How AI Engines Choose and Cite Sources: A 7-Month Analysis

There's a model most AEO teams are still operating from. It goes something like this: track your citation share, improve your content, watch your numbers go up. All AI engines are treated as one neutral distribution channel. You optimize for it. It rewards you.

Seven months of citation data says otherwise.

Every AI engine in this analysis revealed a source preference. An editorial identityEntity
An entity is a thing/concept that search engines and AI models can identify and relate to other entities, forming the foundation of semantic search.
Learn More: a consistent pattern of which domains it trusts, for which intent types, that holds month after month. Wikipedia anchors one engine across four intents all year. YouTube dominates three others across nearly every category. One engine consistently routes users back to its own properties for Purchase queries. Same user. Same intent. Completely different source logic.

The part that breaks the model most: three of the engines we tracked are Google products. Google AI Overviews (AIO), AI ModeAI Mode
AI mode is a search feature using AI to provide comprehensive answers by synthesizing information from multiple sources into direct responses.
Learn More, and Gemini share infrastructure, a user base, and a parent company. They don’t share a source preference. Not for Education. Not for Purchase. Not for Support. They have developed distinct editorial identities, and those identities lead them to completely different places for the same query.

If your brand has one strategy for all AI engines, you're getting it right for some and leaving citations on the table for the rest.

This research shows you exactly where.

Understanding the seven intents of AI search

This analysis uses Conductor's Intent Taxonomy to categorize how AI engines respond to different user goals. The seven intent categories map to the universal stages of how we all use the internet today.

Methodology: What we analyzed

We tracked citations across ChatGPT, ChatGPT Search, Perplexity, Google AI Overviews, AI Mode, Gemini, and Claude. The dataset covers seven months from September 2025 through March 2026, across all seven Conductor intent categories. For each engine, each intent, and each month, we identified the top-cited domain and its share of total citations. That's 1,056 data points total. We looked for the patterns that held, and the ones that shifted.

Claude is covered separately in a callout below. Our dataset captured two months of Claude citation data, not seven. That's enough for early signals, not a verdict.

Every AI engine has a source preference

AEO conversations fixate on two numbers: citation volume and share of voice. Both are useful. Both are incomplete. They tell you how often you show up. They don't tell you why.

That why lives in the source preference layer. Which types of domains does each engine trust? Which content formats earn citations for which intents? Which engines are moving, and in what direction? Those answers compound. Citation volume is a lagging indicator of source preference decisions that already happened three months ago.

Four findings defined this research:

• ChatGPT and ChatGPT Search are the only Wikipedia-anchored engines in the dataset.
• Google AI Mode consistently routes users to Google property, aka Google Shopping for Purchase queries. No other engine did this in our study window.
• Google AI Overviews, Perplexity, and Gemini all lead with YouTube across nearly every intent.
• ChatGPT and ChatGPT Search are separate products with meaningfully different citation behavior. They are not the same engine.

Claude was tracked in this dataset but is not included in the engine breakdowns below. At the time of this analysis, Conductor's dataset only captured two months of Claude citation data. That allows us to detect early-stage signals but not a stable pattern.

ChatGPT and ChatGPT Search are the only engines that surface Wikipedia

These two AI engines share a name and a company. They don't share a source preference, and treating them as the same product will cost you.

Before getting into the citation data, it's worth being clear on what each one actually is:

• ChatGPT is a conversational platform. It answers by drawing on the model's training knowledge and the context you provide. Unless you've specifically turned on browsing, it's not necessarily checking the live web.
• ChatGPT Search is ChatGPT with live web retrieval layered in. It goes out to the web first, identifies sources, extracts relevant information, and then uses the model to turn that evidence into an answer. The difference isn't just one has internet access and one doesn’t; it's the retrieval step. ChatGPT Search actively looks for sources to cite, not synthesizing from what it already knows.

Now here's what our data shows.

ChatGPT Search prefers Wikipedia across four intents: Education, Recommendations, Comparison, and Purchase. That preference didn't shift across seven months. For Education queries, Wikipedia was the top-cited domain every single month. The share barely moved. Recommendations queries, same story.

That's the cleanest editorial identity in the entire dataset. ChatGPT Search wants citation-grade prose: structured summaries, named entities, and content that reads like a reference section rather than a marketing page. If your content is narrative-first, it likely won't surface.

ChatGPT is less stable. Wikipedia and Reddit trade the top spot depending on intent type. Wikipedia leads on Recommendations queries every month. Reddit leads on Support in five of seven months. The editorial identity is real, but it splits by intent rather than source types.

For marketers, that distinction matters. Optimizing for ChatGPT Search is closer to optimizing for retrievability and source authority. Optimizing for ChatGPT is broader: brand recognition, topical authorityTopical Authority
Topical authority is the expertise and credibility a website demonstrates on a subject through comprehensive, interconnected, high-quality content.
Learn More, and how your brand is represented across the model's training data.

Two engines. Two optimization strategies. Know which one your audience is actually using.

Perplexity prefers YouTube across most intents

Perplexity runs on a different citation logic entirely. YouTube anchored its top citation slot for Education and Recommendations queries every month during our research period.

Same intents where ChatGPT Search points to Wikipedia. Perplexity points to YouTube.

One engine rewards long-form reference prose. The other rewards video. If you're only producing written content for Education and Recommendations intent, you're winning citations from one of these engines and leaving the other untouched.

Google AIO is the most predictable Google engine we tracked

Google AI Overviews prefers YouTube for five of seven intents. It held that preference consistently across the entire seven-month window. For Purchase queries, YouTube was the top-cited domain every single month.

