Agentic Search Optimization (ASO): How to Get Chosen By AI Agents, Not Just Found

Picture this: someone opens their AI assistant and says, "Book me a dentist appointment near downtown for next Wednesday afternoon." The AI doesn't hand them a list of links to click through. It just goes and does it — browses dental practice websites, checks availability, picks the one that fits, and schedules the appointment. The user never sees your website. But the AI agent does. And if your site isn't built for that agent to understand and use, you simply don't exist in that transaction.

That's where we are in mid-2026. Agentic search isn't some far-off thing anymore. Google showcased agentic capabilities prominently at I/O 2026. OpenAI's operator-type agents are already running for paying subscribers. Perplexity is testing agentic actions. This is happening now, and most SEO playbooks don't cover it at all.

What "Agentic" Actually Means (And Why It's Different)

There's a lot of terminology getting muddled right now, so let's be clear about what we're talking about.

Traditional SEO: You rank in search results. A human sees your link, clicks it, visits your site, and decides whether to convert.

AEO / GEO (Answer Engine / Generative Engine Optimization): An AI reads your content and summarizes or cites it in an answer. The human reads the AI's answer. They might click through to you, or they might not.

ASO (Agentic Search Optimization): An AI agent is given a goal by a user. The agent autonomously browses, evaluates, and takes action on the user's behalf. The human isn't browsing — the agent is. Your job is to be the site the agent selects and can successfully complete a task on.

The key difference is intent completion. With AEO, you're trying to get cited. With ASO, you're trying to get chosen and used. That's a meaningfully different optimization goal.

The Three Stages Every AI Agent Goes Through

When an AI agent receives a task, it generally moves through three stages before completing it. Understanding this pipeline tells you exactly where to optimize.

Stage 1: Retrieval

The agent figures out which sources to consider. This is influenced by the agent's training data, its search queries, and any knowledge base it has about trusted providers. If your brand has no presence in the AI's knowledge base and no visibility in the search channels the agent uses, you won't even make the candidate list. This is where your existing SEO, AEO, and brand reputation work pays off — they feed the retrieval stage.

Stage 2: Evaluation

The agent visits candidate websites and tries to extract the specific information it needs: pricing, availability, product specs, service areas, booking options. This is where most websites fail the agent. If your prices are hidden behind a modal that requires JavaScript to load, the agent can't read them. If your availability calendar is a complex embedded third-party widget, the agent gets nothing useful. If your "contact us" page is the only path to get a quote, the agent hits a dead end.

Stage 3: Selection

The agent picks the best option and — depending on how capable it is — either completes the transaction or hands off a pre-populated recommendation to the user. At this stage, clear structured data, a frictionless booking/purchase path, and explicit trust signals (reviews, certifications, contact info) all become competitive advantages.

The Google-Agent User-Agent: What It Tells You

In early 2026, Google started sending traffic under a new user-agent string: Google-Agent (distinct from Googlebot). This is the crawler used when Google's agentic systems browse on behalf of users in AI Mode and related features. Semrush, Cloudflare, and several other platforms have now confirmed they're seeing this in their logs.

Why does this matter? A few reasons:

• It tells you whether Google's agents are visiting your site at all. If you're not seeing Google-Agent in your server logs, you're not in the pool for agentic task completion — at least not from Google.
• It means you can (and should) test how your site responds to that user-agent. Tools like Screaming Frog now let you render pages as Google-Agent to see what the agent actually sees.
• Other agentic platforms use their own strings — Perplexity, OpenAI's browsing agent, Anthropic's Claude — and you'll want to track those too. Check your server logs for unfamiliar crawlers; there are a lot more of them than there were 18 months ago.

Don't block these crawlers. Seriously. Some site owners have started throwing up blocks against AI crawlers indiscriminately because of concerns about content scraping. That's a legitimate concern for some content-heavy publishers, but if you're a service business, an e-commerce store, or any site that wants agentic traffic, blocking these crawlers means opting out of this entire traffic channel.

Practical ASO: What to Actually Change On Your Site

This is the part people want. Here's what genuinely moves the needle for agentic optimization right now.

1. Implement a llms.txt file

Similar to robots.txt (which tells crawlers what to index) and sitemap.xml (which tells them what exists), llms.txt is an emerging standard that tells AI systems how to understand and use your site. It lives at /llms.txt and contains plain-text descriptions of your key pages, products, services, and what tasks a user might be completing on your site. There's no official spec yet — it's more of a de-facto standard being developed by the AI community — but several major platforms are already reading it. Get yours in place now before it becomes table stakes.

