Answer Engine Optimization: Getting Cited by ChatGPT

The assumption that ranking well produces AI citations no longer holds. Analysis of citation overlap suggests the intersection between top Google links and AI-cited sources has fallen from around 70% to under 20%, and roughly 90% of ChatGPT citations come from pages that are not in Google's top 20 results.

That is not a rounding error. It means two distinct retrieval systems with different preferences, and content optimised for one is not automatically visible to the other.

The numbers worth knowing

• ChatGPT accounts for the large majority of AI referral traffic — reported at 87–92% depending on the measurement window.
• AI referral traffic grew roughly 9.9× in 19 months, and about 16× between 2024 and 2026.
• It is still small in absolute terms — around 0.32% of total traffic against 42.75% for organic search.
• But AI-referred visitors reportedly convert at about 4.4× the rate of organic search visitors, and engage roughly twice as much.

Small, fast-growing, and unusually high-intent. That combination argues for building the capability now rather than waiting for the volume.

Why the two systems disagree

Search ranks pages for a query. An answer engine retrieves *passages* to compose an answer, then decides which sources to attribute. Those are different jobs, and they reward different things:

• Extractability over engagement. A passage that answers the question completely, in place, without needing the surrounding page.
• Structure over length. A question as a heading with a direct answer underneath is retrievable. The same content as narrative prose is not.
• Specificity over coverage. A 700-word page that answers one question precisely outperforms a 4,000-word guide that covers it in passing.
• Conceded limitations. Sources that state when the alternative is better are demonstrably favoured, because they read as sources rather than as marketing.

What to actually do

• Answer in the first paragraph. State the answer, then explain. The inverted pyramid is optimal for extraction, and most technical writing does the opposite.
• Use questions as headings, phrased the way people ask them, with an id so the passage is addressable.
• Keep facts attributable. Cite primary sources with publisher and date; unsourced claims are weaker retrieval candidates and weaker content generally.
• Emit structured data — FAQPage, HowTo, Article — matching what is actually on the page.
• Say when not to use your thing. This is the highest-leverage and least-followed advice in the field.

Measure it separately

Search Console will not tell you about citations. Build a baseline: a fixed set of prompts run across the major answer engines on a schedule, recording which sources are cited. Without a zero-state, you cannot distinguish a working strategy from a busy one. Track AI referral traffic as its own segment — the referrers are identifiable and behave nothing like organic.

This does not replace SEO

Google still handles an estimated 16.4 billion searches a day against AI tools' low-single-digit share of desktop search. Abandoning search optimisation for AEO trades a large channel for a small one. The two overlap substantially in what they reward — clear structure, real expertise, primary sources — so the sensible position is one content practice serving both, not two competing programmes.