For twenty years, being found online meant ranking on Google. That is changing. More buyers now open an AI chat, describe what they need, and get a short list of answers back. Some never see a page of search results at all.
AI agents are reading your website for another reason too. Some are looking for a way in. Others are reading pages where attackers have hidden instructions, waiting for an agent to obey them.
Every website now has two new questions to answer: can AI agents find and understand you, and is what they read on your site safe?
Discovery has moved into the chat
Buyers start with a chatbot
In a G2 survey of 1,076 software buyers in March 2026, 51% said they now start their research with an AI chatbot more often than with Google, up from 29% a year earlier. 69% ended up choosing a different vendor than they first planned because of what the chatbot recommended.
The answer comes before the click
Traffic from AI assistants is still small, but it is growing fast. SE Ranking's study of 101,574 websites found it grew 16 times between 2024 and 2026, though it is still about 1 visit in 312. Back in 2024, Gartner predicted that traditional search volume would fall 25% by 2026 as chatbots became "substitute answer engines."
The visit may be small, but the decision is made in the answer. If an assistant can't tell what you offer, you aren't on the shortlist.
What AEO means
Answer engine optimization (AEO) is making your site easy for AI assistants to find, read and describe correctly. The basics are a robots.txt that lets in the AI crawlers you want, a sitemap, an llms.txt file, structured data (JSON-LD) that describes your company and products, and clear page titles and descriptions.
Agents are also attacking websites
Autonomous agents are breaking in
According to Reuters, as reported by Fortune, one attacker set AI agents loose on more than 1,200 websites from September 25 and breached about 100 companies, at roughly $50 per breach. We covered it in AI Agents Are Now Hacking for Real.
Hidden instructions are waiting for agents
Google scanned billions of web pages and found a 32% rise in malicious hidden instructions aimed at AI agents between November 2025 and February 2026, Help Net Security reports. Most were hidden from people with one-pixel or nearly transparent text.
In July, Zscaler ThreatLabz found a fake documentation site that told coding agents to pay $3 in cryptocurrency for a "license key." Four of 26 models it tested paid, Infosecurity Magazine reports. The instructions were moved off-screen with CSS or placed inside JSON-LD, the same structured data that helps agents understand a site.
The places that help agents find you are the same places attackers use to steer them. If a page on your site is compromised, or lets visitors post content, hidden text there speaks to every agent that reads it.
How Guardian Samurai helps
Website AI Scan: see your site the way an agent does
The Website AI Scan reads your website the way an AI agent would and scores it from 0 to 100 across four parts:
- Accuracy. A model reads your pages and writes down what you offer and who it's for. Guardian checks whether your own titles, descriptions and structured data lead an agent to the same answer.
- Discoverability. Your robots.txt, including which AI crawlers are allowed or blocked, plus your sitemap, llms.txt and well-known files.
- Structured data. The JSON-LD that tells agents who you are and what you sell.
- Agent safety. Text that reads like instructions to an AI agent, hidden text, and, in a full scan, cloaking: showing agents different content than people see. A serious finding here caps your score at 40.
Every finding comes with its evidence and a fix. A quick scan works on any public website. A full scan of a site you've verified you own reads more pages, checks for cloaking, and builds a summary of your site that you can review. You can also scan a local or staging site from the Guardian command line tool before it goes live.
Samurai Security: test like an attacker's agent
If your site connects AI agents to your systems through an MCP server, Samurai Security tests that server the way an attacker's agent would, in production and before a branch ships.
The Website AI Scan isn't a penetration test of your web app. It checks what agents read on your site. Samurai Security tests what agents can do through your MCP server.
Is your site ready? A short checklist
- Ask an AI assistant about your company and your category. Does it describe you correctly?
- Check that robots.txt lets in the AI crawlers you want, and blocks the ones you don't.
- Publish a sitemap and an llms.txt file.
- Add JSON-LD for your organization and products.
- Look for hidden text, and anything visitors can post, that could carry instructions.
- Show agents the same content you show people.
- Test any MCP server your site exposes.
Get ready for the agentic web
Create a free account to see how agents read your site, or learn more about Samurai Security.
Photo by Zulfugar Karimov on Unsplash.