2. Make critical data machine-readable in the HTML

Your price has to be in the DOM. Your availability status has to be in the DOM. Your address, phone number, hours — all of it needs to exist in plain, crawlable HTML, not just as dynamically loaded JavaScript after a user interaction. This is one of those things that's easy to say and genuinely annoying to implement if your site was built around a JS-heavy framework, but it's non-negotiable for agentic visibility.

3. Structured data — and use it properly

Schema markup has never mattered more than it does right now. For agentic systems, structured data is the difference between an agent that can confidently extract "this hotel room costs $189/night and has 3 available on July 15" and an agent that has to guess or skip you entirely. Implement Product, Offer, LocalBusiness, Service, FAQPage, HowTo — whatever is relevant to your content. Keep it accurate and keep it updated. An agent that gets burned by stale data (says you have availability when you don't) will lower its confidence score for your domain over time.

4. Speed matters more for agents than for humans

AI agents often have timeouts. If your page takes 6 seconds to load all its content, an agent on a tight clock may cut off before it gets the information it needs and move to the next candidate. Your Time to First Byte and Largest Contentful Paint matter for agentic discovery even if the agent doesn't care about your UX. Get your server response times under 500ms and your main content loading inside 2 seconds on the first render.

5. Build frictionless paths for agentic actions

If an agent can complete a task — book, purchase, request a quote — it will. If it can't, it'll move on to a competitor that lets it. This means: booking widgets that work without login, purchase flows that don't require account creation, request forms with straightforward required fields. The harder you make task completion, the less often agents will pick you. Some sites are even starting to expose lightweight APIs specifically for agentic access, though this is still early-stage for most businesses.

Content Strategy for ASO: What Agents Actually Want to Read

Agents don't read content the way humans do. They're scanning for specific facts they can use to evaluate and compare options. Write your key pages — especially service pages, product pages, and landing pages — with that in mind.

A few things that work well:

• Explicit, fact-dense paragraphs up front. Put your prices, service areas, key specs, and turnaround times in the first 200 words of a page. Don't bury the lede.
• Comparison-friendly formatting. Tables and lists that let an agent quickly extract "X does Y for $Z" are gold. Run-on paragraphs that describe your company's philosophy are not useful to an agent trying to pick the best plumber in Seattle.
• Explicit trust indicators. Number of reviews, average rating, years in business, certifications — write these out in text, not just in a widget that requires JavaScript to render. Agents weigh trust signals heavily in the selection stage.
• FAQ sections with specific, answerable questions. "Do you offer same-day service?" and "What's included in the basic package?" are exactly the types of questions agents try to answer before selecting a provider. If your FAQ answers them directly, you win more selections.

How This Fits With What You're Already Doing

Here's the good news: ASO isn't a separate silo. It builds on top of the work you've already put into SEO and AEO. Strong E-E-A-T signals? Those help with agentic trust evaluation. Good schema markup? Already an ASO asset. Fast Core Web Vitals? Helps with agent timeouts. Appearing in AI Overviews and getting AEO citations? Those help feed the retrieval stage of the agentic pipeline.

What ASO adds on top is the transactional layer — making sure that when an agent arrives to complete a task, it actually can complete that task on your site. That's the gap most sites haven't addressed yet, because most SEO advice still assumes a human is the end user.

Think of it as three layers:

1. Be findable → Traditional SEO + AEO/GEO
2. Be understandable → Structured data, plain HTML content, llms.txt
3. Be usable → ASO — frictionless task completion, machine-readable data, fast load times

Most sites have layer 1 somewhat covered. Many have started working on layer 2. Almost nobody has seriously worked on layer 3 yet. That's your window.

Bottom Line

The shift from "a human searches and finds your site" to "an AI agent completes a task on a human's behalf" is already happening — it's not a future scenario. Google I/O 2026 made it clear that agentic search is a core product direction, and the user-agent data backs it up. The sites that get this right early will own the agentic traffic channel while everyone else is still arguing about whether AI Overviews are hurting their click-through rates.

Start with the basics: audit your pages for machine-readability, add or fix your structured data, check whether critical information is in your HTML on first load, and set up a llms.txt file. None of this is rocket science — it's mostly the kind of technical hygiene work that's been good practice for years, now with higher stakes.

The bar for agentic selection isn't impossibly high. It's just different from the bar for human UX. And right now, most of your competitors haven't figured that out yet.