If you run a hair salon, a restaurant or a clinic, you probably think AI is about other things: big companies, search engines, marketing. The reality is much closer to home. Programs that work for AI assistants may already be reading your website right now. Here's how to check, without needing to be technical, and what to do next.
What an AI agent is (and how it differs from Google's robot)
You've been living alongside robots on your website for years. The search engine's crawler comes in, reads your pages and leaves to index them: its job ends with filing and classifying. An AI agent does more: it's a program that enters your website on behalf of a person, with a specific task.
When someone asks their assistant “find a dentist that opens on Saturdays and tell me how much a first appointment costs”, the agent visits real websites: it checks opening hours, compares prices and, in some cases, prepares a booking or a purchase so that the person only has to confirm.
In your statistics or your hosting logs, these programs appear under names such as GPTBot, ClaudeBot or PerplexityBot. They're not viruses: they're the “public faces” of assistants your customers may already use every day.
Why your small website already interests them
It's easy to think “why would they look at my website? I'm just a local business”. The reason is simple: when someone asks an assistant “which restaurant near me opens on Sundays?”, the answer doesn't come out of thin air. It comes from reading websites like yours: up-to-date opening hours, a menu with prices, services, availability.
More and more questions of this kind are answered that way, which is why automated traffic is growing for small and local businesses too, not just for large portals. You don't need to be famous: it's enough to be nearby and to have a website that answers the question.
This has two sides to it. The opportunity: an assistant mentioning you when someone searches for what you sell, something we covered in another post. The risk: that alongside the useful visitors come others who take advantage. This article is about both sides.
Signs that you're already getting AI visits (without being technical)
You don't need to know how to program. These five signs will give you a fairly accurate picture:
- Look at your hosting statistics or logs. Search for user agents with names like GPTBot, ClaudeBot or PerplexityBot. Many control panels list them under a “crawlers” or “bots” section.
- Check your analytics for odd visits. “Direct” traffic, with no referral, that doesn't click on anything and never gets past the first page is usually automated reading.
- Count your ghost form submissions. Repeated submissions, or bookings nobody confirms or picks up the phone about, may well be forms filled in by programs.
- Watch your server resources. Consumption spikes, or a website that suddenly becomes slow with no real visitors, are classic signs of automated traffic.
- If you can't see any of this, ask. Request the list of crawlers from the past month from your hosting provider. It's a perfectly normal request, and they should hand it over.
Legitimate AI traffic and abusive AI traffic aren't the same thing
An agent reading your opening hours, your menu or your prices isn't an attack. That's the legitimate side, and a customer can come out of it: someone asked, an assistant searched, your website answered. You can even make that job easier by making your website readable for assistants without exposing data.
The problem is something else: programs that try to extract data that isn't public, that spam forms, that test prompt injections to manipulate other AIs, or that hammer your server until it falls over.
Blocking everything blindly leaves you without the good part and without protection against the bad: you stop appearing wherever an assistant compares local businesses, and you don't truly stop the ones abusing. The sensible question isn't “yes or no to AI?” but “is this specific request legitimate or not?”.
What a firewall designed for AI agents should do
A classic firewall looks at IP addresses and fixed rules designed for people. Agents call for a different approach. If you're evaluating a solution, these criteria make a handy shopping list:
- Decide on every request. A fast, deterministic filter (with no AI) that allows, blocks or quarantines on the spot, without waiting for something serious to happen.
- When in doubt, block. What's known as fail-closed: requests that look suspicious because of their content (a prompt injection, an attempt to extract data) are judged afterwards, and if no verdict arrives, they get blocked.
- Look after privacy, even while defending yourself. There's no need to store every visitor's IP address in plain text: a daily hash is enough to spot repetition without piling up personal data.
- Learn from attacks. Anything judged to be an attack should become a rule that protects other websites for a while, not just yours.
- Make the legitimate easy at the same time. Your website should be readable by assistants (for example, through a /.well-known/agent.json manifest and an MCP server) while exposing only what is already public.
WordNext Guardian is an example of this approach: a deterministic fast lane decides on every request, applies fail-closed when there's no verdict, stores a daily hash instead of the IP in plain text, turns attacks into rules that protect all websites for 30 days, and keeps the website readable by assistants while exposing only what's public.
What to do today: a four-step plan
- Look at your statistics or logs and note which AI user agents appear, and how often.
- Notice which pages they touch most. Opening hours, menus and prices are usually the most read: they're the ones that answer the usual questions.
- Decide what you want agents to be able to do. Read what's public? Book, always with the person's confirmation? Nothing at all? It's your website and your rule.
- If you can't see any of this, ask your provider, or consider a platform that includes cookie-free analytics and control over this traffic as standard.
An afternoon spent looking at your statistics with curiosity will tell you more than a month of vague worry.
Frequently asked questions
Can I find out which AI agents visit my website if I don't have access to the server logs?
Yes. Your hosting provider's statistics panel and your platform's analytics are usually enough. And if not, ask your provider for the list of crawlers from the past month.
Is the robots.txt file enough to control AI agents?
No. robots.txt is a recommendation that well-behaved bots usually respect, but it isn't a security barrier, and not every agent follows it. A firewall does decide and enforce, on every request.
Should I block AI agents altogether?
Blocking everything also removes you from the answers where an assistant compares local businesses. The sensible approach is to allow reading of what's public and stop the abuse, request by request.
Can an AI agent book or buy on my website without my permission?
That's exactly what an agentic firewall should prevent. In WordNext, an AI can only book or prepare a purchase if the business enables it, and always with the person's confirmation or a mandate signed by them; the price is always calculated by the server.
AI agents are no longer just a big-company matter: they read the website of any business that answers a question. Finding out whether they already visit you is the first step; deciding what they may do is the second. If you'd like to see all of this working together, take a look at the WordNext Guardian page: an agentic firewall and a website readable by assistants, with no magic promises.