For twenty years the unit of search success was a position. You ranked, someone clicked, and the visit was yours. That unit has changed. The unit now is a sentence — the one an answer engine reads out loud before anyone decides whether to click.
Answer Engine Optimization (AEO) is the discipline of being that sentence. Not the result underneath it, not the ten blue links beside it — the answer itself. Where SEO optimises a page to rank, AEO optimises a passage to be selected, quoted and attributed.
GEO.thejourney.hkThe reason this matters is not philosophical. Pew Research Center measured it in July 2025 across 68,879 Google searches: when an AI summary appeared, users clicked on a traditional result in 8% of visits. When no summary appeared, they clicked in 15%. Same intent, same results page — roughly half the clicks. And the citation links inside the answer barely get touched at all: someone followed one out of the summary in just 1% of those visits.
The same study shows how fast the layer spreads wherever people ask rather than browse. AI summaries appeared on 18% of the searches Pew captured — and on 60% of searches that opened with a question word. Who, what, when, why. Ask a question, receive an answer, skip the list.
That is the entire case for AEO in three numbers. The answer layer is not a channel you can opt out of. It is the surface where your audience now forms its first judgement about you.
The Answer Layer, In Numbers
| Signal | Figure | What it tells you |
|---|---|---|
| Click-through when an AI summary appears | 8% | Pew Research Center, July 2025 — 68,879 Google searches |
| Click-through when no summary appears | 15% | Same study; the summary itself is what changes behaviour |
| Relative click loss with a summary | ~47% | The measured gap between 8% and 15% |
| Zero-click rate in Google AI Mode | 93% | Semrush, September 2025 — AI Mode rarely sends anyone anywhere |
| Click loss at position 1 with an AI Overview | −58% | Ahrefs, December 2025 — up from −34.5% nine months earlier |
| Queries that trigger an AI Overview | 13% | Semrush, March 2025 — roughly one search in eight |
| AI Overview citations drawn from top-10 pages | 76% → 38% | Ahrefs, 2025 versus 2026 — ranking still helps, but no longer decides |
| ChatGPT citations drawn from the first third of a page | 44% | Kevin Indig, February 2026 — 18,012 verified citations |
Read that middle column the way you would read a weather forecast. These are not opinions about the future. They are measurements of the present.
- 01QueryWhat the person actually asked
- 02RetrievalCandidate passages are gathered
- 03SynthesisOne answer is assembled from many
- 04AnswerThe sentence they read
- 05CitationThe only layer that carries your name
You can influence every stage. You are only rewarded at the last one.
The layer has a shape. A question is asked, an engine gathers candidate passages, a model assembles one answer out of them, and then — sometimes — it names the sources it leaned on. You can influence every stage of that pipeline. You are only rewarded at the last one.
AEO Is Not a Smaller SEO
| SEO | AEO | GEO | |
|---|---|---|---|
| What it optimises | A page, to rank | A passage, to be answered with | A brand, to be cited and recommended |
| Where it plays | Google and Bing result pages | Featured snippets, AI Overviews, AI Mode | ChatGPT, Gemini, Claude, Perplexity |
| What success looks like | Position, clicks, CTR | The answer you wrote, attributed to you | Share of answers, citations, mentions |
These are not three names for one job. SEO decides whether you are in the candidate set at all. AEO decides whether the sentence that reaches your customer is yours. GEO decides whether an engine recommends you by name when it is weighing you against a competitor.
We covered the wider split in Why GEO is not just SEO, and the way each engine rewards a different move in The Great Fragmentation. The full comparison, column by column and with the click data behind it, is in SEO vs GEO: a list of links against a cited answer.
GEO.thejourney.hkHow to Become the Answer
Answer engines do not rank pages. They lift passages. Every technique below is a variation on one instruction: make a passage easy to lift, and easy to attribute.
01 — Lead with the answer. The first forty to sixty words under a question-shaped heading should be a complete, standalone answer. If a reader stops there, they should still have what they came for. If a model stops there, it should have a quotable sentence. The data agrees with the instinct: 44% of ChatGPT's verified citations come from the first third of a page, and cited passages are almost twice as likely to open with a plain definition — 'X is', 'X refers to'.
02 — Write headings as questions. Engines map questions to headings before they map them to pages. A heading that mirrors how people actually ask is a shortcut into the answer layer. Of the citations in that same study that tied back to a question, 78.4% came from a heading — the model reads the H2 as the prompt and the paragraph beneath it as the answer.
03 — Keep every passage self-contained. A paragraph that opens with 'as we covered above' cannot be lifted cleanly, so it usually is not. Pronouns and cross-references are the enemy of extraction.
04 — Use tables for anything comparative. A clean table of a few rows and columns is one of the most reliably extracted formats on the web. It survives synthesis with its meaning intact.
