What are AEO and GEO, and how do you get cited by ChatGPT, Perplexity and Google AI Overviews?

What are AEO and GEO, and how are they different from SEO?

AEO (answer engine optimization) grew out of featured snippets and voice assistants: shape a page so a system can lift one clean answer from it. GEO (generative engine optimization) was named in a 2023 paper by Aggarwal and colleagues, presented at KDD 2024, about engines that write an answer from several sources and link a few. In 2026 the two words are mostly used interchangeably.

SEO earns a position in a list of links; AEO/GEO earns a sentence inside an answer, often without a click. The foundation is shared. Google states that a page must be indexed and eligible to show a snippet to appear in AI Overviews or AI Mode, with no additional requirements.

How do ChatGPT, Perplexity and Google AI Overviews choose which pages to cite?

None of them reads the whole web while answering. Each turns the question into searches, pulls candidate pages from an index and grounds the answer in a few of them. Google calls this query fan-out; Microsoft reports the phrases as grounding queries. In OpenAI's own example, ChatGPT adds a year to a search, or the user's city to a "near me" question.

Can you see AI retrieval in Search Console?

Our own data: in the viton13.com Search Console property for 23 June to 23 September 2026, the query report holds many searches that do not look typed by a person. 32 of the top 160 queries by impressions start with % or +, such as %kylian mbappe current club 2026 real madrid (508 impressions). Together those 32 rows have 2,407 impressions, 0 clicks and average positions of about 3 to 10.

• Month-stamped: 51 of the 350 rows we exported add a month, as in "current club september 2026": 1,302 impressions, 0 clicks.
• A quoted headline plus the word date, as if checking when a story came out: 143 impressions for one of them.
• A garbled site: operator aimed at the American Academy of Dermatology, %site.aad.org "how to test skin care products".

They land on VJOURNAL's most-shown stories: Mbappé, OpenAI's leadership and a skincare patch test. Our hypothesis: software builds these queries (in a URL, + stands for a space and % starts an encoded character) to check freshness and provenance, much as ChatGPT adds a year to its searches. It could be answer engines, rank trackers or scrapers; Search Console does not say.

The lesson holds either way: whatever is asking wants current, dated, attributable facts.

Do you need llms.txt, content chunking or special schema?

Google's guide to generative AI features, updated 10 July 2026, lists as myths several things that are sold as GEO.

Where Google is silent, treat claims as unproven. The GEO paper found quotations, statistics and cited sources lifted its visibility metric by 30-40% on a simulated GPT-3.5 engine, and quotations by 22% on Perplexity at the time; keyword stuffing did little. A direction, not a forecast.

Which on-page changes make a page quotable?

• Answer first: 40-60 words under the heading, with the name, the number and the condition in one sentence.
• Date and source every figure. "Prices checked 23 Sep 2026" beats "recently".
• Keep the entity consistent: same name, address, services and prices on your site, profiles and directories, and an About page with a named founder.
• Put facts in the server-rendered HTML. OpenAI's and Perplexity's crawler pages say nothing about running JavaScript, so don't hide prices in images, PDFs or widgets.
• One question per section, with a heading that asks it.
• Leave nosnippet, data-nosnippet and a tight max-snippet off pages you want quoted: Google uses them to limit what Search, AI features included, shows from a page.

Which off-site signals matter: mentions, reviews, directories, forums?

Many sources in an answer are not yours. Run your prompts, list every domain cited and work on the ones that recur:

• profiles you control: Google Business Profile, Bing Places and, for Russian-speaking buyers, Yandex Business, which lists the company in Yandex Search and Maps;
• reviews on the platforms your buyers read, with factual replies;
• directories and marketplaces that compare your category;
• forums where you post under your own name: Google says chasing inauthentic mentions helps less than it seems, and paid posts dressed as opinions are exactly that.

How can you measure AI visibility without paid tools?

Those reports cover Google, Bing and Yandex. For ChatGPT and Perplexity, keep your own prompt panel:

1. Write 15-20 promptsIn buyers' words: half about the problem, half comparing providers, some with a price or city.
2. Run them in four enginesChatGPT, Perplexity, Google AI Mode and Copilot, in a clean session.
3. Record four columnsBrand mentioned, your page cited, competitors cited, third-party domains cited.
4. Repeat monthlySame wording, same day. Answers vary, so read trends, not single runs.

What can you do in one hour, and when is a paid audit worth it?

In one hour: check robots.txt against the crawler table, run URL Inspection on your three key pages, rewrite their first paragraphs as answers, date your prices and note today's numbers from the free reports.

An audit is worth buying when, after that hour, you still cannot say which sources engines trust in your category, why a competitor is cited and you are not, and what to test first. AI-assisted: AI drafts the prompt runs and gap map, a person directs and checks. Human-led: a specialist does the analysis, worth it for regulated categories or an ambiguous brand name.

What can AEO and GEO not promise?

• Citations. The engine picks the sources for each answer; Yandex states that even its own Alice AI report gives no way to control whether you appear.
• Clicks. In Pew Research Center's March 2025 data, people who saw a Google AI summary clicked a regular result in 8% of visits, against 15% without one, and a link inside the summary in 1%.
• Stable answers. One prompt can cite different sources on different days; Yandex says Alice AI builds each answer afresh from current results.
• Complete data. Every report above leaves out clicks, queries or both; a prompt panel is a sample.