What is answer engine optimization (AEO)?

Buyers used to search and get ten links. Now many of them ask an assistant and get one answer, which names a few companies.

AEO is everything you do to be one of those companies. An "answer engine" is any assistant that answers in sentences instead of a list of links.

The goal is the same, being found by a buyer. What you win and where you check it are different.

Because there is no rank, the useful measure is how often you are named across many questions. Count that apart from questions that already contain your name, because those name you almost every time. The method page explains why that split matters.

An assistant answers from two places. One is what it learned in training. The other is pages it reads on the web while it answers, when web search is on.

• From training, it repeats what was written about a category often and clearly before its training ended. You cannot change that quickly.
• From web search, it reads pages that match the question and names the brands those pages name. This part can change from week to week.
• Either way, a brand that is described plainly on many pages is easier for it to name than one described on few.

Research supports working on the pages. The GEO paper, published at the KDD conference in 2024, found that changes such as adding sources, quotations and statistics could raise how visible a page was in generated answers by up to 40 per cent in its tests.

None of that matters if the assistant cannot read your site. Each company sends its own crawler, a program that fetches pages, and your robots.txt file can let it in or keep it out.

You can do the first round by hand in an afternoon. The steps are the same with or without a tool.

1. Check that AI crawlers may read your site. Open your robots.txt file and look for lines that block GPTBot, ClaudeBot, PerplexityBot or Google-Extended. Google lists its own crawlers in its crawler documentation.
2. Write the questions a buyer asks before they know your name. The guide to finding buyer questions shows how.
3. Ask each question in each assistant your buyers use, with web search on, in a fresh chat.
4. Keep every answer in full, with its links, and note who was named.
5. Open the pages the answers cited. Those pages are why the answer came out the way it did.
6. Change what you can: a clearer page of your own, a fair comparison page, plain prices, or a place on the lists the answers cite.
7. Ask the same questions again later and compare, using the rules in how to report AI visibility.

The numbers in this example are made up to show the arithmetic. They are not a measurement of any real company.

Say a company that sells project planning software writes 20 questions without its name and asks them on four assistants. That gives 80 answers. It is named in 6 of the 80. A rival is named in 31 of the 80.

The company opens the pages cited in the answers that named the rival. Most of them are two "best project planning tools" articles on review sites, and neither article mentions the company. That is the first thing to work on, and no change to its own homepage would have found it.

The answer tells you who won. The pages it cited tell you why.

• It writes or collects the questions your buyers ask.
• It asks those questions of AI assistants on a schedule.
• It records which companies each answer named and which pages it cited.
• It reports how often you were named, and against whom.

Agentcites does those four things with ChatGPT, Perplexity, Gemini and DeepSeek, and two more: it suggests possible reasons you were missing, and it can start an editable draft for a question you are missing from. Other tools in the field are compared on our compare page.

• Nobody can promise that an assistant will name you. You can change what it reads, not what it writes.
• Answers change on their own. The same question asked twice can name different brands, so one reading is not a result.
• A page you publish today may not be read for some time. How long depends on when each assistant next fetches it.