Answer engine optimization examples: ten changes, each with its source

Answer engine optimization examples, in one table

The ten examples below are concrete changes you can make to a page or a site. Each row names the change and the public source that supports it, so you can read the reasoning for yourself.

Answer engine optimization, often shortened to AEO, means shaping your pages so that AI assistants such as ChatGPT, Perplexity and Gemini can use them in an answer. Our guide to what answer engine optimization is covers the idea. This post is only the examples.

The examples use a made-up company, Example Bakery Supplies, which sells ovens to small bakeries. It is invented to show the change. It is not a real business, and no result is claimed for it.

Examples 1 to 3: let the assistants reach the page

An assistant cannot quote a page it is not allowed to fetch or cannot read. These three examples come first because every other change depends on them.

Example 1: allow the search crawlers

A crawler is the program an AI company sends to fetch web pages. OpenAI's crawler page says OAI-SearchBot is used to surface websites in ChatGPT's search features. It also says that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links.

Perplexity's crawler page says PerplexityBot is designed to surface and link websites in Perplexity's search results, and that it is not used to train AI models. So the change for Example Bakery Supplies is a robots.txt that allows OAI-SearchBot and PerplexityBot. It can still block GPTBot, which OpenAI says is for training, if the company prefers.

A site can block these crawlers without meaning to, through a firewall setting or an old rule. Our page on AI crawler access explains what to look for.

Example 2: keep the answer in the page text

Microsoft's Bing team published guidance on AI search answers in October 2025. It says to avoid relying on PDFs for core information, to avoid putting key information only in images, and not to hide important answers in tabs or expandable menus.

For the bakery supplier, the price list lived in a PDF and the oven sizes lived in a product photo. The change is to write both out as text on the product page, with the PDF kept as a download for people who want it.

Example 3: stay indexable for Google

Google's page on AI features and your website says a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode. It says there are no additional technical requirements.

So a page marked noindex, or one with a nosnippet rule, is out of those features. The change is to check the pages that answer buyer questions and remove any rule that hides them from Google.

Examples 4 to 6: put the answer where it can be lifted

Assistants tend to quote a passage, not a whole page, so the answer should sit in one clear place. Bing's guidance says headings act like chapter titles that define clear content slices, and it recommends concise answers of one to two sentences.

Example 4: write the heading as the question

Bing says direct questions with clear answers mirror the way people search. A heading that reads "Our ovens" tells an assistant little. A heading that reads "What size oven does a small bakery need?" matches the question a buyer actually types.

Example 5: answer first, explain second

Under the question heading, the first sentence should be the answer. The detail, the caveats and the story come after it. This post is written that way: every section opens with its answer.

Example 6: use a table for a comparison

Bing names bulleted lists, numbered steps and comparison tables as formats for breaking down complex details. When a buyer asks which of two things to choose, a table with one row per option and one column per trait is easier to lift than three paragraphs.

None of these changes makes the page longer for its own sake. Each one moves a fact to where an assistant, or a hurried person, would look for it first. The page also becomes easier for a human buyer to scan, which is a good reason to make the change even if no assistant ever quotes it.

Examples 7 to 9: give the assistant evidence to repeat

Pages that carry checkable facts did better in the one controlled study of this question. The study is GEO: Generative Engine Optimization, by Aggarwal and colleagues, published at KDD 2024.

The paper tested changes to web pages and measured how visible they became in AI-written answers. Its abstract says these methods can boost visibility by up to 40 per cent. That is a relative change on the paper's own benchmark, not a promise for your site.

Example 7: add a statistic and its source

In the paper's results, adding statistics was one of the methods that did best, with a relative improvement of 30 to 40 per cent on its main measure. For the bakery supplier, that means "our ovens are efficient" becomes a measured figure for energy use, with a link to the test it came from.

Example 8: add a quotation from a named source

Adding a quotation and citing sources did as well in the same paper. A quote from a named baker who uses the oven, with their bakery's name, gives an assistant something specific to repeat.

Example 9: replace vague words with specifics

Bing's guidance says to avoid vague language. "Affordable" and "fast delivery" are vague. "From $4,200" and "delivered in 10 working days in the continental US" are specific, and an assistant can compare them with a rival's figures.

The same paper found that keyword stuffing, the old SEO habit of repeating a phrase, gave little to no improvement and sometimes made things worse.

Example 10: measure over many runs, not one answer

An assistant gives a different answer each time, so one check proves nothing. Measure how often you are named across many questions, each asked more than once.

SparkToro's study, published in January 2026, had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. It found less than a 1 in 100 chance that ChatGPT or Google's AI would give the same list of brands in any two responses.

The same study concluded that a visibility percentage across many prompts, each run several times, is a reasonable measure. It called any tool that reports a ranking position in AI "full of baloney". So the tenth example is a habit, not a page change: count how often you are named before and after each of the other nine.

You can do this by hand with a spreadsheet. Give each question a row and each run a column, and write in each cell the brands that answer named. Ten questions asked three times give you thirty answers, and a count out of thirty tells you far more than any single answer.

What Google says you do not need

Several popular AEO tactics are not needed for Google's AI features, by Google's own account. Its guide to optimizing for generative AI in Search, last updated in July 2026, says so directly.

If you already publish an llms.txt file, there is no need to remove it. The same guide says it is fine to keep such files for other services that use them, and that they neither help nor harm a site in Google Search.

That is not the same as saying these things are useless everywhere. Other assistants have not published equivalent statements, so their behaviour on these points is unknown.

How to try one example this week

Pick one buyer question you want to be named for, and work through the examples on the single page that should answer it. One page done properly teaches you more than ten pages done halfway.

1. Write down the exact question a buyer would type, in their words.
2. Ask it of two or three assistants, three times each, and note who is named each time.
3. Check that robots.txt lets the search crawlers reach the page.
4. Make sure the answer is in the page text, near the top, under a heading that asks the question.
5. Add one statistic with a source, one specific figure and, where it fits, one table.
6. Ask the same question the same number of times two to four weeks later, and compare the counts.

If you keep the counts, write them as "named in 2 of 9 answers", never as a bare percentage. Our method page explains why every figure on this site carries both numbers.

Where Agentcites fits

Agentcites does the measuring in Example 10 for you. Today it asks ChatGPT, Perplexity, Gemini and DeepSeek up to 50 buyer questions each week, keeps every answer word for word, and checks whether each assistant's crawler is allowed on your site. For a question you lost, it shows who was named instead and can draft a page aimed at that question. It costs $149 a month per website, as the pricing page says.