GEO and AEO: getting cited in AI answers

Search is shifting into AI answers. What GEO and AEO mean, why it matters now, and what you can concretely do.

Your rankings hold steady. Traffic drops anyway.

This is not a measurement error. The Pew Research Center examined the click behavior of 900 US users in July 2025. When an AI summary appears above the search results, only 8 percent click through to a classic result. Without an AI answer, it is 15 percent. As for the source cited in the summary, only 1 percent clicks it.

Users get their answer directly. They never see the page behind it. And the ranking report still shows position 3.

This is exactly where GEO and AEO come in.

What are GEO, AEO, and SEO?

SEO (search engine optimization) optimizes for the classic results list. Ten blue links, sorted by relevance. The goal is the click to your page.

GEO (generative engine optimization) optimizes for the AI answer itself. The goal: that a model picks up your content and names you as a source. This means ChatGPT, Perplexity, Google AI Overviews, and Gemini. The term comes from a Princeton study from 2024.

AEO (answer engine optimization) is narrower. It targets the one direct answer: a featured snippet, a voice assistant, or an AI box. In practice, AEO and GEO overlap heavily. Many use the terms synonymously.

The difference in one sentence: SEO fights for the click, GEO for the mention. GEO is essentially SEO for generative systems. Some also call the same thing "LLM SEO." The field is young, and the terms are not yet cleanly separated.

An example. A user asks ChatGPT: "Which tool is right for X?" The model names three providers and links two sources. If you appear in neither the list nor the sources, you do not exist in that moment. Your Google ranking for "X tool" changes nothing about that.

Why this is tipping right now

The shift is not a forecast. It is underway.

Google's AI Overviews reached around 2 billion users per month in July 2025. That is stated in Alphabet's quarterly report, a good year after launch. In October 2025, ChatGPT reported 800 million weekly users (OpenAI CEO Sam Altman).

And these AI answers are no fringe phenomenon. According to Pew, roughly one in five Google searches showed an AI summary in March 2025. Nearly six in ten users got at least one. After such an answer, one in four sessions ended (26 percent), and without an AI box only one in six (16 percent).

Many of these people no longer search on Google and click through. They ask a model and read the answer. The click, once the core of every SEO strategy, becomes optional.

For you, that means the stage is shifting. If you optimize only for the ten blue links, you optimize for a surface that is shrinking. Your report from Search Console and GA4 shows this loss only indirectly. The click-through rate drops, the position stays.

What AI answers cite

Why does a model cite one page and not another? The first major study on this comes from researchers at Princeton (KDD 2024). They tested nine content strategies across 10,000 queries. Their core finding: targeted changes to the content raise visibility in AI answers by up to 40 percent.

What worked most strongly:

• Numbers and statistics in the text. Models preferentially pick up substantiated facts.
• Quotes from sources. They act as an authority signal.
• Clear sourcing, prominently placed.

On top of that come patterns from the practice of AI search engines. Perplexity, for example, pulls only a handful of sources per query and cites a portion of them. It favors pages that answer the question early and are clearly structured.

Across the engines, one pattern shows up: AI answers concentrate their citations on a few authoritative sources. Established trade publications land in the answer more often than personal blogs. Topical authority often weighs more heavily than raw domain strength. A specialized page can beat a large generalist if it hits the question better. Freshness counts too: recently maintained content is measurably preferred.

The principle behind it: a model cites what it can extract cleanly. Buried answers, vague statements without structure: those it skips.

8 measures for your AI visibility

These points work on the substantiated levers. There are no guarantees, more on that below. None of them is a trick. They are moves that make your content more readable for a model.

1. Core answer up front. Answer the main question in the first 100 words. No long intro.
2. Facts with a source. Build in numbers, data, and named sources. This is the strongest substantiated lever.
3. Structure to extract. Meaningful headings, short paragraphs, lists for anything enumerable.
4. Use the question format. Phrase subheadings as questions that users actually ask.
5. Topical authority over breadth. Cover one topic deeply. Topical authority beats raw domain strength.
6. Show freshness. Set a date, maintain your content. Several AI systems prefer fresh pages.
7. Stay crawlable. Allow the AI crawlers (GPTBot, PerplexityBot, Google-Extended in robots.txt) and keep the technical basics clean.
8. Get mentioned off-site. Secure mentions on pages that AI draws on anyway: industry directories, trade publications, forums.

The foundation remains good SEO. GEO is not a replacement, but a layer on top.

How to measure AI visibility

The ranking report barely helps here. It shows positions, not mentions.

Manually, it works like this. Put the same prompts into ChatGPT and Perplexity regularly, such as "best [your category] providers" or "what is [your topic]." Check whether your brand shows up and which source the model names. Log it over weeks.

The catch: AI answers fluctuate. The same prompt delivers a different result tomorrow. A single test says little. It only becomes reliable across many prompts and repetitions.

That is exactly what specialized tools automate. They sample AI answers across hundreds of prompts and count how often and with which source your brand appears. AI agents are taking over this reporting to a large extent by now.

What does this mean for agencies?

For agencies, GEO shifts the reporting logic. Before long, the client no longer asks only about rankings. They ask whether ChatGPT recommends them. Whoever measures and explains that delivers something most competitors don't yet have on their radar.

The effort per client is manageable. The substantiated levers overlap heavily with good content SEO. So you build on work you are doing anyway. What's new is the measurement. A regular look shows whether the brand appears in AI answers and which source the model cites.

That is exactly where a reason to talk arises. A short AI visibility check opens the door to a client whom a pure ranking report no longer impresses.

Limits and an honest assessment

GEO is a young field. A few things need to be said clearly.

There is no guarantee of a citation. AI answers are not deterministic. You optimize probabilities, not fixed slots. Anyone who promises "rank 1 in ChatGPT" is selling you something.

The body of research is thin. The Princeton work is strong, but one of the few. The AI systems keep changing their selection. What works today may look different in six months.

And measurement remains a sample. You get a picture, not the full picture. AI answers are partly personalized and vary by region. What a user in Vienna sees need not match what a model outputs in Hamburg. Clean sampling covers ranges, not individual cases. That is still worth more than guessing.

The honest advice: no gaming, no tricks. The substantiated levers are simply substance, well sourced and cleanly structured.

Where do you stand right now?

Want to know whether the models even know you exist? The geo-audit skill in the Honeyfield Marketing MCP checks this systematically. Across many prompts, it measures whether and where your brand is cited in ChatGPT, Perplexity, and AI Overviews. The first assessment runs via marketing-mcp.honeyfield.at.