What an AI Search Engine Is, and How It Decides What to Cite

An AI search engine does not hand you a list of links to sort through. It searches the web, reads parts of a few pages, writes one answer and cites some of those pages. Ranking has not gone away, though: Google's own guide for site owners says its AI answers find pages with the same core ranking systems that run its normal search results.

If you run an online store and keep hearing that ChatGPT and Google's AI now answer shoppers, the question for your store is which pages end up cited, and why. This guide walks through how ChatGPT, Perplexity and Google AI Overviews decide, using what each company publishes and what outside studies have measured, and it keeps those two kinds of evidence apart. The short answer comes first, then the detail.

S2 | tldr

The short answer

An AI search engine is a search tool that answers in its own words and links the web pages it used, as Google AI Overviews, ChatGPT search and Perplexity do. It searches, reads parts of a few pages, writes one answer and cites some of them.

• It works in steps. It decides whether to search, splits your question into several searches (query fan-out), reads parts of the pages it finds, then writes one answer and cites some of them. The method is called retrieval-augmented generation (RAG).

• Google says its AI answers find pages with its normal ranking systems, and that a page must be indexed and able to show a snippet to be linked.

• Google and OpenAI publish eligibility rules, and Perplexity publishes a general description. None of them publishes a formula for which pages get cited.

• Outside studies find that many cited pages do not rank in Google's top 10 for the original question, and that AI assistants lean toward fresher pages and third-party sources.

• For a store, the work is plain: let the engines reach your pages, keep product data accurate, answer buyers' questions with facts other stores do not have, and get mentioned by others.

Before the detail, it helps to be clear on what counts as an AI search engine and what does not.

S3 | toc | jump list; "FAQ" anchors to the first FAQ question (S12)

What an AI search engine is

An AI search engine is a search tool that uses a large language model (LLM) to search the web for your question, then writes one answer that cites the pages it used. Google AI Overviews, ChatGPT search and Perplexity work this way.

Three kinds of tool now answer questions, and they work differently:

• A classic search engine, such as Google Search, returns a ranked list of pages for you to open.

• A chatbot answering from memory writes from what its model learned in training. There is no fresh page behind the answer to cite.

• An AI search engine does both jobs. It searches first, then writes, and links its sources.

These are the ones you are most likely to meet, described the way each company describes them:

• Google AI Overviews are the summaries at the top of some Google results. Google says they appear only when its systems judge them "additive to classic Search".

• Google AI Mode is a separate, conversational mode inside Google Search. Google says AI Mode and AI Overviews "may use different models and techniques".

• ChatGPT search is ChatGPT going to the web on its own when, in OpenAI's words, "your question would benefit from current information".

• Perplexity calls itself an answer engine that searches the internet in real time, with numbered footnotes linking to the original sources.

• Microsoft Copilot, Microsoft's assistant, also answers with web sources.

The words "AI search" also turn up inside business software. ServiceNow AI Search, for example, searches a company's own records. This guide is about engines that answer from the public web, and whichever one you use, the answer is built in a series of steps.

S5 | body

How an AI search engine builds an answer

An AI search engine builds its answer in five steps, and only the last one is the part you read. It decides whether to search, turns your question into several searches, pulls pages from a search index, reads the parts that match, then writes one answer and cites some of those pages.

1. It decides whether to search. OpenAI says ChatGPT searches when "your question would benefit from current information". When ChatGPT does not search, it answers from what its model learned in training, and there is no fresh page to cite.

2. It turns your question into several searches. Google calls this query fan-out, and OpenAI says ChatGPT "typically rewrites your query into one or more targeted queries". An Ahrefs study of 118,931 searches run by ChatGPT found an average of 1.78 searches per question (August 2025).

3. It pulls candidate pages from a search index. Google's AI features use the Google Search index and Google's core ranking systems. OpenAI says ChatGPT works with third-party search providers, and Perplexity says it searches the web in real time.

4. It reads the parts of those pages that match each search. None of the three companies publishes how much of a page it reads. The 2020 research paper that named the method split Wikipedia articles into 100-word chunks.

5. It writes one answer and attaches sources. Google says its models "identify more supporting web pages" while the answer is being written. ChatGPT adds inline citations and a Sources button, and Perplexity adds numbered footnotes.

The citations you see are chosen in steps 3 to 5, which is why a page can be found and still not be cited. Two terms sit behind these steps: RAG and query fan-out.

