AI Search Isn't One Channel. It's Three.

The Number That Should Reset Your Budget

Zero pages were cited by all three engines, out of 6,083. ChatGPT and Google AI Overview did not share a single citation across the 500 identical queries; their source lists never touch. The only overlap anywhere in the study sits between Google AI Overview and Perplexity, and it covers 352 URLs. Six percent of the total.

What this means for the "one channel" slide

If two of these engines never recommend the same page, they are not one channel. The work your team does to win a citation in one of them does close to nothing in the other two. You are not buying AI visibility. You are buying visibility inside one specific engine, and you should know which one before you sign a proposal.

They Don't Even Show Up for the Same Questions

Zero citation overlap is half the finding. The other half comes one step earlier, before any page gets chosen.

Each engine first decides whether a question deserves cited sources at all; some questions it answers straight from memory, with nothing attached. The three engines make that call in completely different ways. Here is how often each returned citations, by query type:

The three engines behave like three separate products.

• Perplexity cites something for almost everything. Between 99 and 100 of every 100 queries drew at least one citation, in every category we tested. It is the only engine where any intent type is a dependable citation opportunity, and the only one answering local questions with sources at all.
• Google AI Overview is an informational specialist. It returned citations for 90 to 96 of every 100 informational and health-and-money queries, then fell away sharply: 32 of 100 transactional queries, 10 of 100 local ones. For "dentist near me," Google trusts its Map Pack over an AI answer, so the AI answer stays home.
• ChatGPT is the outlier. Across the whole study it returned zero citations for informational questions and zero for YMYL, answering both from its own training with no sources shown. It reached the live web only on commercial, transactional, and local queries, and even there on a fifth to a third of them. Publish health or educational content and the read is blunt: there is no ChatGPT citation to win in your category right now. Your entire ChatGPT play is being in the model's memory and its direct answers, not being cited.

So "get cited by AI" has no single answer for your business. It resolves to a different engine, and a different kind of page, depending on what your buyers are asking.

Independent work is landing on the same structure. Ayomide Joseph's Query Fan-Out Experiment (May 2026) ran 270 "best alternative to [product]" queries across ChatGPT, Gemini, and Perplexity: Perplexity searched the web on every query, ChatGPT on under half, and each engine pulled from a visibly different source mix. His method differs from ours, and his set swaps Gemini for Google AI Overview, so read it as corroboration of the shape rather than a replication of our numbers.

What This Changes, One Decision at a Time

The table matters less than what it lets you stop paying for.

If you run a local business (clinics, restaurants, home services, retail), Perplexity is the only engine reliably citing sources for the questions your customers ask, at 100 of every 100 local queries. Google AI Overview reaches 10 of 100. An "AI visibility" budget pointed at Google for local intent buys you coverage of one question in ten.

If you sell products or run comparisons (e-commerce, SaaS, affiliate), commercial and transactional intent is the one place all three engines are awake at once. It is also where their selection rules diverge the most, which is exactly why a single budget genuinely splits three ways here. Our companion report on how each AI engine treats schema markup shows how differently: ChatGPT's commercial picks carry structured data at close to 100%, against roughly half for Google's.

If you publish informational or YMYL content (health systems, finance brands, education), your contest is Google AI Overview and Perplexity, and only those two. ChatGPT is not citing this territory. The upside is that both remaining engines return citations for 90 to 99 of every 100 questions here, so the opportunity is larger than in any other intent.

Before you fund any of it, run one check on your own numbers

Ask whoever owns your analytics to pull two things: referral traffic by source for the last 90 days, filtered for chatgpt.com, perplexity.ai, and Google AI referrals; and your Search Console pages, to see which of your own URLs already surface in Google's AI answers. One engine is almost certainly already sending you buyers, or already citing your pages, more than the other two. Fund that engine first. Deprioritize the ones your own data says are quiet for your category. The study tells you which engines are reachable by intent. Your referral log tells you which one is already working for you. Start where those two agree.

If you're the one holding the whole budget, one question separates a real AEO plan from an expensive one: which engine, for which query types? A proposal that can't answer it is optimizing for a channel that doesn't exist.

The rest of the study takes each of these decisions deeper: the pages that win both Google rankings and AI citations (and the outbound-linking habit that separates them), how schema requirements differ engine by engine, and why no page in the study won more than 8 queries, which retires the pillar-page playbook.

Three Channels Doesn't Mean Triple the Budget. But…

The wrong takeaway from this study is "AI visibility now costs three times as much." The honest reading runs the other way. Because the engines specialize by intent, most businesses can put two of them down entirely. A hospital system can ignore ChatGPT citations today. A restaurant group can nearly ignore Google AI Overview. Focus costs less than coverage.

The split is also not a passing glitch. Each engine picks winners from different inputs: Google AI Overview inherits two decades of ranking signals, Perplexity runs its own retrieval, ChatGPT leans on what a page declares about itself. The exact percentages will drift quarter to quarter. The three-way split is the part that holds.

That is also why per-engine tracking earns its keep. Your citation rate per engine is how you run the self-check above and catch the next shift early, while a single blended "AI visibility score" hides the one distinction that decides where the money goes: which engine, for which intent.

Plan against that structure. Then check it against your own referral data before the next budget cycle closes.

Methodology

Timeline. All engine responses were collected in April 2026, in a single automated run per query per engine, under identical conditions. The study was first published on 10 July 2026. The percentages describe that April 2026 window: AI engines ship retrieval and citation changes continuously, so treat the exact figures as a dated snapshot and the three-way structural split as the durable finding.

Dataset. 500 queries, 100 in each of five intent categories (informational, commercial, transactional, YMYL, local). US market, English. For each query we captured Google's organic top 10, Google AI Overview citations, ChatGPT citations (web search enabled), and Perplexity citations via the DataForSEO APIs. Sources deduplicated to 6,083 unique AI-cited URLs (8,397 including organic-only URLs).

Overlap. A URL counts as cited by an engine if that engine cited it for at least one query in the study. Overlap is computed on unique URLs across all 500 queries.

Coverage. For each category, the share of its 100 queries where the engine returned at least one citation.

Limitations. Five apply, and they bound what this study can and cannot tell you:

1. Single point in time. One collection window, April 2026. Engines update retrieval behavior frequently; the exact percentages will drift between snapshots even where the structure holds.
2. Single run per query. Each query was executed once per engine. AI responses vary between runs, so per-query results carry sampling noise that aggregate figures across 500 queries smooth but do not eliminate.
3. API conditions, not interactive sessions. ChatGPT's citation behavior is inconsistent upstream: the DataForSEO integration honors web search intermittently, so our numbers describe what ChatGPT cited under identical automated conditions. Interactive ChatGPT sessions, with user history and different toggles, may behave differently.
4. One market, one language. US market, English only. Citation behavior in other markets and languages, where engines lean on different indexes and local sources, is not covered.
5. Citations only. We counted linked citations, not unlinked brand mentions inside answer text. A brand can be named without being cited; that influence is real and this dataset does not measure it. The intent balance (100 queries per category) also cannot represent every industry equally.

FAQ: Which AI Engine Cites What