# AI search

> For the complete documentation index, see [llms.txt](https://attensira.com/llms.txt).
> Every page on this site is also served as Markdown: append `.md` to any URL, or send `Accept: text/markdown`.

> AI search is search where a model reads pages and writes the answer instead of returning ten links. How it works and how engines differ, with our own data.

- Canonical page: https://attensira.com/glossary/ai-search
- Family: glossary
- Last updated: 2026-10-04
- Author: Karl-Gustav Kallasmaa (https://attensira.com/about)
- Claims as data: https://attensira.com/claims.json?slug=ai-search

**In short:** AI search is any search experience where a language model retrieves pages and writes the answer itself, instead of handing back a ranked list of links. ChatGPT search, Perplexity, Claude with web search, Gemini and Google's AI Overviews all work this way. The result is a short answer that names a few brands and links a few sources.

**ChatGPT, asked 2026-10-03:** "AI search optimization for seed-stage startups"

> AI search optimization (GEO) for seed-stage startups is about becoming a trusted source that AI engines cite—not just ranking on Google. … Avoid spending early budget on “AI hacks” or special markup promises. GEO is mostly strong SEO fundamentals plus making your expertise easy for AI systems to retrieve and cite.

Sources cited, in order: crowdstax.com, support.storeconnect.com, cituna.com, arxiv.org, redliolabs.com.

Attensira: not named.

_Source: attensira.com workspace, read over the Attensira MCP, as of 2026-10-04 (n=24 answers to this question in 30 days, Attensira named in 0)._

## Key facts

- Also known as: Answer engines, generative search
- Technique underneath: [Retrieval-augmented generation](/glossary/rag)
- Unit that competes: A passage, not a page
- What you can win: [A mention, a citation, or both](/glossary/source-citation)
- Related term: [AI Overviews](/glossary/ai-overview)
- In the docs: [Tracking › Models](https://docs.attensira.com/tracking/models)

## What does AI search look like from the website's side?

It looks like a short list of sources you either made or did not. Take one question from our own attensira.com workspace, "AI search optimization for seed-stage startups", asked of both engines every day. Over the 30 days to 4 October 2026, read over the Attensira MCP, here are the domains each engine attached most often:

|  | ChatGPT (12 answers, 429 citations) | Perplexity (12 answers, 114 citations) |
| --- | --- | --- |
| 1st | developers.google.com, 79 | topify.ai, 17 |
| 2nd | help.openai.com, 55 | averi.ai, 10 |
| 3rd | arxiv.org, 20 | stackmatix.com, 10 |

Same question, same days, no overlap in the top three. ChatGPT leaned on the platforms' own documentation and research papers; Perplexity leaned on vendor and agency guides. Attensira was named in none of the 24 answers. That is AI search from a website's side: the engine decides which few pages speak for a topic, and each engine decides differently.

## What is AI search?

It is search where a language model does the reading. A question goes in; a search step retrieves pages; the model reads passages from them and writes an answer, usually with a few source links. The technique underneath is [retrieval-augmented generation](/glossary/rag).

The output changes the economics of being found. A results page shows ten links and lets the reader choose. An AI answer names two or three options in prose. If you are not one of them, you were not considered. Pew Research Center found AI summaries on 18% of 68,879 Google searches it tracked in March 2025, and Google is only one of the places this now happens.

## How does AI search find your pages?

Through crawlers, and each operator runs several with different jobs. OpenAI [documents](https://developers.openai.com/api/docs/bots) OAI-SearchBot for ChatGPT search, GPTBot for model training and ChatGPT-User for fetches a person asks for. Perplexity documents PerplexityBot for its index and Perplexity-User for user-initiated fetches. Anthropic runs ClaudeBot, Claude-User and Claude-SearchBot. Google's AI Overviews draw on the ordinary search index Googlebot builds.

That split matters for [robots.txt](/glossary/robots-txt). Blocking a training crawler and blocking a search crawler are different decisions with different consequences. And user-initiated fetchers may not follow robots.txt at all: Perplexity says Perplexity-User generally ignores it, and OpenAI says robots.txt rules may not apply to ChatGPT-User. The [crawler directory](/bots) lists each agent and what it does.

