There's No Single 'AI SEO': ChatGPT, Perplexity and Google Cite Completely Different Sources
For the past two years, marketers have chased a single prize called "AI SEO" — one set of tactics that would get them cited by ChatGPT, Google AI Overviews and Perplexity all at once. A study published in July 2026 says that prize does not exist. Analysing citations across the major answer engines, CiteLens found that 91% of cited URLs appear in only one AI engine, and just 2% show up across all three. The engines are reading the web through completely different lenses, and optimising for one barely moves the others.
That finding, corroborated by parallel research from Profound and several independent audits this month, reframes the whole game. There is no unified "AI search" to rank in — there are separate ecosystems with separate rules. Below we unpack exactly what the data shows, why the engines diverge so sharply, and how to build a strategy that wins citations on more than one of them.
What the CiteLens study actually found
CiteLens ran the same prompts across Google AI Mode, Perplexity and ChatGPT, then compared which domains each engine chose to cite. Instead of the overlap you'd expect if a single "AI SEO" existed, the citation sets were almost disjoint.
| Overlap measure | What the data shows |
|---|---|
| Cited URLs appearing on all three engines | ≈ 2% |
| Cited URLs appearing on only one engine | ≈ 91% |
| Domains cited by both ChatGPT and Perplexity for the same query | ≈ 11% |
| Correlation of citations with Google ranking — Google AI Mode | ≈ 0.92 |
| Correlation of citations with Google ranking — Perplexity | ≈ 0.87 |
| Correlation of citations with Google ranking — ChatGPT | ≈ 0 (near zero) |
Read the last three rows together and the split becomes obvious. For Google AI Mode and Perplexity, citation frequency tracks almost perfectly with how highly Google ranks a domain — a correlation of 0.92 and 0.87 respectively. These are, in effect, search-ranking machines: win classic SEO and you win their citations. ChatGPT sits in a different universe. Its citations correlate with neither Google ranking nor brand size, and only about 21% of the sites it cited were Wikipedia-backed, versus a much higher share for the others.
The mechanism behind ChatGPT's independence is its retrieval behaviour. CiteLens found ChatGPT generates 91% unique fan-out queries — it rarely searches the web the same way twice, even for the same prompt. Where Google AI Mode and Perplexity lean on a stable ranking signal, ChatGPT improvises a fresh set of sub-searches each time, which is why its source list looks so different from the SEO-driven engines.
Why the engines diverge so sharply
Three structural differences explain why one optimisation playbook can't cover all of them.
1. Different retrieval backbones
Perplexity and Google AI Mode are built on top of a search index and treat the ranked results as their candidate pool. That is why their citations correlate so tightly with Google ranking — the ranking is the shortlist. ChatGPT, by contrast, blends its own web tool with model priors and a wide fan-out, so a page that never ranks can still be pulled in if it answers a sub-query cleanly. Same question, three different ways of deciding who is allowed in the room.
2. Different trust signals
The engines weigh authority differently. The SEO-driven engines lean on the signals Google already rewards — links, rankings, established entities — which is why big, Wikipedia-backed domains dominate their citations. ChatGPT's near-zero correlation with brand size means a small, specific, well-structured page can be cited on merit. This is the same passage-over-page dynamic we covered in our guide to optimizing for AI search: what gets quoted is often a single clear paragraph, not a domain's reputation.
3. Different query expansion
All the engines expand a user's prompt into sub-queries — the "fan-out" mechanic — but they do it with different breadth and stability. We explained the core idea in what query fan-out is and how to optimize for it. The new wrinkle from CiteLens is how unstable ChatGPT's fan-out is: 91% unique queries means the source pool reshuffles constantly, so a page cited today may be swapped for a rival tomorrow. Google AI Mode's fan-out, tied to a ranking index, is far more repeatable.
Two ecosystems, not one
The practical takeaway is that AI visibility now splits into two broad camps, and you should think about them separately rather than as one "AI SEO" target.
| Search-ranking engines (Google AI Mode, Perplexity) | Model-native engine (ChatGPT) | |
|---|---|---|
| What decides citation | Classic Google ranking (corr. 0.87–0.92) | Passage fit + fresh fan-out (corr. ≈ 0) |
| Who tends to win | High-ranking, established, Wikipedia-backed domains | Specific, well-structured pages regardless of size |
| How stable are the sources | Relatively stable, tied to the index | Volatile — 91% unique queries per prompt |
| Primary lever | Traditional SEO: rank, links, authority | Clear extractable answers, entity clarity, structured data |
| How to measure | Rankings + manual citation checks | Repeated manual prompt checks; watch AI referral traffic |
Because 92% of standalone AI referral traffic now comes from ChatGPT — a figure we broke down in our look at AI referral traffic in 2026 — the model-native engine is not a niche you can ignore. Yet it is precisely the one that traditional SEO barely touches. That tension is the whole story of this study.
What to do about it
You cannot game three engines with one trick, but you can build a foundation that earns citations across all of them, then add engine-specific work on top.
1. Keep doing real SEO — it wins two of the three
With citation correlations of 0.87 and 0.92, ranking well in Google directly buys you visibility in both Perplexity and Google AI Mode. That is two of the three major engines unlocked by the fundamentals you already know. If anything, the study is a reason to double down on classic SEO, not abandon it. Our primer on improving your SEO rankings covers the groundwork.
2. Write liftable, self-contained answers for ChatGPT
Because ChatGPT ignores rankings and rewards passage fit, structure each section to be quotable out of context: lead with a one- or two-sentence answer, use question-style H2s, keep paragraphs short, and put the key fact where a model can extract it without reading the whole page. This is the work that helps you on the engine SEO can't reach.
3. Strengthen your entity, not just your pages
Consistent brand mentions, clear authorship and coverage on reputable third-party sites help a model feel safe quoting you when it can't lean on Google's ranking as a proxy for trust. Structured data matters here too — schema markup has been linked to markedly higher AI selection rates because it hands the engine machine-readable proof of who you are and what a page is about.
4. Measure per engine, because one dashboard won't do
If 91% of citations are engine-specific, a single "AI visibility" score hides more than it reveals. Track ChatGPT, Perplexity and Google AI Mode separately — run the same priority prompts in each, log who gets cited, and watch AI referral visits in analytics. What earns you a citation in one tells you little about the others.
The mistakes to avoid
- Chasing a mythical unified "AI SEO." With 91% of citations appearing on a single engine, there is no one setting that wins them all. Plan per ecosystem.
- Abandoning SEO because "AI changed everything." SEO still drives citations on the two engines that correlate 0.87–0.92 with Google ranking. Cutting it loses you most of the AI-answer surface.
- Ignoring ChatGPT because SEO doesn't move it. It is the source of ~92% of AI referral traffic. Skipping it means skipping the biggest AI channel.
- Reporting a single AI visibility number. An average across engines that barely overlap is close to meaningless. Break it out.
The takeaway
The CiteLens study puts a number on something practitioners have suspected all year: "AI SEO" is not one thing. Google AI Mode and Perplexity are search-ranking engines you win with classic SEO; ChatGPT is a model-native engine that shrugs off rankings and rewards clear, extractable, entity-backed content. With only 2% of citations shared across all three and 91% unique to one, the winning move is not to find a single hack but to build a strong SEO and content foundation — then layer engine-specific optimisation on top and measure each surface on its own terms.
