Blog · What counts as AI search
What Is AI Search? Which Products Actually Count in 2026
Priya Bothra · September 22, 2026
AI search is any search experience where a system retrieves information from the web or a search index at the moment you ask, uses a language model to write an answer, and points you to the sources it used. By that definition, the main AI search products in 2026 are Google AI Overviews and AI Mode, ChatGPT search, Perplexity, Claude with web search, Microsoft Copilot and Copilot Search in Bing, the Gemini app, and a new group of AI browsers such as ChatGPT Atlas and Perplexity Comet.
The boundary matters because it decides which crawlers you allow, which platforms you monitor and which traffic you attribute. A chatbot answering purely from its training data behaves differently from one that searches the live web, and the work needed to appear in each is different. This guide gives a practical three-part test for what counts as AI search, a category map of current products, and the edge cases that confuse most teams.
This article focuses on the products themselves. If you want to understand how people phrase their questions to these tools, see our guide to what AI search queries are.
A working definition of AI search
AI search is the combination of retrieval, generation and attribution in a single experience. Each part does a specific job:
- Retrieval: the product fetches documents from a search index or the live web when the question is asked, rather than relying only on what the model learned during training.
- Generation: a large language model reads what was retrieved and writes a synthesized answer in natural language.
- Attribution: the answer links to or names the sources it drew from, so the user can verify or click through.
Google's own description of AI Mode fits this pattern. According to TechCrunch's report on the March 5, 2025 launch, AI Mode uses "query fan-out" to run multiple related searches across data sources and then synthesizes the results into a response with source links. Google's documentation on AI features describes the same fan-out technique for both AI Overviews and AI Mode.
The three-question test
We recommend this simple framework when deciding whether a product belongs in your AI search program. It is a practical rule of thumb, not an industry standard.
- Does it fetch content at answer time? If yes, your pages can be selected today, and crawl access matters.
- Does it write the answer instead of listing links? If yes, the model is summarizing and framing your brand, not just ranking your page.
- Does it show or link sources? If yes, citations and referral traffic are possible and measurable.
A product that passes all three is core AI search. A product that passes only the second, such as a chatbot with browsing turned off, still shapes perceptions of your brand but is not search in the retrieval sense.
Which products count as AI search?
AI search products fall into four categories: AI layers inside traditional search engines, AI assistants with web search, AI-native answer engines, and AI browsers and agents. The table below summarizes the main products as of September 2026.
| Category | Product | Retrieves live web? | Cites sources? | Crawler or index it depends on |
|---|---|---|---|---|
| AI inside search engines | Google AI Overviews | Yes | Yes | Googlebot and Google's index |
| AI inside search engines | Google AI Mode | Yes | Yes | Googlebot and Google's index |
| AI inside search engines | Copilot Search in Bing | Yes | Yes | Bing's index |
| AI assistants with search | ChatGPT search | Yes | Yes | OAI-SearchBot plus third-party search providers |
| AI assistants with search | Claude with web search | Yes | Yes | Claude-SearchBot, Claude-User |
| AI assistants with search | Gemini app | Can ground in Google Search | Often | Google's index |
| AI assistants with search | Microsoft Copilot | Can ground in Bing | Often | Bing's index |
| AI-native answer engine | Perplexity | Yes | Yes | PerplexityBot, Perplexity-User |
| AI browsers and agents | ChatGPT Atlas, Perplexity Comet | Yes, while browsing | Varies by task | The parent product's systems plus the pages the user opens |
AI layers inside traditional search engines
AI layers inside search engines are generated answers placed on top of or beside a classic results page. They matter most because they reach the largest audience by default.
Google AI Overviews launched to everyone in the U.S. on May 14, 2024, and Google said at the time it expected to bring them to "over a billion people by the end of the year" (Google). Ahrefs' 2026 benchmark found AI Overviews on 21% of the keywords it studied, appearing on 9.5% of one-word queries versus 46.4% of queries with seven or more words.
Google AI Mode is a full conversational search tab. It started as a Search Labs experiment in March 2025, and at Google I/O in May 2026 Google said AI Mode had passed 1 billion monthly users, with queries more than doubling every quarter.
Copilot Search in Bing is Microsoft's equivalent. Search Engine Land reported its official launch on April 4, 2025, describing curated answers with "clearly cited sources."
For all three, the ranking foundation is the search engine's own index. Google states there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" beyond being eligible for regular Search (Google Search Central).
AI assistants with web search
AI assistants with web search are chat products that decide when to search the web during a conversation and then cite what they found.
ChatGPT search launched on October 31, 2024 and draws on third-party search providers and content from news partners. Whether your site can appear depends on OpenAI's search crawler. OpenAI's bot documentation states: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." ChatGPT also adds utm_source=chatgpt.com to referral links, according to OpenAI's publisher FAQ, which makes it one of the easier AI search sources to attribute.