## Which crawlers fetch pages for AI search?

On attensira.com, in the 30 days to 4 October 2026, three of the agents that fetch for search indexes made 7,643 of 143,601 AI-crawler requests. OpenAI's search agent led at about 6,481, a derived share of OpenAI's total; PerplexityBot and bingbot followed. Every row is the user agent the request claimed; none was IP-verified.

**Requests from search-index agents** (attensira.com, by the user agent each request claimed (none IP-verified); OAI-SearchBot derived as 86% of OpenAI's 7,536, 2026-09-05 to 2026-10-04, n=143,601 requests)

|  | Requests | Share |
| --- | --- | --- |
| OAI-SearchBot (~86% of OpenAI's 7,536) | 6,481 | 4.5% |
| PerplexityBot | 621 | 0.4% |
| bingbot | 541 | 0.4% |

_Source: attensira.com workspace, read over the Attensira MCP, as of 2026-10-04 (n=143,601 AI-crawler requests)._

## How is AI search different from Google search?

|  | Ordinary search | AI search |
| --- | --- | --- |
| Output | A ranked list of links | A written answer with a few links |
| Unit that competes | A page | A passage |
| Who reads the pages | The person | The model |
| What you can win | A position | A mention, a citation, or both |
| Can you see it in your analytics? | Clicks and impressions | Mostly not: you sample the answers |

The last row is the practical difference. Search engines report impressions. AI engines do not tell you when they mention you, so the only way to know is to ask the questions and read the answers, repeatedly. See [AI search performance](/glossary/ai-search-performance).

## What are the objections to taking AI search seriously?

**"AI search is a fad; Google still sends the traffic."** Google itself now answers with AI on a share of searches. The question is not which channel wins but whether you are named where buyers ask.

**"If we rank on Google we are covered."** Each AI engine has its own index and source preferences, as the seed-stage table shows. Ranking helps you be a candidate on Google's surfaces; it does not carry over automatically elsewhere.

**"Blocking AI crawlers protects our content."** It can also remove you from answers. Decide training and search separately.

**"We can't measure it."** You can, by sampling. Every rate needs its sample size beside it, and a change needs to clear the noise before it means anything.

## How does Attensira work with AI search?

Attensira reads the questions your buyers ask on the AI engines every day and records whether each answer names you, links you, and which sources decided it. Its crew then does the work: fixes your pages, earns placements on the sources each engine prefers, and ships each change after your approval. That is the service of an SEO agency at the price of a tool. [Which engines Attensira reads](/platform-visibility) lists each engine and what is read on it.

## FAQ

### What is AI search?

It is search where a language model retrieves pages and writes the answer, rather than returning a ranked list of links. ChatGPT search, Perplexity, Claude with web search, Gemini and Google's AI Overviews are examples. The reader gets prose naming a few options and a few source links.

### What separates AI search from ordinary search?

Where the answer is assembled. Ordinary search returns ranked links and leaves the reading to you. AI search reads the pages for you and writes one answer, so only the handful of sources and brands that make it into that answer are seen.

### Which crawler does an AI search product use?

Usually a dedicated search crawler, separate from its training crawler. OpenAI documents OAI-SearchBot for ChatGPT search and GPTBot for training. Perplexity documents PerplexityBot for search and Perplexity-User for user-initiated fetches. Google's AI Overviews use Googlebot's ordinary search index.

### Do user-triggered fetches obey robots.txt?

Not necessarily. Perplexity states that Perplexity-User generally ignores robots.txt because a user requested the fetch, and OpenAI states robots.txt rules may not apply to ChatGPT-User. Blocking a crawler does not reliably block a fetch a person asked for.

### Do all AI search engines use the same sources?

No. On one question in our own workspace, 'AI search optimization for seed-stage startups', ChatGPT's most-attached source over 30 days was developers.google.com and Perplexity's was topify.ai, and the two engines' top three sources did not overlap. Each engine has its own index and preferences, so visibility has to be measured per engine.