The short answer
In part, yes. Each assistant finds sources its own way. Google's AI features use Google's search index, ChatGPT search draws on search providers and partner content, and Claude and Perplexity run their own search tools with citations. None publishes exact rules for choosing sources, so the same question can give different answers.
In part, yes. Each assistant has its own way of finding and citing pages, so the same question can bring different answers from each. Here is what each company says about how its assistant works, and what none of them says.
How does Google’s AI search find its sources?
Gemini is Google’s assistant. Google documents how its AI features in Search work, which is what we describe here, and says less about the Gemini app itself, so we do not assume the two behave identically.
For AI Overviews and AI Mode, Google says answers are grounded in pages from its own Search index. Its guide says core ranking systems pull relevant, current pages, the AI reviews them, and the response links to the pages that support it. It also uses “query fan-out”, sending several related searches to gather more material. Google adds that the two features may use different models, so the links they show can differ.
How does ChatGPT search find its sources?
OpenAI says ChatGPT search draws on third-party search providers and on content from partners, and shows links to its sources. OpenAI also runs a crawler called OAI-SearchBot for its search features, and says that sites blocking it will not appear in ChatGPT search answers, though they can still show up as plain links.
How does Claude find its sources?
According to Anthropic’s help centre, Claude uses a search tool when a question benefits from current information, grounds its answer in live web content and gives citations so you can check the sources. Anthropic runs Claude-SearchBot, which it describes as crawling the web to improve search results, and says blocking it can lower a site’s visibility in them.
How does Perplexity find its sources?
Perplexity’s help centre describes it as reading your question, searching the web and answering with numbered citations that link to the original pages. Its crawler, PerplexityBot, is described as helping sites appear as linked results, and the company recommends allowing it in robots.txt.
Why do the answers still differ?
As far as we can find, none of the companies publishes exactly how it decides which pages to rely on, so we cannot rank their sources. What we can see are the reasons answers differ.
- The model. Each company uses its own, and each writes the answer differently.
- The search step. Some use their own crawlers and index, some use search partners, and some use a mix.
- The questions they ask. Several break your question into smaller searches, and the sub-searches they choose are not the same.
- What they show. Some lead with a short answer and links, some with a longer report.
They do have things in common. All four cite pages, so the pages they cite need to be readable and trustworthy to a person who clicks. None can use what it cannot read, so a blocked page, a page behind a login or one locked inside images is hard for any of them to use. And none offers a way to buy a place in its answers. Separate fetchers, such as ChatGPT-User, Claude-User and Perplexity-User, visit a page when a person asks a question, and their companies describe different rules for how robots.txt applies to them.
What does this mean for my business?
In our view, you do not need a separate strategy for each assistant. You also should not assume a mention in one is a mention in all.
The sensible approach is the same foundation for every one of them. Keep your site indexed and open to search crawlers, write each page to answer a clear question in plain words with the key facts in text, and keep your business details consistent across your website and listings. Our answer on the difference between SEO and AI search optimisation explains why that foundation overlaps so heavily.
Treat what you find as a log, not a scoreboard. Answers change between assistants and between days. The useful part is spotting facts about your business that are missing or wrong, because those are the ones you can fix at the source.
Sources
Where a claim depends on how Google or another company says something works, we link to their own words so you can check it.
- AI features and your website, Google Search Central. Supports that Google's AI Overviews and AI Mode use query fan-out and show supporting links.
- Google's guide to optimizing for generative AI features on Google Search, Google Search Central. Supports that Google's AI features ground answers in pages from the Search index.
- Introducing ChatGPT search, OpenAI. Supports that ChatGPT search draws on third-party search providers and partner content, and links to its sources.
- Overview of OpenAI crawlers, OpenAI. Supports what OAI-SearchBot, GPTBot and ChatGPT-User do.
- Enable and use web search, Claude Help Center, Anthropic. Supports that Claude searches the web when a question benefits from current information and cites its sources.
- Claude web search now available globally on all plans, Anthropic. Supports that Claude's web search is available on all plans.
- What is ClaudeBot and the Claude crawlers, Anthropic. Supports what Claude-SearchBot and Claude-User do.
- How does Perplexity work?, Perplexity Help Center. Supports that Perplexity searches the web and attaches numbered citations to its answers.
- PerplexityBot and Perplexity-User, Perplexity. Supports what PerplexityBot and Perplexity-User do.
About this answer. We are an independent studio and this is our own research, experience and opinion, shared as general guidance. It is not legal, financial or other professional advice, and what works for one business may not work for another. Where the law or a regulator is involved, check the official guidance for your situation or ask a qualified adviser. Our sources are listed above.