Claude web search arrived in March 2025, initially as a preview for paid U.S. users, with Anthropic saying Claude "provides direct citations to allow users to verify sources" (Search Engine Journal). Anthropic runs separate bots for training, search indexing and user-initiated fetches, and each needs its own robots.txt rule (Search Engine Land).
The Gemini app and Microsoft Copilot are assistants that can ground answers in their parent company's search index. Because they can ground in Google and Bing, your standing in those indexes is a sensible first thing to check for both assistants.
AI-native answer engines
An AI-native answer engine is a product built from the start around search plus generated, cited answers, with no legacy results page to protect. Perplexity is the clearest example. Its crawler documentation explains that PerplexityBot surfaces and links websites in Perplexity results and is not used to train foundation models, while Perplexity-User handles user-initiated requests and generally ignores robots.txt.
Answer engines also cite differently from Google. In Profound's analysis of 680 million citations, Reddit was Perplexity's most cited domain at 6.6% of citations, while Wikipedia led for ChatGPT at 7.8%.
AI browsers and search agents
AI browsers and search agents are the newest category. They put an assistant inside the browser or let it act on the user's behalf, so "search" can happen while someone reads, compares or buys.
- ChatGPT Atlas launched on October 21, 2025 as "the browser with ChatGPT built in," with an agent mode that can research and complete tasks while the user browses.
- Perplexity Comet became free worldwide in October 2025, according to CNBC.
- Google announced search agents and a new AI search box that accepts text, images, files, videos and Chrome tabs at I/O 2026.
Agents blur the line between searching and doing. When an assistant can check out through protocols such as OpenAI's Agentic Commerce Protocol or Google's Universal Commerce Protocol, being selected becomes a transaction, not just a mention.
Edge cases: what sits on the boundary
Several products look like AI search but fail one or more parts of the test. Classifying them correctly keeps your program focused.
Chatbots answering from training data only. When an assistant answers without searching, nothing is retrieved and nothing is cited. Your content can still influence the answer, but only through what the model learned, which you cannot update quickly. This is brand perception work, not search optimization in the retrieval sense.
Visual and voice search. Google Lens handled more than 100 billion visual searches in 2025 by the time of I/O that May, and the category keeps growing. Visual and voice products count as AI search when they retrieve and summarize, but they reward different assets. Our guide to multimodal AI search covers them in detail.
Vertical AI search. Assistants built into specific platforms, such as travel, legal research or developer tools, retrieve from narrow datasets. They count as AI search within their domain. See vertical AI search for how they differ from general engines.
Classic featured snippets. Featured snippets extract a passage but do not generate new text, so they belong to traditional answer engine optimization rather than AI search.
Why the classification changes what you do
Each category needs a different mix of access, content and measurement. Treating "AI search" as one channel leads to blind spots.
| If the product is... | Access priority | Content priority | Measurement approach |
|---|---|---|---|
| AI inside a search engine | Standard search crawling and indexing | Strong SEO coverage of the topic and its sub-questions | Search Console AI impressions plus sampled prompts |
| AI assistant with search | Allow each vendor's search bot | Clear, citable facts and third-party mentions | Sampled prompts, referral UTMs, citations |
| AI-native answer engine | Allow the answer engine's crawler | Fresh, specific, well-sourced pages and community presence | Sampled prompts, citation sources |
| AI browser or agent | Pages usable by agents, not just readable | Complete product, pricing and policy details | Agent visits in logs, assisted conversions |
Three practical consequences follow from this classification.
Crawler policy must be set per product. Training bots and search bots are separate. OpenAI separates GPTBot from OAI-SearchBot, and Anthropic separates ClaudeBot from Claude-SearchBot. Blocking training while allowing search is a legitimate choice, but blocking everything labeled "AI" can remove you from AI search results. Our guide to how AI search works from crawling to citation walks through each stage.
Measurement must be sampled, not ranked. AI answers vary between runs. SparkToro and Gumshoe found less than a 1 in 100 chance that the same prompt returned the same brand list twice. Track visibility percentage across repeated runs for each product, not a single position.
Traffic expectations should differ by category. Pew Research found users clicked a result on 8% of Google visits with an AI summary, versus 15% without (Pew Research Center). Ahrefs estimates Google still sends about 190 times more traffic to websites than ChatGPT (Ahrefs via Business Wire). AI search influence often shows up as brand demand and assisted conversions rather than direct clicks.
A hypothetical example
Consider a hypothetical B2B payroll startup that checks ChatGPT by hand once a month and calls that "tracking AI search." The three-question test reveals gaps:
- Its robots.txt blocks every bot with "AI" or "GPT" in the name, including OAI-SearchBot, so ChatGPT search cannot show its pages.
- It never checks Google AI Overviews, even though its buyers ask long, specific questions, the query type most likely to trigger one.
- It ignores Perplexity, where comparison threads on community forums describe an outdated pricing model.