05 — Say the entity out loud. Name the brand, the product and the category inside the passage itself. A fact stated without its subject is a fact that gets attributed to someone else. Proper nouns are the tell: ordinary English runs 5–8% proper nouns, while the passages engines cite most average 20.6%. Density of named things is a citation signal, not a style choice.
06 — Keep the page machine-legible. Structured data, a clean heading order and crawlable text still decide whether your passage enters the candidate set. AEO does not replace technical SEO; it depends on it. Though the dependency is loosening: in 2025, 76% of AI Overview citations came from pages ranking in the top ten. By early 2026 that had fallen to 38%. The answer layer now draws more of its material from beyond page one than from it.
07 — Keep the facts current. Answers are assembled from whatever looks most reliable at the moment of asking. An undated statistic is a liability; a dated one is a citation waiting to happen.
GEO.thejourney.hkWhere AEO Fails
Two failure modes are worth naming, because both of them look like success in a dashboard.
Cited but not named. Your passage is used, your phrasing survives into the answer, and your domain never appears. This is the mention-versus-source divide, and it is the most common way a strong AEO effort goes unrewarded.
Look at who actually gets cited and the scale of the competition is obvious. Across a thirteen-week Semrush study of 230,000 prompts and more than 100 million citations, the top sources were platforms — Reddit and Wikipedia on ChatGPT, LinkedIn and YouTube on Google's AI Mode. Peec AI's analysis of 30 million sources found the same: Reddit the single most-cited domain across ChatGPT, AI Mode, Gemini, Perplexity and AI Overviews, with YouTube, LinkedIn, Wikipedia and Forbes behind it. You are not only competing with your competitors. You are competing with the platforms your customers talk on — which makes being present in those conversations an AEO task, not a public relations one.
Read but not counted. The engine quotes you inside the answer and nobody clicks. Session-based analytics record nothing, so the work looks worthless — while your brand is being repeated to your audience every day.
The fix is unglamorous. Put the name inside the sentence you want quoted, and give the engine a reason to prefer your version of a fact — a number, a definition, a dated figure it cannot get anywhere else.
GEO.thejourney.hkMeasuring AEO
Rank tracking cannot see this layer, because there is no list to be on. Three measures can.
Share of answer — how often your passage is the one synthesised into the answer for a question you care about.
Citation rate — how often your domain is named among the sources an engine leaned on.
Answer-only questions — questions that earn impressions and no clicks at all. These are not failures. They are the answer layer working exactly as designed, with you either inside the answer or outside it.
Your own Search Console data will tell you more than any third-party estimate. A growing set of question-shaped queries with impressions and zero clicks is the signature of the answer layer arriving in your market. The only question worth asking is whether your name is in the answer.
Which engine is doing the answering matters too — the share of AI referrals has been shifting quickly — but the mechanics of being the answer are the same wherever the question is asked.
GEO.thejourney.hkSources
Every figure in this piece comes from the studies below. Where a number is a prediction rather than a measurement, we have said so.
• Pew Research Center (July 2025). Google users are less likely to click on links when an AI summary appears in the results — the 8% versus 15% click-through gap, across 68,879 Google searches and 12,593 searches that produced an AI summary. Also the source for citation links being followed in 1% of visits, summaries appearing on 18% of searches, and 60% of question-word queries producing one.
• Semrush (September 2025). Google AI Mode's Early Adoption and SEO Impact — the 93% zero-click rate in AI Mode.
• Semrush (2025). How to Win in a Zero-Click Search Market — AI Overviews triggered on 13.14% of queries in March 2025.
• Ahrefs (December 2025). Update: AI Overviews Reduce Clicks by 58% — the click-through loss at position 1, measured against the forecast had AI Overviews never rolled out. The original study, showing −34.5% in March 2025, is here.
• Ahrefs (2025). 76% of AI Overview Citations Pull From the Top 10 — 1.9 million citations, and the baseline for the ranking dependency.
• Ahrefs (2026). Update: 38% of AI Overview Citations Pull From The Top 10 — 863,000 SERPs, showing how far that dependency fell within a year.
• Kevin Indig, reported by Search Engine Land (February 2026). 44% of ChatGPT citations come from the first third of content — 18,012 verified citations from 3 million ChatGPT responses. The source for the 44% figure, the 78.4% heading figure, and the 20.6% entity density. Full report: The science of how AI pays attention.
• Semrush (October 2025). The Most-Cited Domains in AI: A 3-Month Study — 230,000 prompts and 100 million citations across ChatGPT, Google AI Mode and Perplexity.
• Peec AI, reported by Search Engine Land (2026). AI search engines cite Reddit, YouTube, and LinkedIn most — 30 million cited sources across ChatGPT, AI Mode, Gemini, Perplexity and AI Overviews.
Ranking is a position you hold. Being the answer is a sentence someone repeats. Only one of them survives the answer layer.