Agentic search: prepare your site for AI Mode and Google agents
Google has just announced the agentic era. Gemini 3.5 Flash is now the default model of AI Mode, agents monitor the web while your client sleeps, and the search box changed for the first time in 25 years. We optimize your site so you appear when those agents recommend brands in your sector.
- 1B Monthly users on Google AI Mode
- 2024 Working LLM visibility since
- 25 years Since search last changed like this
On May 19, 2026, Sundar Pichai closed the Google I/O keynote framing this era as agentic. It was not marketing rhetoric. Gemini 3.5 Flash is now the default model of AI Mode for every user. Information agents arrive on Google AI Pro and Ultra this summer. The window to prepare your site before your competitors do is measured in weeks.
What agentic searches are and why they affect you
An agentic search does not look like a traditional search. Your client does not type three words and skim ten blue links. They ask an assistant to solve a problem, and the assistant, on its own, reads pages, compares options, calls businesses or completes bookings. The AI moves from answering questions to doing things.
This changes two things for your brand.
First, the output format. The agent does not return ten results. It returns one recommendation, or a pair. If your site is not in that shortlist, you do not appear.
Second, the criteria. The agent does not decide on classic keyword relevance. It decides on the ability to extract structured information with confidence, on demonstrable source authority and on the ability to act against the site without friction. Three axes your 2023 SEO playbook did not consider.
A concrete example. An online travel agency gets 70% of its traffic from Google. In the agentic era, part of that traffic no longer enters its site. The Gemini agent goes in, checks availability, compares prices against three competitors, presents two options to the user and books the one the user picks. The merchant sees the booking as if the user came in directly. What it does not see is that its site competed against four others in a process it never consciously took part in.
Landing in that agent shortlist is what we call agentic SEO. It is the branch of GEO work we have been running since 2024, focused specifically on agents that act, not just on assistants that answer.
What changed at Google I/O on May 19, 2026
Five keynote announcements directly affect how your customers find you. The figures come from blog.google itself, not from sector commentary. If you prefer the editorial breakdown with the 7 concrete actions we recommend, you have it in Google I/O 2026: what changes for SEO.
- 01
Gemini 3.5 Flash, default model of AI Mode
Sundar Pichai said it openly in the keynote. Gemini 3.5 Flash is now the default AI Mode model for every user, in every country. AI Mode is hovering around 1 billion monthly users. The practical consequence is that the AI deciding which pages to read and what to answer when someone runs a complex search on Google is this one. If your content does not fit how Gemini processes the question, you stop appearing in the answer.
- 02
A new search box (first redesign in 25 years)
For the first time in more than 25 years, Google's search box changes. It is no longer a three-word field. It expands dynamically, accepts multimodal inputs and pushes the user to describe what they want in long sentences. What used to be "best lawyer Madrid" now becomes "I need a lawyer in Madrid specialized in commercial law for an 80,000 euro claim". Your content has to answer questions like that, not short keywords.
- 03
Information agents: 24/7 monitoring in the background
Information agents are personal agents the user enables and leaves running in the background. They check listings, prices, slots, deals, whatever it is, without the person searching again. The example Google showed: someone is looking for a rental flat and the agent keeps crawling the web 24/7, pinging them when one fits. They arrive on Google AI Pro and Ultra this summer. If your site is not readable by this kind of agent, you sit outside the monitoring loop.
- 04
Agentic bookings extended to local services
Agentic bookings already existed for restaurants. At I/O 2026 they extend to local services: home repairs, hairdressers, pet care and similar. The agent can call the business on behalf of the user to confirm availability. It starts this summer in the United States. It will land in other markets later. If your business is local, what the agent reads about you on Google and on your own site decides whether you make it into that call.
- 05
Agentic coding in Search: answers built on the fly
Search can now build answers with dynamic layouts and interactive widgets. Not just paragraphs. The engine decides in real time how to present the information depending on the question. For your brand this means content is no longer consumed exactly as you wrote it. It gets broken down into extractable data that the agent recomposes as it needs. Structuring information well starts to weigh as much as writing it well.
How we prepare your site for agentic search
We work across six fronts adjusted to the sector, the current state of your site, and the competitive intensity in AI answers for each vertical.
-
Agentic readiness audit
Diagnosis of what an agent can or cannot read from your site today. Current schema coverage, quality of extractable data, prior presence in related generative engines, behavior of your site against AI crawlers, and rendering on Gemini, ChatGPT and Perplexity.
-
Advanced schema and structured data
Implementation of Organization, Person, Service, Product, FAQ, HowTo, BreadcrumbList and vertical schema specific to your sector. Agents prioritize sources where information is marked up clearly. Schema stops being optional.
-
Agent-extractable content
Refactor of content into atomic blocks: direct answer up front, explicit definitions, verifiable data with sources, structured lists. The agent comes in, extracts, cites or summarizes. If it cannot find the piece it is looking for in a clean paragraph, it does not use it.
-
Optimization for long conversational queries
The new search box pushes users to phrases of 15 to 30 words. Your content has to answer that full intent, not just three short keywords. We build pages aimed at sector questions written the way a person speaks them, not the way they type them.
-
Llms.txt and access signals for agents
Llms.txt is not yet an official standard and Google has not taken a public position in favor. We implement it in sectors where the line between editorial use and agent use starts to matter (media, large-catalog ecommerce, technical B2B). In others it adds nothing. Case by case.
