AI Search 101
What is AI search?
AI search is search that returns a written answer instead of a list of links. A language model reads across many sources, sometimes retrieving pages live, and composes one response. The reader gets a conclusion rather than a set of candidates, and often never visits the sources it was built from.
The basics
| Also called | Generative search, answer engines, conversational search |
| Main examples | ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews |
| Built on | Large language models combined with live retrieval |
| Output | A composed answer, sometimes with citations, sometimes with only unattributed mentions |
| Optimised through | Answer engine optimisation |
How does AI search differ from traditional search?
A traditional search engine is a matching system. It holds an index of pages, scores them against a query, and returns the ones it judges most relevant in order. The work of deciding is left to the reader, who opens three or four results and forms a view.
AI search moves that work to the machine. The system reads across sources and writes a single answer. The reader receives a conclusion, already synthesised, with the reasoning and the sources mostly out of sight.
This changes what visibility means. In traditional search the goal is a position: rank third and a predictable share of clicks follows. In AI search there is no position. There is either a mention in the answer or nothing at all, and the mention may not carry a link.
How does an AI search system actually work?
Two mechanisms are at play, and separating them explains most of the confusing behaviour.
Training. The model learned patterns from an enormous body of text during training. Anything it read there is baked into its weights, unattributed and undated. When it answers from training alone, it is drawing on a compressed impression of what the internet said, not looking anything up.
Retrieval. Most systems now also fetch live pages at the moment of the question, read them, and write the answer from what they find. This is retrieval-augmented generation. It is why an assistant can discuss something that happened yesterday, and why citations appear at all.
A single answer usually blends both. The model's trained impression shapes the framing and the vocabulary; the retrieved pages supply the specifics. A company can be described confidently from stale training data even when its current site was never fetched.
Which sources does AI search actually use?
Rarely the ones a company would choose. Retrieval favours pages that answer the question directly, load quickly, and are structured clearly. In practice that means a mixture of trade press, comparison and listicle pages, documentation, forums such as Reddit, directories, and review sites.
A company's own website is one input among many, and frequently not the most quoted one. This is uncomfortable but it is also the mechanism: the systems are built to describe a subject from independent sources rather than from its own marketing.
How much traffic does AI search actually send?
Little, so far. For most business-to-business companies, referrals from AI assistants sit around one percent of sessions, occasionally two or three in technical categories where the audience lives in these tools.
Judging the channel on that number is a mistake, for two reasons. The first is direction: the share is rising steadily while traditional organic clicks fall, partly because Google now answers many questions above its own results. The second is that traffic is the wrong measure. A buyer who reads a description of a company in ChatGPT, forms an impression and later arrives through a direct visit or a branded search shows up in analytics as direct or branded, not as AI. The influence is real and mostly unattributed.
Can a company influence what AI search says about it?
Partly, and not by the means most people reach for first.
What works is being clear and consistent about what the company is, in a form machines can extract, across every source a model is likely to read. That means an unambiguous definition on the company's own pages, the same description repeated on independent sources, structured data that names the entity properly, and content that answers the questions buyers actually ask rather than the keywords a tool suggested.
What does not work is the tactic that worked in search for twenty years: producing volume against keywords. A model does not count pages. It forms an impression of a subject, and twelve vague pages make that impression vaguer.
How AI search is used in marketing
Production. Marketing teams write the reference material buyers ask about, structured so it can be quoted: a direct answer near the top, one question per page, plain declarative sentences.
Analysis. Teams monitor what assistants say when asked about their category and their company, tracking mentions, sentiment and which competitors appear alongside. This is a new measurement discipline and most organisations have no baseline for it.
Distribution. Because independent sources carry disproportionate weight, effort moves toward the places models read: trade publications, comparison sites, documentation, and communities where the category is discussed.
How this impacts your business
A language model does not look a company up in a register. It composes a description from everything it read during training and whatever it retrieves at the moment of the question. Trade press, directories, forums, competitor comparison pages, and somewhere in that mix the company's own website too.
So the description a buyer sees is assembled, not published. No company controls it, but every company supplies a share of the material it is built from.
The traffic impact is still small, roughly one percent of sessions for most companies. The influence is not small, because the reader is often a buyer forming a first impression of what a company does and who it serves, before any page gets a click.
Checking takes ten seconds. Type the company name into ChatGPT and read the answer.
Frequently asked questions
Is AI search the same as generative search?
The terms are used interchangeably. Generative search emphasises that the answer is written rather than retrieved; AI search is the broader label covering assistants, answer engines and generated summaries inside traditional search results.
Does AI search use Google's index?
Some systems do, directly or through partnerships. Others run their own crawlers, and several use more than one source. This is why the same question can produce different sources in different assistants.
Will AI search replace traditional search engines?
It is displacing a share of queries rather than replacing the category. Google has responded by generating answers above its own results, which means the shift is happening inside traditional search as well as outside it.
How do I find out whether my company appears in AI search?
Ask the assistants directly. Type the company name, then the category question a buyer would ask, and read what comes back. For anything beyond spot checks, dedicated monitoring tools track mentions across models over time.
Do AI search answers link to sources?
Sometimes. Retrieval-based answers usually cite. Answers drawn from training data usually do not, which is why a company can be described in detail with no link and no traffic.
Related definitions
- What is answer engine optimisation?
- What are AI Overviews?
- What is an AI citation?
- What is a large language model?
- AI Search 101
Written by Lari Numminen, Generate More. Also available in Finnish. Updated 31 August 2026.