Every major assistant answers current questions the same basic way: it rewrites the question into one or more searches, retrieves pages from an index, and cites some of them. What differs is whose index. Google AI Overviews and AI Mode use Google Search. Copilot uses Bing. ChatGPT uses its own crawler plus third-party providers including Bing. Perplexity uses its own crawler. Claude has its own search crawler, and its web search has been reported to use Brave. So one site needs to be readable by five sets of crawlers, and ranking well in Google alone covers only part of it.
This page compares what each company documents about how its assistant finds and cites sources, plus the largest independent studies of what they actually cite. It's about mechanics. For what makes an assistant recommend one business over another, see How AI assistants pick which businesses to recommend.
The comparison
| Assistant | Where web results come from | Crawler to allow | Searches per question | Citation overlap with Google's top 10 | How you see the traffic |
|---|---|---|---|---|---|
| Google AI Overviews and AI Mode | The Google Search index | Googlebot (Google-Extended doesn't affect Search) | Several: Google says both "may use a 'query fan-out' technique" | AI Overviews: 76% (Jul 2025), 38% (Mar 2026) | Search Console Performance report, "Web" |
| ChatGPT | OAI-SearchBot's index plus "third-party search providers such as Bing" | OAI-SearchBot (GPTBot is training only) | Rewrites the question; count not published | 8.0% (in-text citations) | utm_source=chatgpt.com on referral links |
| Microsoft Copilot | The Bing index | Bingbot | Generates "a search query… different from the user's original prompt" | Not broken out; ChatGPT, Gemini and Copilot averaged 12% | Referrals from Copilot and Bing |
| Perplexity | Its own index, built by PerplexityBot | PerplexityBot (Perplexity-User for live fetches) | Not published | 28.6% | Referrals from perplexity.ai |
| Claude | Claude-SearchBot's index; TechCrunch reported evidence that web search uses Brave Search | Claude-SearchBot and Claude-User | Not published | Not studied | Referrals from claude.ai |
Sources: Google, OpenAI help, OpenAI publisher FAQ, Microsoft Learn, Perplexity crawlers, Anthropic crawlers, TechCrunch, Ahrefs, Mar 2026, Ahrefs, Aug 2025. Checked November 2026.
Google AI Overviews and AI Mode
Google's documentation is the most specific of the five. To be a supporting link, "a page must be indexed and eligible to be shown in Google Search with a snippet", and "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." No special schema, no llms.txt.
The important mechanism is query fan-out: both features "may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources" to build an answer. That's why ranking for the exact query matters less than it used to. In July 2025, Ahrefs found 76% of AI Overview citations ranked in Google's top 10 for the query. Its March 2026 rerun, on 4 million cited URLs, found 38%, with 31% ranking outside the top 100 entirely. Ahrefs suggests AI Overviews now lean more on pages ranking for the fan-out searches, and notes its method improved between studies, so the two figures aren't exactly like for like.
Controls: nosnippet, data-nosnippet, max-snippet and noindex limit what appears. Blocking Googlebot removes you from Search entirely. Traffic: Google says clicks from AI features are counted in Search Console's Performance report under "Web", not broken out separately.
ChatGPT
OpenAI says ChatGPT "will automatically search the web if your question might benefit from information on the web", and that it "may share disassociated search queries with third-party search providers such as Bing", along with a general location inferred from IP. Its publisher FAQ says "any public website can appear in ChatGPT search", provided it isn't blocking OAI-SearchBot. Blocking GPTBot only opts you out of training.
What ChatGPT cites looks least like a search results page of any assistant. Across 15,000 long-tail prompts, Ahrefs found only 8.0% of ChatGPT's in-text citations ranked in Google's top 10 for the prompt, and 8.1% in Bing's top 10. ChatGPT rewrites questions into its own searches, so the page it cites often ranks for a query you never targeted.
Live fetches matter too. When someone asks ChatGPT about a specific business or URL, ChatGPT-User loads the page there and then. In our Observatory, ChatGPT-User was the single most active AI agent on small-business sites, at 28.1% of AI fetches, and OpenAI agents together made up 53%. Traffic: OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs.