S6 | body

RAG and query fan-out, in plain words

Retrieval-augmented generation (RAG) means the model writes its answer from documents it has just fetched, instead of only from what it learned in training. Query fan-out means the engine splits one question into several related searches, then builds its answer from what all of them return.

The term RAG comes from a 2020 paper by researchers at Facebook AI Research, University College London and New York University, presented at NeurIPS 2020. Google calls the same step "grounding" and says it "relies on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index".

Google defines query fan-out as "A set of concurrent, related queries generated by the model to request more information". It does not show site owners which searches it ran.

For example, a question such as "which sunscreen holds up in Dubai summer heat" could be split into searches about SPF ratings, water resistance and reviews. That example is ours, written for this guide; it is not a query captured from Google or ChatGPT.

Once the pages are fetched, each company says something different about what decides which of them get cited.

S7 | body

What Google, OpenAI and Perplexity say about choosing sources

Google, OpenAI and Perplexity describe how their AI search works, but none of them publishes how it picks one page over another. Google requires a page to be indexed and able to show a snippet; ChatGPT requires its OAI-SearchBot crawler to be allowed, and OpenAI says placement is not guaranteed.

Two details on these pages trip site owners up. Google Search Central lists snippet controls (nosnippet, data-nosnippet and max-snippet) that limit what Google can show from a page, and that includes its AI features. OpenAI runs separate crawlers: blocking GPTBot tells OpenAI not to use your pages for model training, but it does not remove you from ChatGPT search, because OAI-SearchBot controls that.

Lists of "citation factors" on agency and tool blogs are those writers' own readings, not platform rules. What outside studies have measured about the pages that do get cited is a separate question, with its own evidence.

S8 | body

What studies find about cited pages

Outside studies have measured which pages AI engines cite, and three patterns repeat: ranking for the exact question is not the whole story, AI assistants lean toward newer pages, and third-party sources carry weight. None of these is a platform rule.

Ranking still matters, but not only for your question

Ahrefs found that 76% of the citations in Google AI Overviews ranked in Google's top 10 in July 2025, across 1.9 million citations. A larger Ahrefs study in March 2026 found 37.9%, across 4 million AI Overview links. Ahrefs changed its method between the two and says the results are not directly comparable, so the two figures do not show a trend.

For ChatGPT, 8% of in-text citations ranked in Google's top 10 for the original question, in a test of 15,000 prompts (August 2025). Ahrefs suggests query fan-out as the reason: pages get cited for the side searches, not for the question you typed. Google does not publish those side searches.

Newer pages show up more on assistants

Pages cited by ChatGPT, Perplexity, Gemini and Microsoft Copilot were on average 25.7% newer than the pages in Google's organic results, across about 17 million citations (Ahrefs, July 2025). Google AI Overviews were the exception and cited slightly older pages. Even so, the average cited page was close to three years old.

Other people's pages count

A University of Toronto preprint from September 2025 found that AI search shows "a systematic and overwhelming bias towards Earned media", meaning third-party reviews and publishers rather than a brand's own pages. Pew Research Center found that Wikipedia, YouTube and Reddit were the most cited sources in Google's AI summaries (US adults, March 2025).

Most of these figures come from Ahrefs, which sells an AI-visibility tool and used that tool's own data. They show patterns, not causes.

No study has found one factor that guarantees a citation from any engine, but each pattern points to a practical reason one page gets quoted and another does not.

S9 | body

Why some pages get quoted and others do not

A page gets passed over at one of five points, and most of them happen in the steps before the answer is written. The engine cannot reach it, does not find it for any of the side searches, finds no part that answers a side question on its own, sees it repeat what other pages say, or sees others cited on the topic instead.

• The engine cannot reach or show the page. It blocks OAI-SearchBot, carries a noindex tag, is missing from Google's index, or uses nosnippet. OpenAI and Google both document these.

• It was not found for any of the side searches, because the engine searched for parts of the question your page does not cover. Google documents the fan-out step, and the Ahrefs studies above show its effect.

• No part of it answers a side question on its own. Engines quote parts of pages, so an answer spread across several paragraphs is harder to lift. This is an inference from how retrieval works; no platform states it as a rule.

• It says what other pages already say. Google Search Central calls this commodity content, gives the example "7 Tips for First-Time Homebuyers", and says that content people find "unique, compelling, and useful" will likely influence a site's presence.

• Others are cited on the topic and you are not. Studies show AI search leans on third-party sources.

One 2023 lab test, from the research paper that named Generative Engine Optimization (GEO), found that adding quotations raised a source's share of the AI answer by about 41% on its main measure, and that repeating keywords lowered it by about 8%. Both results applied to pages that were already among the five sources the AI read, and the test did not measure rankings, clicks or sales.

