Agentic search is search carried out by an AI agent that works through a task in steps: it plans several queries, reads and compares the sources it finds, decides what else it needs to know, and may then take an action such as filling in an enquiry form or making a reservation. The user receives a finished answer or outcome rather than a page of links.
How agentic search works
A traditional search engine matches one query to a ranked list of pages. An agentic system starts from a goal. Asked “which accountants near Guildford handle sole traders and offer fixed monthly fees?”, it might split the task into searches for accountants in the area, visits to their service pages, a check of pricing pages, a look at reviews and a comparison of the results. The language model behind it decides at each step whether it has enough evidence to stop.
Under the surface, it relies on the same retrieval methods that power AI answers, often called retrieval-augmented generation: fetch relevant passages, then write from them. The difference is persistence. An agent may read far more pages than a person would, and it favours sources where the needed fact is stated plainly and is easy to extract.
At the time of writing (October 2026), examples include Google’s AI Mode, which runs many related searches behind a single question and has begun to complete some tasks for users in certain markets, along with the research and agent modes in ChatGPT, Perplexity, Gemini and Copilot. Features and UK availability change often.
Why it matters
For a UK business, agentic search changes who reads your website. Increasingly, the first visitor is software working out whether you fit the brief. If the agent cannot quickly confirm that you serve Guildford, work with sole traders and publish fixed fees, it will recommend someone who makes those facts obvious.
It also cuts clicks. When the agent compares options and presents a shortlist, the user may contact one business directly without visiting the others. Traffic reports can fall even as enquiries hold up, so measure leads and their sources, not only sessions.
There is an upside for smaller firms. An agent comparing options on stated facts does not care how large your advertising budget is. A local business whose site states its services, coverage, prices and reviews clearly can be shortlisted alongside national brands that rely on reputation and vague copy.
Common mistakes
- Service pages that describe the business in general terms but never state who it serves, where, at what price or how quickly.
- Different facts in different places: one price on the website, another in Google Business Profile, a third on a directory listing.
- Blocking AI crawlers in robots.txt without deciding whether that is the intention, then wondering why AI tools never mention you.
- Rewriting every page in a question-and-answer format, which helps nobody if the answers stay vague.
- Ignoring third-party sources. Agents check reviews, directories and press coverage, not only your site.
How to act on it
List the questions a careful buyer would research before choosing you, then check that your site answers each one clearly, under a heading, in text rather than images. Make your key facts consistent across your website, Google Business Profile and the directories that matter in your sector. Decide deliberately which AI crawlers may access your content.
Then test. Ask the main AI assistants the kind of multi-part question your customers would ask and see whether you appear and whether what they say is right. Track AI referrals and enquiry sources separately in your analytics. This sits within generative engine optimisation and is part of the work I do in my AI search optimisation service.