After reclassifying, the team allows search bots while keeping training bots blocked, adds AI Overviews and Perplexity to a monthly sampled prompt set, and fixes the outdated third-party descriptions. None of these steps needs special markup or new files.
Common mistakes when defining AI search
Equating AI search with ChatGPT. ChatGPT is one product. Google's AI features reach far more people, and Perplexity and Claude cite from different source mixes.
Blocking all AI crawlers by default. A blanket block removes you from AI search products that respect robots.txt while doing little about user-initiated fetchers that may not.
Counting training-only chatbot answers as search results. These answers cannot be influenced on a weekly cycle, so mixing them with retrieval-based results muddies reporting.
Ignoring AI browsers and agents. Agents read pricing pages, policies and product details to complete tasks. Pages that hide key facts behind scripts or vague copy may be skipped.
Chasing special files. Google has said you do not need new machine-readable or AI text files for its AI features, and a Google Search Advocate called llms.txt "purely speculative for now" in June 2026.
How Bob Builds AI helps
Bob Builds AI is an AEO and GEO platform and agency that tracks brand appearance across ChatGPT, Gemini, Claude, Perplexity, Copilot and Google AI Overviews and AI Mode, measuring the real chat and search interfaces rather than raw model APIs. Visibility Monitoring reports visibility rate, citation rate, competitor positioning and citation sources for each product, which fits the per-category approach described above. Agent Analytics shows how AI crawlers access your site, so you can check that search bots are reaching the pages you want cited. Prompt Research helps you build the prompt set to sample across each AI search product.
FAQ
What is AI search in simple terms?
AI search is a search experience where a system looks up information from the web or a search index when you ask a question, uses a language model to write an answer, and shows the sources it used. Google AI Overviews, Google AI Mode, ChatGPT search, Perplexity, Claude with web search and Copilot Search in Bing are the main examples. The key difference from classic search is that you get a written answer instead of a list of links.
Is ChatGPT an AI search engine?
ChatGPT counts as AI search when it uses its search feature, which OpenAI launched on October 31, 2024. In that mode it retrieves web content, writes an answer and links to sources. When it answers purely from training data without searching, it behaves as a chatbot rather than a search engine. Sites that block OAI-SearchBot will not be shown in ChatGPT search answers, according to OpenAI's documentation.
Are Google AI Overviews and AI Mode the same thing?
No. AI Overviews are generated summaries that appear at the top of a normal Google results page for some queries. AI Mode is a separate conversational search tab designed for complex, multi-part questions and follow-ups. Both use Google's index and a technique Google calls query fan-out, and Google says neither requires special optimization beyond being eligible to appear in regular Search.
Is Perplexity a search engine or a chatbot?
Perplexity is best described as an AI-native answer engine. It searches the web for each question, writes a cited answer and lets users ask follow-ups. Its crawler, PerplexityBot, surfaces and links websites in Perplexity results and is not used to train foundation models. Because it cites heavily from community sources such as Reddit, off-site presence matters for visibility there.
Do AI browsers like ChatGPT Atlas count as AI search?
AI browsers count as AI search when their built-in assistant retrieves and summarizes web content for the user. They also go further, because agent modes can research, compare and complete tasks on the user's behalf. For brands, this means pages need to be usable by agents, with complete pricing, product and policy details that an assistant can read and act on without guessing.
Should I block AI crawlers to protect my content?
It depends on your goal. Most AI companies run separate crawlers for model training and for search, so you can block training bots while allowing search bots. OpenAI separates GPTBot from OAI-SearchBot, and Anthropic separates ClaudeBot from Claude-SearchBot. Blocking every AI-related bot can remove your site from AI search answers while doing little about user-initiated fetchers, some of which may not follow robots.txt.
Which AI search products should a B2B brand track first?
Start with the products your buyers already use. For most B2B brands that means Google AI Overviews and AI Mode for reach, ChatGPT for research-heavy buyers, and Perplexity for technical audiences. Gartner found 45% of B2B buyers used generative AI in a recent purchase, mainly to research vendors. Add Claude, Copilot and Gemini as your prompt set and resources grow.
How is AI search different from AI search queries?
AI search refers to the products and systems that retrieve, generate and cite answers. AI search queries are the questions people type or speak into those products, which tend to be longer and more conversational than classic keywords. Understanding the products tells you where to be visible, while understanding the queries tells you what questions your content needs to answer.
Conclusion
AI search is not one channel. It is a family of products that retrieve web content, write an answer and cite sources, spread across four categories: AI layers in search engines, assistants with web search, AI-native answer engines, and AI browsers and agents. Chatbots answering from training data, featured snippets and in-chat apps sit near the boundary and deserve separate treatment.
The practical implication is that each category needs its own crawler rule, content priority and measurement method. A useful next step is to list the AI search products your buyers use, check robots.txt for each vendor's search bot, and run the same ten buyer prompts across those products to see where you appear. If you want to run that check continuously across models, Bob Builds AI can help you monitor visibility and act on the gaps it finds.