The one exception is the Navigational intent. There, brand-owned domains (a brand's own website, as opposed to a third-party source like Wikipedia or YouTube) took the top citation slot for most of our study period.

That exception matters. Navigational queries are where users are actively looking for a specific brand or product. For every other intent, AI Overviews leads with YouTube. Your video content strategy directly affects your citation performance here, across the majority of the customer journey.

Google AI Mode routes users back to Google—no other engine does that

AI Mode is the most interesting engine in this dataset, and the one that requires the most strategic attention for brands working in Google's ecosystem.

AIO and AI Mode are both Google products. They handle the same Purchase intent in completely opposite ways.

AIO anchored on YouTube for Purchase queries every month for seven months. AI Mode never put YouTube in the top slot for Purchase. It consistently cited Google properties instead, routing users back into Google's own product ecosystem (think Google ShoppingGoogle Shopping
Google Shopping is Google’s product-based advertising service.
Learn More, Google MapsGoogle Maps
Google Maps is one of, if not the biggest, readily accessible and free online map services in the world.
Learn More, and Google's own blog) rather than outward to third-party sources.

The divergence for Education queries is equally sharp. AI Mode shifted to institutional sources for four months (November 2025 through February 2026) while AIO stayed on YouTube the entire time. Both changed course in March 2026. AI Mode reverted to YouTube. AIO stayed consistent.

AI Mode is also the only engine in this dataset that cites LinkedIn for Education intent. None of the other five AI engines did this at any point during our analysis.

If your audience reaches you through Google's AI engines, you are not optimizing for one engine. You are optimizing for two with different source instincts, and a single content strategy won't satisfy both.

Google Gemini cites YouTube across every single intent

Gemini is YouTube-anchored across nearly all intents, with one notable pattern for Navigational queries. Brand-owned domains and Media sources rotate in as the second and third most-cited sources for Navigational in three of seven months. It's the only intent where a category other than YouTube consistently challenges for the top position.

For Support queries, YouTube held the top citation slot every month across the full seven-month window. Gemini is the most consistent YouTube-first engine in the dataset.

AI engine editorial identities summarized (September 2025 – March 2026)

What this data means for your content strategy

The aggregate picture across all AI engines is useful, but it's not a complete content strategy.

Good foundational AEO practices lift visibility across every AI engine. Structured content, authoritative sourcing, and clear intent alignment get you in the door on all of them. Those are the entry requirements. What separates strong AEO performance from average is what comes next: matching the specific editorial identity of the engine your audience actually uses.

Our analysis shows clearly that individual AI engines rarely share source preferences, even those within the same mother company (think ChatGPT vs. ChatGPT Search, Google and its three engines). If you optimize for what all AI engines have in common, you end up with content that sort of works everywhere but excels nowhere.

Here’s an example: YouTube dominates Education queries in the aggregate because three engines favor it. But if your audience mainly uses ChatGPT Search, YouTube won't move your numbers. Wikipedia will. And if your audience uses Google more often, you're optimizing for three engines with three separate editorial identities. Use the aggregate as a benchmark. Then go one level deeper.

1. Optimize per engine, not per intent category

Treat AI engines as distinct channels with different editorial identities. Structured dataStructured Data
Structured data is the term used to describe schema markup on websites. With the help of this code, search engines can understand the content of URLs more easily, resulting in enhanced results in the search engine results page known as rich results. Typical examples of this are ratings, events and much more. The Conductor glossary below contains everything you need to know about structured data.
Learn More and clear intent alignment get you in the door on all of them. But the ceiling is engine-specific.

Google AIO and AI Mode handle Purchase intent in completely opposite ways. One anchors on YouTube. The other routes users to Google's own properties. A single-category strategy can't bridge that gap.

2. Match the editorial identity before scaling content volume

AI engines cite sources that look like the sources they already trust. Before scaling content production, audit the top-cited domains in your priority engine for your target intent. Is it rewarding reference-grade prose, video content, or community discussion?

If the AI engine wants encyclopedic structure and you're publishing narrative marketing pieces, domain authority won't compensate. Volume without an identity match is expensive and ineffective.

3. Watch the swings, not just the share

Citation share is a lagging indicator. The actionable signals are in the pivots.

When ChatGPT Search shifted Support citations from Reddit to brand domains, that wasn't noise in the data. It was the engine redefining what quality Support content looks like. Google AI Mode's four-month shift toward institutional sources for Education intent is the same kind of signal.

Those pivots are your intervention windows. Spot them early and you're adjusting your strategy before the shift shows up in your competitors' share numbers.

4. Diversify content across formats, not just topics

Different engines trust different source types. ChatGPT Search rewards encyclopedic prose. Perplexity and AI Overviews reward video. Claude (based on our two-month observation) rewards compliance-grade institutional content.

If your content strategy primarily produces one content format (like long-form, written content), you're limiting citation gains to only some of the engines your audience uses and leaving the rest uncontested.

Topic coverage isn't enough anymore. Format coverage is what gets you into the citation set across the full AI engine landscape.

AI engines aren't neutral. They never were.

Google AIO, AI Mode, Gemini, ChatGPT, ChatGPT Search, Perplexity, and Claude aren't neutral delivery mechanisms. They're curated platforms with distinct, often conflicting editorial identities. Our seven months of data confirms it. These citation preferences aren't random. They're persistent behavioral patterns that redefine what quality means for every intent and engine.

Citation volume tells you how often you show up. Source preference tells you why you're showing up and what it takes to show up more.

Showing up isn't the goal anymore. Being the most citation-worthy source for your intent, on the AI engines your audience actually uses is.