-
LLM and agent visibility monitoring
Monthly tracking of brand mentions in ChatGPT, Claude, Perplexity, Gemini and AI Overviews. For the agentic layer we add task probes (bookings, comparisons, recommendations) to see whether your site lands in the agent shortlist. Without this measurement, agentic work is blind.
How GEO and Agentic SEO fit together
GEO and agentic SEO are not separate disciplines. Agentic SEO is the branch of GEO that deals specifically with agents that execute tasks, not with assistants that only answer. If you work the full GEO layer with us, agents are already covered. If you come in just for the agentic piece because your sector moves faster, that works too.
| Aspect | GEO (broad layer) | Agentic SEO (specific layer) |
|---|---|---|
| Target engine | ChatGPT, Perplexity, Claude, Gemini, AI Overviews | Agents that execute tasks (Information agents, agentic booking, agentic coding) |
| Answer format | Text, citations and recommendations | Direct action: bookings, calls, comparisons, purchases |
| Primary metric | Brand mentions and citations in answers | Inclusion in the agent recommendation shortlist |
| Tactical levers | E-E-A-T, structured content, source authority | All of the above, plus actionable schema, public APIs and real-time data |
The parent service lives at /en/services/geo/. This page goes into the detail of the agentic layer.
Why brands pick us for agentic SEO
Some agencies added GEO to their catalogue once clients started asking. We began investigating LLM visibility in 2024, when the term was still "LLM SEO" and nobody said agentic. The agentic piece sits inside the same team and the same methodology.
The work is led by a senior consultant with experience in technical SEO and semantics for generative engines. The wider team brings specialised profiles: technical for schema and architecture, content for extractable refactor, link building and E-E-A-T, and analytics for LLM visibility monitoring. More than 30 in-house people between Barcelona and Madrid.
A note on honesty. The agentic layer is still very early. Information agents arrive this summer. Agentic bookings roll out gradually. What we deliver today is rigorous preparation, not guaranteed rankings in a system that is not fully shipped yet. We say so in the first meeting.
How we run an agentic SEO engagement
We work agentic SEO as a standalone scope or as a layer inside your monthly SEO execution. The methodology is the same.
- 01
Agentic readiness audit
Week 1. Diagnosis of what an agent can read from your site today. Schema, extractable content, presence in related generative engines. Executive findings report with prioritization.
- 02
Prioritized strategy
Week 2. Optimization plan by topic cluster and query type, scored by impact and effort. Quick wins first, structural refactor next.
- 03
Implementation
Months 1 to 3. Advanced schema, content refactor for extraction, llms.txt where it fits, optimization for conversational queries, E-E-A-T reinforcement. We work against your CMS and with your technical team.
- 04
Continuous monitoring
Monthly. Tracking of mentions in the main generative engines and task probes against agents. Continuous adjustment based on results. What is not measured does not improve.
Frequently asked questions about agentic search
-
GEO is the broad discipline: getting cited in answers from ChatGPT, Claude, Perplexity, Gemini and AI Overviews. Agentic SEO is the branch focused specifically on agents that execute tasks on behalf of the user, not on assistants that only answer. The levers overlap in E-E-A-T, schema and structured content. They differ in how we measure visibility and which signals we prioritize. The agentic layer is included when we work GEO with you. More context in our GEO service.
-
Because of the window. Google officially framed the agentic era on May 19, 2026. Information agents land on Google AI Pro and Ultra this summer, agentic bookings extend to local services in the United States also this summer, and Gemini 3.5 Flash is now the default model of AI Mode. The sites ready when agents start operating at volume will capture the citations of the next 18 months. By the time it is obvious for the whole market, it will already be late.
-
Technical schema and structured data work can show up between 4 and 8 weeks for some engines. Extractable-content optimization: 8 to 16 weeks. Building consolidated presence as a source agents cite or use: 6 to 12 months. Agentic SEO does not escape the compound effect typical of SEO.
-
Partly yes. A site with solid technical SEO, correct schema and well-written content starts with an advantage. But there are important differences. Classic SEO optimizes so Google understands your site and ranks it against ten others. Agentic SEO optimizes so an agent reads it in one pass, extracts concrete data and includes it in its recommendation shortlist. The initial audit tells you exactly how much extra work your case needs.
-
Almost never. Most improvements live in templates, schema, content and headers. We only recommend changing CMS when there is a prior structural problem that was already blocking your SEO. The audit tells you whether your current CMS covers what you need.
-
The audit as a standalone deliverable starts at €1,500 depending on scope and site size. It is a fixed quote after the initial brief. If you already work SEO with us, this layer is included inside the monthly SEO Execution when the sector demands it.
-
Llms.txt is a file at the domain root, similar to robots.txt, that tells language models how to treat your content. It is not yet an official standard and Google has not committed to it. We recommend it in sectors where the line between editorial use and agent use starts to matter: media, large-catalog ecommerce, technical B2B. In others it adds nothing and we do not implement it.
-
It works for the main engines. The methodology centers on cross-cutting signals (structure, authority, extractable data, schema) that ChatGPT, Perplexity, Claude, Copilot and Gemini share. The tactical execution adjusts to each engine. The most mature agentic surface today is Google with Gemini, but the foundation we build also serves OpenAI, Anthropic and Perplexity when they activate their own task agents at volume.
Request your free Agentic SEO diagnosis
We'll analyze your current situation and send you a prioritized report with the most profitable opportunities. 30 minutes, no commitment.
No commitment.