Quick Answer: An answer engine is a system that retrieves web pages for a question, writes one synthesised reply and cites the sources it used. ChatGPT search, Perplexity, Google AI Overviews and AI Mode, Gemini, Claude and Grok all work this way. A search engine returns ranked links; an answer engine returns the answer, with links attached.
The phrase comes from the engines themselves. Perplexity's help centre defines an answer engine as "a tool designed to give you direct, detailed answers to your questions", and contrasts it with search engines that "make you sift through a list of links", per its help article.
For a brand the definition matters because the unit of competition changes. On a search engine you compete for a position; on an answer engine you compete for a sentence.
What is an answer engine
An answer engine is software that takes a natural-language question, fetches relevant pages from the live web, and composes one cited response rather than a list of results. The three verbs, retrieve, synthesise and cite, are what separate it from a chatbot answering from memory and from a search engine returning links.
Every current answer engine keeps a large language model at the centre and a retrieval layer around it. OpenAI describes ChatGPT search as a fine-tuned model that "leverages third-party search providers, as well as content provided directly by our partners", per its launch post. Google says AI Overviews and AI Mode "surface relevant links" and may fan a question out into many sub-searches, per its AI features guide.
What it is not
An answer engine is not a chatbot with no web access, and it is not a search engine with a summary bolted on. A model answering from training data alone cannot cite a page published last week, and a results page with a snippet still asks the user to choose a link. The answer engine does the choosing and shows its working.
Answer engine vs search engine
An answer engine synthesises and cites; a search engine indexes and ranks. The table sets the two side by side on the axes that change what a brand has to do.
| Axis | Search engine | Answer engine |
|---|---|---|
| Input | A keyword or short phrase | A full question, often with constraints, in a running conversation |
| Core process | Crawl, index, rank | Interpret, retrieve, synthesise, cite |
| Output | Ten ranked links plus snippets, ads and packs | One written answer with a handful of citations |
| Where sources come from | Its own index | Its own index, third-party search providers, or live fetches at question time |
| What the user does next | Clicks a result | Reads the answer, asks a follow-up, occasionally clicks a citation |
| Session shape | One query, one page | Multi-turn, with context carried forward |
| How a brand wins | Rank a URL | Be named, cited or linked inside the answer |
| How it is measured | Rank, impressions, clicks | Mention rate, citation rate, share of the answer |
The click gap
The gap between the two shows up in behaviour. Pew found Google users clicked a traditional result on 8% of visits when an AI summary appeared, against 15% without one, and clicked a link inside the summary on 1% of visits, per its July 2025 browsing study.
The answer is the destination for most users. What zero-click answers do to brand traffic follows that consequence through.
How an answer engine works
An answer engine works in four steps: it interprets the question, retrieves pages, synthesises a reply and attaches citations. Perplexity documents exactly that sequence for its own product, per its how-it-works page, and the other engines describe the same shape in their own terms.
Step 1: interpret
The model reads the question in context, including earlier turns in the conversation, and decides whether it needs the web at all. ChatGPT "will choose to search the web based on what you ask", according to OpenAI's launch post, or the user can force a search. A question the model thinks it can answer from memory may never trigger retrieval.
Step 2: retrieve
The engine turns the question into one or more searches and fetches the results. Google calls this "query fan-out", issuing "multiple related searches across subtopics and data sources", so a broad prompt becomes several narrow ones.
ChatGPT uses third-party search providers and partner content; Perplexity runs its own real-time web search. Each engine fetches with its own agent, such as OAI-SearchBot and ChatGPT-User for OpenAI, per its crawler documentation.
Step 3: synthesise
The model reads the retrieved passages and writes one answer from them. This is where a brand gets named or skipped: a claim that appears in several fetched sources is more likely to survive into the reply than a claim that appears in one.
Step 4: cite
The engine attaches links to the sources it used. Perplexity numbers its citations inline; ChatGPT shows a Sources button that opens a sidebar of references; AI Overviews show a panel of supporting links beside the summary.
A brand can be named in the text without a citation, or cited without being named, and the two are counted separately. How ChatGPT decides which pages to cite goes deeper on the ChatGPT case.
Which answer engines matter in 2026
Seven answer engines are the ones brands track in 2026: ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Gemini, Claude and Grok. Published scale runs from 100 million to 2 billion monthly users.
| Engine | Owner | Retrieval source | Crawler or fetch agent | Scale, as published |
|---|---|---|---|---|
| ChatGPT search | OpenAI | Third-party search providers plus partner content | OAI-SearchBot (index), ChatGPT-User (live fetch) | More than 1 billion weekly active users, per OpenAI, 31 Aug 2026 |
| Google AI Overviews | Google's index via query fan-out | Googlebot | 2 billion monthly users, 200 countries, per Google's Q2 2025 call | |
| Google AI Mode | Google's index via query fan-out | Googlebot | Over 100 million monthly active users, same call | |
| Perplexity | Perplexity | Its own real-time web search | PerplexityBot (index), Perplexity-User (live fetch) | Not published in its help centre |
| Claude | Anthropic | Web search with live page fetches | Claude-SearchBot (index), Claude-User (live fetch) | Not published in its crawler doc |
| Gemini, Grok | Google, xAI | Vendor-specific grounding | See each vendor's documentation | Not published alongside the above |
The ChatGPT figure is from OpenAI's 31 August 2026 ads update; the Google figures are from its Q2 2025 earnings call, as reported by TechCrunch. The best AI search engines compared ranks them on the user side.
Why the crawler column matters
Each engine reads a site with a named agent, and each agent obeys its own robots.txt block. A rule written for one does nothing to the others, so a site can be visible on Perplexity and invisible on ChatGPT through one careless line. Robots.txt rules for AI bots lists the blocks per engine.
What an answer engine changes for a brand
An answer engine changes the goal from earning a click to earning a sentence. Three shifts follow from the mechanism above.
The page is read as passages, not as a whole
Retrieval returns pages, but synthesis uses passages. A page whose answer is buried under six paragraphs of preamble can be fetched and still contribute nothing. Structuring content so an engine can lift it is the practical response: the answer in the first sentence, the evidence in a table, the number with its source attached.
Being named, cited and linked are three outcomes
A brand can be named in the answer with no link, cited in the source panel without being named, or both. Only the third sends a click, but the first two shape the buyer's shortlist before any click happens. Measuring AI visibility records the three separately for that reason.
The answer is not fixed
A search ranking persists until the next crawl or update. An answer is generated at the moment of the question, from whatever retrieval returned that time, so two runs of the same prompt on the same day can name different brands. Visibility on an answer engine is a rate over repeated readings, not a position.
How to tell whether an answer engine can see your site
Ask the engine a question your page answers, with web search on, and check whether it fetches and cites the page. Then confirm in your server logs that the engine's agent returned a 200 status on that URL.
- Check robots.txt for a block on OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot, Claude-User or Googlebot.
- Check rendering by fetching the page with a plain HTTP client; if the answer only appears after JavaScript runs, assume an agent may not see it.
- Check the logs for each agent's user-agent token and status code over 30 days.
- Run the prompt on each engine, signed out where possible, and record named, cited and linked.
- Repeat weekly, because a single reading proves nothing about the rate.
A page that is fetched but never cited has a content problem. A page that is never fetched has an access problem. The pillar on answer engine optimization covers what to do in each case.