The Same 10 Questions, Three AI Engines, Three Different Answers
Everyone tells you to optimise for “AI search”, as if AI search were one thing. We put ten identical questions about Google algorithm recovery through Google’s AI Overviews, Perplexity and ChatGPT on the same day and recorded every source each one cited. They barely agreed on anything.
Do different AI engines cite the same sources? Largely no. Across 10 identical questions about Google algorithm recovery, Google AI Overviews cited 50 unique domains and Perplexity cited 63 — but only 18 domains appeared in both. Roughly two-thirds of each engine’s citation set was unique to that engine. ChatGPT, asked the same questions, cited almost nothing at all: it answered from its own training rather than the live web, surfacing a single source in both tests. There is no such thing as “ranking in AI search”. There are at least three different competitions, with different winners.
What we tested
Ten queries, every one taken from real Google Search Console data for this site — questions US searchers actually used to reach us, not keywords we invented. Each was run through all three engines on 21 August 2026, and every cited domain was recorded.
How much do the engines actually overlap?
About a third. That is the headline.
Eighteen domains, out of fifty and sixty-three. If your AEO strategy is built around one engine’s behaviour, it is not automatically transferring to the others.
Why does ChatGPT cite almost nothing?
Asked “hcu recovery”, ChatGPT produced a complete, confident answer and cited exactly one external source. Asked again with an explicit instruction to search the web and cite its sources, it returned the same single citation.
For this topic, in default use, ChatGPT is answering from training data rather than live retrieval. That means “getting cited in ChatGPT” is largely not a competition you can enter here — which quietly undercuts a good deal of the AEO advice currently being sold.
Does Perplexity favour big brands?
The opposite, on this evidence. Perplexity’s 93 citation slots spread across 63 domains, and only 15 of those 93 slots (16%) went to household-name SEO publishers — Search Engine Journal, Search Engine Land, Semrush, Moz, Backlinko, Surfer, SE Ranking, Sistrix and Google’s own blog combined.
The most-cited domains in Perplexity’s set were thestacc.com and crowdo.net, at four citations each — tied with Search Engine Journal. Ahrefs was cited zero times.
The specialists who did recur — Marie Haynes, Glenn Gabe, Patrick Stox — were cited for precisely the topic they are known for. Narrow and specific beat broad and large.
What happened when we measured our own site?
Here is the uncomfortable part, published because leaving it out would make the rest untrustworthy.
This site has six backlinks from five referring domains and an average US Google position of 17.3 — page two, effectively invisible. In Google’s AI Overviews for these ten questions, seorevive.in was cited on nine of ten, making it the most-cited domain in the Google set, ahead of YouTube, LinkedIn and Search Engine Journal. On “google hcu baked into core” the AI Overview cited two of our pages at once.
In Perplexity, the same site on the same day was cited on two of ten.
Our first instinct was that the Google number had to be personalisation — those searches ran in a logged-in account belonging to this site’s owner. So we re-ran a subset with personalised results disabled and the locale forced to the US (&pws=0&gl=us). seorevive.in was still cited on all three re-tested queries, and on one of them it was the first source listed. That is meaningful corroboration, though not a clean-room test — see the limitations below.
What seems to earn the citation
Ten queries cannot establish causation. But the pages that were cited share three traits, and all three are copyable:
- They answer one narrow question. Our most-cited page exists to answer “what changed when HCU merged into core”, and it was cited for almost exactly that phrase.
- They open with a liftable answer. Both cited pages lead with a short direct answer before any narrative.
- They are specific enough to quote. Dates, mechanisms, named consequences — the things a model can repeat without hedging.
The pages of ours that were not cited are the broad, general ones.
Limitations — read these before quoting us
We would rather be cited accurately than widely.
- The Google measurements ran in a logged-in account belonging to this site’s owner, based in India. We could not complete a fully clean-room replication: signed-out attempts from a separate browser were served a CAPTCHA, which we would not bypass. The
pws=0&gl=usre-test is corroboration, not proof. A 9-of-10 citation rate for one’s own site, measured from one’s own account, is exactly the result a reader should interrogate. - ChatGPT was also a logged-in account with prior SEO conversations in memory.
- One day only — 21 August 2026. AI answers drift, sometimes quickly.
- Ten queries is a small sample, and only two were put through ChatGPT.
- One query was contaminated. “How long does HCU recovery take” returned medical sources, because “HCU” is ambiguous outside SEO. Weight that row lower.
- We are an interested party. We sell recovery services and this study found us cited. Judge it accordingly — and re-run it yourself.
Frequently asked questions
Do all AI engines cite the same sources?
No. Across 10 identical questions about Google algorithm recovery, Google AI Overviews cited 50 unique domains and Perplexity cited 63 — but only 18 domains appeared in both. Roughly two-thirds of each engine's citation set was unique to that engine, so optimising for one does not automatically win you the others.
Does ChatGPT cite websites?
For these queries, barely. Asked “hcu recovery” it produced a full answer citing a single external source, and asked again with an explicit instruction to search the web and cite sources it returned the same single citation. In default use on this topic it answers from training data rather than live search, so it is not really a citation channel.
Do you need a high-authority domain to be cited by AI engines?
Not on this evidence. 84% of Perplexity's 93 citation slots went to domains without meaningful brand recognition, and its two most-cited domains were tied with Search Engine Journal. Ahrefs was cited zero times across the whole set.
Is being cited in AI search the same as ranking in Google?
No. seorevive.in has an average US position of 17.3 — page two — while appearing in 9 of the 10 Google AI Overviews we tested. The blue-link ranking and the citation slot behave like separate competitions.
How do you get your site cited by an AI engine?
On this data: answer one narrow question per page, open with a short liftable answer before any narrative, and be specific enough to quote — dates, mechanisms, named consequences. The broad overview pages in our sample were not cited; the narrow, specific ones were.
Want to know whether AI engines cite your site?
We run the same test on your brand — the questions your buyers actually ask, across Google AI Overviews, Perplexity and ChatGPT — and tell you exactly where you appear and where you don’t.
See the AI Visibility Audit →Not in India? We work with clients across the US, Canada, the UK, Australia and Singapore — same playbook, delivered async, billed in USD. See our USD recovery pricing →
SEO & Growth Specialist with 8+ years of experience and 100+ campaigns delivered across India, USA and Canada. Founder of SEORevive, based in Guwahati. Specialises in Google penalty and algorithm recovery, technical SEO and organic growth.