Knowing what does not decide a citation is just as useful, whatever it is sold as.

S10 | body

What does not decide a citation

Google's guide for site owners, published on Google Search Central, lists what you do not need for its AI features, and it says "optimizing for generative AI search is optimizing for the search experience, and thus still SEO". OpenAI adds that ChatGPT's product results are not ads.

• No llms.txt file, other "AI text files", special markup or Markdown copies of your pages (Google Search).

• No need to break your content "into tiny pieces" (Google Search).

• No special schema.org markup, and no special way of writing for AI (Google Search).

• No paid route into ChatGPT's product results. OpenAI says they "are not ads, nor influenced by any OpenAI partnerships" (ChatGPT).

Each of these is stated by one platform for its own product; none of them is a rule for every engine. Search engine optimization (SEO) basics still apply, and for an online store they come down to a short list of checks.

S11 | body

What this means for an online store

For an online store, being the cited page now matters alongside ranking, because fewer people click through when an AI summary answers first. Five checks help: let OAI-SearchBot and Google reach your pages, keep product data accurate in Google Merchant Center and your Shopify catalog, answer buyers' questions with your own facts, get mentioned by others, and track AI referrals.

The click data is from the US. In browsing data from 900 US adults in March 2025, Pew Research Center found that 18% of Google searches showed an AI summary. People who saw one clicked a regular result in 8% of visits, against 15% when there was no summary.

On the shopping side, OpenAI says ChatGPT's product results use "Structured metadata from first-party and third-party providers (e.g., price, product description)". Shopify stores reach ChatGPT through Shopify Catalog, and OpenAI says "no additional work is required"; other merchants can apply for a direct product feed. Google says Merchant Center feeds and a Google Business Profile can help products appear in its AI answers.

1. Check that the engines can reach you. OAI-SearchBot should not be blocked, and your key pages should be indexed in Google and able to show a snippet. These are the same basics as SEO.

2. Keep product data accurate and complete in Google Merchant Center and, on Shopify, in the product details your catalog sends out.

3. Answer the questions buyers ask, one per section, with figures and named sources that are yours rather than a rewrite of what other stores say.

4. Get mentioned by others, through reviews, publishers and comparison pages.

5. Track what AI sends you. Links from ChatGPT carry "utm_source=chatgpt.com", so you can see those visits in your analytics.

At MG Lumeo, we work on the product data AI systems read and on original, checkable content, and we tell you when a lever does not exist. The questions below are the ones people search for next.

S12 | faq | FAQ heading printed as a question

Is ChatGPT an AI search engine?

ChatGPT is an AI search engine when it searches the web. OpenAI launched ChatGPT search on 31 October 2024 and says ChatGPT "will choose to search the web based on what you ask". When it does not search, it answers from what its model learned in training, so there is no fresh web page behind the answer to link. When it does search, the pages come from a search index, and the index differs by engine.

S13 | faq | FAQ heading printed as a question

Which search engine does AI use?

Each AI search engine uses a different source of pages, and only Google says exactly which. Google AI Overviews and Google AI Mode use the Google Search index and Google's core ranking systems. OpenAI says ChatGPT works with third-party search providers, and its help page does not list them all. Perplexity says it searches the web in real time and does not say whose index it uses.

One Ahrefs study found that only 6.82% of ChatGPT's search results appeared in Google's top 10 for the same searches (August 2025), so ChatGPT is not simply repeating Google. Which engine matters most for you depends on where your buyers ask.

S14 | faq | FAQ heading printed as a question

What is the best AI search engine?

The best AI search engine depends on what you use it for. PCMag (24 June 2026) and Zapier (14 September 2026) publish tested rankings for searchers; for a store owner, the more useful question is which engines your buyers use. Check how your products appear in Google AI Overviews, Google AI Mode, ChatGPT search, Perplexity and Microsoft Copilot. The work of showing up in them goes by several names.

S15 | faq | FAQ heading printed as a question

What is AI search optimization called?

AI search optimization is called several things: Generative Engine Optimization (GEO), Answer Engine Optimization (AEO), LLM SEO or AI SEO. Google spells out both answer engine optimization and generative engine optimization in its guide, and treats them as part of search engine optimization (SEO). The difference from classic SEO is the goal: being cited inside an answer, rather than only listed in results. The name Generative Engine Optimization comes from a 2023 research paper. Below is who wrote this guide and the sources it relies on.