Microsoft Copilot
Microsoft's documentation describes the steps plainly. Copilot "parse[s] the user's prompt and identifies terms where information from the web would improve the quality of the response", then "generates a search query that it sends to the Bing search service". That query "consists of a few words informed by the user's prompt", not the prompt itself. Bing's results are used "to compose the response returned to the user", and Copilot Chat shows the exact queries it sent in its citations.
So Copilot visibility is Bing visibility: allow Bingbot, and check your pages in Bing Webmaster Tools, not only Google Search Console. Ahrefs found Copilot had the highest overlap with Bing's results of the assistants it studied. There's no separate Copilot crawler or robots.txt token.
Perplexity
Perplexity says it "search[es] the internet in real time" and that every answer includes "numbered footnotes linking to the original sources". Its own crawler, PerplexityBot, builds the index it uses to "surface and link websites in search results"; Perplexity says it doesn't use it to train foundation models. Perplexity-User fetches pages live for a user's question and, per Perplexity, generally ignores robots.txt because a person asked.
Of the assistants Ahrefs studied, Perplexity's citations were closest to Google's: 28.6% ranked in Google's top 10 for the prompt. It also tends to cite many sources per answer: Surfer's October 2026 study counted 10.3 sources per answer in Perplexity's app and 19.5 through its API.
One caveat: in August 2025, Cloudflare accused Perplexity of using undeclared crawlers to fetch sites that had blocked PerplexityBot. Perplexity disputed it. For a small business that wants to be cited, the practical point is simply to allow PerplexityBot.
Claude
Anthropic says Claude "invokes a search tool to inform and ground its generated responses" when a question benefits from current information, that "every response includes citations", and that it may use your IP-inferred location for local questions. It documents three agents: Claude-SearchBot, which crawls "to improve the quality of Claude's search results"; Claude-User, for live fetches when a user's question needs a page; and ClaudeBot, for training. Blocking the first two, Anthropic says, may reduce your visibility in Claude's answers.
Anthropic doesn't name its text search provider on its support pages (it names Bing for image results). In March 2025, TechCrunch reported evidence that it's Brave Search: Brave appeared on Anthropic's subprocessor list, a developer found a parameter named BraveSearchParams, and some searches returned identical citations. Treat that as reported, not confirmed. We found no large published study of Claude's citations against Google or Bing.
What this means for one website
- Allow the search and user agents of all five. Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot and Claude-User. You can block the training-only agents (GPTBot, ClaudeBot, Google-Extended) without losing any of the above; our robots.txt guide has the exact rules, and the crawler registry lists every agent.
- Be in Bing, not just Google. Copilot runs on Bing, and OpenAI names Bing as one of ChatGPT's providers.
- Write for the fan-out, not the keyword. Because Google, ChatGPT and Copilot all rewrite questions into their own searches, pages that answer the related questions (prices, areas, comparisons, how-tos) get more chances to be pulled in.
- Put the facts in the HTML. None of these systems promises to run your JavaScript; most AI crawlers don't. Test your page with the AI access checker.
- Track by engine, not in aggregate. With citation overlap this low, a tool that tracks one or two assistants can miss most of the picture. Our tool reviews list which engines each tracker covers.
Questions people ask
Does ChatGPT use Google or Bing?
OpenAI says ChatGPT search may share queries with third-party search providers such as Bing, alongside sites surfaced by its own crawler, OAI-SearchBot. Ahrefs found ChatGPT's citations overlap Google's top 10 about 8% of the time and Bing's about 8%, so ranking in either is no guarantee.
Do AI Overviews only cite pages in Google's top 10?
No. Ahrefs found 76% of AI Overview citations ranked in the top 10 in July 2025, but only 38% in March 2026, which it links to Google's query fan-out: AI Overviews and AI Mode run several related searches and cite from all of them.
Which AI assistant is closest to Google rankings?
Perplexity, in Ahrefs' August 2025 study: 28.6% of its cited URLs ranked in Google's top 10 for the prompt, against 8% for ChatGPT, and 12% on average across ChatGPT, Gemini and Copilot.
How do I see traffic from ChatGPT?
OpenAI says ChatGPT adds utm_source=chatgpt.com to referral links, so visits show up in analytics under that source. Google says AI Overviews and AI Mode clicks are counted in Search Console's Performance report under the Web search type.