Agentic search

Agentic search is a form of AI search in which an agent carries out a multi-step task for the user: it plans, runs many searches, reads and compares pages, then reports back or takes an action such as a booking. ChatGPT deep research and Deep Search in Google AI Mode are examples.

Why agentic search matters

Agentic search hands the research itself to the machine. OpenAI says its deep research agent will "find, analyze, and synthesize hundreds of online sources to create a comprehensive report," in 5 to 30 minutes. Google says Deep Search in AI Mode "can issue hundreds of searches." The person reads only the conclusion: a shortlist, a recommendation, a link to book. The pages and brands the agent set aside along the way never reach them, so a brand is judged on what the agent could find, read and verify.

How agentic search works

Google defines AI agents as "autonomous systems that can perform tasks on behalf of people, such as booking a reservation or comparing product specifications." In search, the task usually runs in four steps:

Plan. The agent breaks the goal into questions and decides what to look up first.
Search, many times. It issues far more queries than a person would, an extended form of query fan-out, and changes course based on what it finds.
Read pages. Browser agents read a site through screenshots, raw HTML and the accessibility tree, according to Google's web.dev guide.
Report or act. Deep research returns a cited report. Google AI Mode can check live restaurant availability across booking platforms and link straight to the booking page. In May 2026, Google announced information agents that run in the background, monitor the web for a question and send updates.

Example: a vendor comparison done by an agent

A finance lead asks ChatGPT deep research: "Compare the five most used accounting tools for UK freelancers on price, bank feeds and Making Tax Digital support, summarize what reviewers complain about, and recommend one." For the next half hour or less, the agent runs searches, opens pricing pages, review sites and forums, and builds a report with a citation for each claim. A vendor that shows its prices only in an image or behind a sales form gives the agent nothing to quote.

How to prepare for agentic search

Rank for the sub-questions. Agents still find pages through search, so the pages that rank for each sub-question are the ones they read.
Make actions readable. web.dev advises semantic HTML, such as real <button> and <a> elements, labels tied to their fields and a stable layout.
State facts in text. Prices, plans, limits and integrations should be plain text on the page.
Read what the engines already say. Agents run inside the same assistants your buyers use for everyday questions. Hawkscan shows the full answers that ChatGPT, Gemini, Microsoft Copilot and Google AI Mode give to the prompts you track, so you see which brands and pages they favor before an agent builds its shortlist. When agents buy, see agentic commerce.