A new study says ranking on Google buys your client AI citations in Perplexity and AI Mode — and almost nothing in ChatGPT

What the study measured

The primary source is CiteLens's study announcement, dated July 8, 2026 and independently written up by MarTech Series. The method is the part worth copying: 320 templated buyer queries, run through four answer engines — ChatGPT, Perplexity, Claude and Google AI Mode — with every cited source then compared, query by query, against the same query's Google and Bing organic results. In other words, it does not ask "who got cited"; it asks "did getting cited track with ranking on Google." Declare the interest up front, as we do with every vendor study: CiteLens sells a GEO citation-tracking product, and so do we. Read the measured numbers; discount the conclusions.

The finding: two families of engine

The four engines split cleanly into two camps on whether Google ranking predicts citation:

Two details sharpen the ChatGPT column. Fewer than 4% of ChatGPT's citations appeared in Bing's top 10 — so despite the common assumption that ChatGPT leans on Microsoft's index, its citations do not track Bing rankings either. And only 21% of ChatGPT's citations went to Wikipedia-backed sites, versus 58% for Claude — the two "model-native" engines are not even similar to each other. So the honest one-line summary is not "SEO is dead" and not "SEO is everything." It is: on the two search-grounded engines, ranking is most of the game; on the two model-native engines, it is a minority input, and being a known entity matters more.

The caveat that keeps this honest

One limitation decides how far you can carry these numbers, and CiteLens states it: the corpus is the Turkish market, across three consumer sectors, in June 2026. That is a real constraint. The exact figures — 0.92, 0.89, 30% — are Turkish-language consumer queries, not US or UK local-service queries, and you should not quote them to a client as if they were measured in their market. What travels is the structure, not the decimals: the engines partition into search-grounded (AI Mode, Perplexity) and model-native (ChatGPT, Claude), and that partition is an architecture fact, not a language fact. It also lines up with what we see from the outside — the engines disagree sharply about who to name even on identical questions, which is why in our own 1,120-answer comparison ChatGPT named the audited business in 16% of answers while Claude named it in 38%. Different engines, different logic, different answers.

What an agency should do with this

• Stop selling one "AEO playbook" that promises all four engines at once. Ranking work that earns a client citations in Perplexity and Google AI Mode may barely move ChatGPT, where 70% of citations came from outside Google's top 10. A proposal that treats the four engines as one lever is priced on a fiction. Scope the search-grounded engines (AI Mode, Perplexity) as an SEO problem and the model-native engines (ChatGPT, Claude) as an entity/brand-presence problem — they are genuinely different work.
• Keep doing the SEO — it is necessary, just not sufficient. The takeaway is emphatically not "skip SEO." On the two search-grounded engines it is close to the whole game, and those two include Google's own AI Mode. SEO is the entry ticket to half the answer surface; it is simply not the ticket to the other half.
• For ChatGPT and Claude, the lever is being a recognized entity, not a ranked page. Consistent name, consistent facts, corroboration across the open web, third-party mentions — the things that make a model "know" a business rather than retrieve a URL. This is slower and less checklist-able than on-page SEO, which is exactly why it is under-sold and under-delivered.
• Measure per engine, because a single check averages the finding away. If SEO moves two engines and not the other two, then a client's AI visibility is not one number — it is four, and they can move in opposite directions. Report them separately or you will tell a client "you're doing well in AI search" on the strength of the two engines your SEO happened to move, while they are invisible in the one their customers actually use.

The takeaway

The useful thing about matching citations to rankings query-by-query is that it converts a slogan — "AEO is just SEO," or its opposite — into a measurement, and the measurement says both slogans are half-wrong. Ranking buys you citations in the search-grounded engines and little in the model-native ones. Which means the first honest deliverable for a client is still not a strategy; it is a per-engine measurement of where they actually stand, so you know which of the two problems you are being paid to solve.

You can run the free 60-second check on any business and see its mention rate across ChatGPT and Perplexity — one search-grounded engine and one model-native one, which is the split this study is about (the $79 audit adds Gemini and Claude, 25 questions sampled three times each, because one run is a sample, not a score). If you do this for clients, the $249 agency 5-pack runs the full repeated-sample audit across five of them, white-labeled — the per-engine breakdown you need before you promise a client an engine your ranking work cannot reach. And if you are choosing a tool to monitor this over time, here is how the AI-visibility trackers compare on which engines they actually cover.