has to Be Vertical, why per-category citation behavior demands a vertical approach
• Documentation Is Your Most Underrated GEO Asset, why developer docs drive Perplexity citations
• GEO Is a Product Discipline, Not a Marketing Tactic, why about pages and product positioning are citation surfaces
• AI Search Is Becoming the Default for B2B Buyers, the buyer shift behind why citations matter
• GEO Compass, vendor-neutral resource on engines, signals, and measurement

Deepak Gupta is a serial entrepreneur and cybersecurity expert who co-founded and scaled a CIAM platform to serve over 1 billion users globally. He leads GrackerAI, a GEO platform built specifically for B2B SaaS and cybersecurity companies to achieve visibility in LLM search engines like ChatGPT, Perplexity, and Google AI Overviews. He writes about AI, cybersecurity, and B2B growth at guptadeepak.com.