An AI answer engine takes your question, retrieves relevant pages from the live web, synthesises them into a single written answer and cites some of what it read. ChatGPT, Google Gemini, Claude, Perplexity, Google AI Overviews and AI Mode, and Microsoft Copilot all work this way when web access is switched on.
That is the whole definition. The interesting word is "some". Every engine reads far more than it shows, and each one shows different things. We ran 240 identical buyer-intent queries across four engines in July 2026 to measure exactly what each cites, and the differences are large enough to change where you spend your marketing budget.
This post is about engine mechanics and citation behaviour specifically. For the broader concept and how it differs from classic search results pages, read what AI search is.
The main AI answer engines
Six engines account for most AI-generated answers today. They differ less in how they write and more in how they retrieve and what they disclose.
| Engine | Company | Retrieval | Citation behaviour |
|---|---|---|---|
| ChatGPT | OpenAI | Built-in web search | Often hides citations in the consumer app; returns full source lists via API |
| Gemini | Google Search grounding | Often hides citations; API sources mostly wrapped in Google redirect URLs | |
| Claude | Anthropic | Web search tool | Exposes its citations |
| Perplexity | Perplexity AI | Search-first by design | Exposes numbered citations on every answer |
| Google AI Overviews / AI Mode | Google's own index | Shows supporting links alongside the answer | |
| Copilot | Microsoft | Bing retrieval | Shows inline source links |
The hide-versus-expose split matters. As a rule, Perplexity and Claude expose their citations while ChatGPT and Gemini often do not, so what a user sees in the app understates what the engine actually read.
Retrieval versus training memory
An answer engine has two places to get an answer from. Training memory is what the model absorbed before its knowledge cutoff. It is fast, free and cannot cite anything, because there is no live source behind it. Retrieval is a live web search: the engine fetches pages, reads them and synthesises an answer it can attribute.
Every major engine blends the two, and the blend shifts per query. That is one reason answers are so unstable. SparkToro found the same AI query changes roughly 70% of the time. In our own repeat runs, the top recommended brand changed between identical consecutive runs 44% of the time on Gemini, 43% on Perplexity, 35% on ChatGPT and 28% on Claude. A single manual check tells you almost nothing, which is why spot-checking fails as a monitoring method.
What each engine actually cites, measured
On 13 July 2026 we ran 20 buyer-intent prompts, three identical runs each, through the APIs of four engines: gpt-5-mini, gemini-2.5-flash, claude-haiku-4-5 and Perplexity sonar. That is 240 requested answers; Gemini completed 54 of its 60 calls and the rest failed at the API. Every completed answer returned sources.
One caveat before the numbers: this is API behaviour. The consumer apps may surface fewer sources, or none at all.
Sources per answer
Average sources cited per answer, by engine
| Engine | Avg sources per answer | Total citations | Distinct domains | Top 3 domains' share |
|---|---|---|---|---|
| ChatGPT | 15.0 | 900 | 445 | 7.6% |
| Gemini | 10.7 | 580 | 109 | 76.4%* |
| Claude | 8.8 | 529 | 194 | 6.0% |
| Perplexity | 8.3 | 498 | 142 | 25.1% |
*Gemini's concentration is an artefact of masking, explained below.
### ChatGPT reads widest
Want to see this in action?
See how every major AI model talks about your brand. Free to start.
ChatGPT averaged 15.0 sources per answer and drew its 900 citations from 445 distinct domains, with the top three domains taking only 7.6% of citations. It is the broadest reader of the four. Getting cited by ChatGPT is a long-tail game: hundreds of mid-sized review sites, comparison pages and niche publications all feed it.
### Claude is broad but lighter
Claude cited 8.8 sources per answer, 529 citations across 194 domains, and its top three domains held just 6% of the total. Like ChatGPT, no single publisher dominates.
### Perplexity concentrates on Reddit and YouTube
Perplexity is the most concentrated open engine. Its top three domains took 25.1% of all citations, led by Reddit with 71 citations (14% of its total) and YouTube with 40 (8%). That is consistent with Semrush's finding that Reddit accounts for 40.1% of all AI citations, the single most-cited source. If Perplexity matters to your buyers, community threads and video are not optional channels.
### Gemini masks its sources
Gemini wrapped 429 of its 580 source URLs, 74%, in vertexaisearch.cloud.google.com grounding redirects that hide the real destination. Its 76.4% top-3 share in the table is mostly that one redirect domain. In practice you often cannot tell which publisher Gemini leaned on, which makes it the hardest engine to audit and the strongest argument for tracking brand mentions rather than link referrals.
### The one constant: Forbes
Forbes was the only domain to appear in all four engines' top-cited lists: 40 citations on ChatGPT, 14 on Perplexity, 12 on Claude and 7 on Gemini. Established editorial coverage still travels across every engine.
What this means for your brand
An answer engine gives a buyer roughly five brand options per answer, based on the average of 4.8 to 5.2 brands per answer across our four engines. There is no page two. If you are not in the synthesised answer or its citations, you are invisible for that query, and Semrush found 62% of AI citations never name the brand at all, so even being a source does not guarantee being seen.
The levers are also different from classic SEO. Ahrefs found AI visibility correlates most with third-party mentions and video, not on-page work. Combined with our data, the playbook is clear: earn coverage on the broad mid-tail that ChatGPT and Claude read, show up in the Reddit threads and YouTube videos Perplexity favours, and pursue the editorial tier that all four engines cite. We cover the tactics in how to get cited by AI.
Because answers change between identical runs, you cannot verify any of this manually. Honeyb, our AI visibility platform, runs your buyers' prompts across engines on a schedule and shows where you appear, who gets cited instead and how that shifts over time.
Start with a baseline: run a free AI visibility check to see which answer engines mention your brand today.
