How to Block AI Chatbots on Your Network, as a Category
Blocking one famous chatbot takes five minutes. Blocking the thousands of others that users switch to next is the real job. This guide shows how to treat AI tools as a category with separate rules for each kind of tool.
Section 1Why organisations block AI chatbots as a group
The risk is rarely one brand. It is the long tail of small tools with unknown owners, unclear data terms and no business contract, used by staff who simply want work done faster.
- Data leaves through hundreds of doorsEach unvetted chatbot, document tool or browser helper is a new place where company text can land and be stored.
- No contract, no controlApproved tools come with retention terms and admin settings. Unknown tools come with neither.
- Wrappers and resellersMany sites put a new interface on someone else's model. They change names and domains often, which defeats brand lists.
- Age and content limitsSchools and parental control apps must keep companion and image tools away from children whatever the brand.
- Filter bypassSome AI tools fetch any web page on request, which turns them into proxies around your other rules.
What a category rule gives you that a list does not
- New tools are covered the day they are categorized, with no ticket to IT.
- Different actions per type of tool: allow tutors, warn on writers, block companions.
- One rule across firewall, DNS and endpoint, all fed from the same data.
- Reports by subcategory, so you can see what people actually use.
- Fewer false blocks, because sites about AI are not the same as AI tools.
Section 2The 18 AI subcategories and a starting policy
Below is a typical starting point for a business network. Schools usually block more and allow chosen education tools. Change each action to suit your AI policy; the point is that you can.
The full definitions, boundaries and neighbouring categories are on the AI / LLM Tools category page.
Section 3Decide your rule order before you build it
Firewalls and filters read rules from top to bottom. The order below avoids the most common mistake: a broad block that also kills the tool you paid for.
Your company assistant, the coding tool for engineers, the district's chosen tutor. Scope each to a group, not to everyone.
Anything that also carries the VPN / Proxy / Anonymizers category. Put this above any warn rule so no one can click through.
Agents, data analysis, companions, voice cloning and model APIs for non-developers.
A warn page with your AI policy and a link to the approved tool, with each continue logged.
Where your proxy can inspect traffic, stop file uploads to AI domains outside the approved list.
Schools often flip rule 4 to a block and add class-time schedules. The order stays the same.
Section 3bStarting points by sector
The same 18 subcategories serve very different networks. These are common first settings we see described by IT teams; adjust them to your own AI policy and local law.
| Sector | Default for AI category | Usually allowed | Usually blocked |
|---|---|---|---|
| Law firms | Block | One vetted legal drafting tool under contract | Document chat, meeting notes, data analysis |
| Healthcare | Block | Approved clinical scribe tools on clinical networks | General chat, voice tools, uploads anywhere else |
| Banks and insurers | Block, warn for research teams | Company workspace, approved coding tools | Agents, model APIs, spreadsheet analysis |
| Software companies | Warn | Coding assistants, model APIs for engineers | Companions, proxies, unvetted agents |
| K-12 schools | Block for students | District-chosen tutors in class time | Companions, image and video tools, general chat |
| Public sector | Block | Tools approved by the central IT or security office | Tools the government has restricted on official devices |
Section 4Which method blocks what
No single layer covers everything. Use this table to choose, then follow the steps in the next section.
| Method | Whole AI category | Per user or group | Upload blocking | Works off-site | Mobile apps |
|---|---|---|---|---|---|
| Home or office router | Rarely | No | No | No | On Wi-Fi only |
| DNS filter | Yes | With agent or network split | Only if uploads use their own host | With roaming agent | On Wi-Fi or DNS profile |
| Next-generation firewall | Yes | Yes | With TLS inspection | No, unless via VPN | On Wi-Fi only |
| Secure web gateway or SSE | Yes | Yes | Yes | Yes | With device client |
| Intune with an endpoint web filter | Via the filter's category | Yes | No | Yes | Separate mobile policies |
| Google Admin (Chrome) | Lists only | By organisational unit | No | Managed Chrome | No |
| MDM | With web filter payload | Yes | No | Yes | Yes |
If you are deciding between resolver-level and proxy-level control, the DNS filtering vs web filtering comparison goes deeper, and what is URL filtering covers the basics.
Section 5Step by step, per platform
Open the platform you use. Each set of steps assumes you have the category available; where a platform only takes lists, feed it a list exported from a category database.
Router or small business gateway
- Change the DHCP DNS server to a filtering resolver that offers an AI category.
- Block outbound DNS to any other server so devices cannot pick their own.
- If the router only has a domain block list, add the major chatbots below, knowing it will not cover the long tail.
- Use a separate guest network with its own policy.
Firewall (vendor-neutral)
- Create a web filter profile with the AI category set to warn or block.
- Add a custom allow category above it for approved tools, bound to user groups.
- Add an external dynamic list fed with AI subcategories you want to treat differently from the firewall's own category.
- Turn on TLS inspection for upload and DLP rules, with exemptions for health and banking sites.
- Apply to staff, guest and server segments; servers rarely need chatbots at all.
DNS filter or resolver
- Enable the AI category in each policy.
- Load extra subcategories as an RPZ zone or allow and deny lists.
- Return a block page for HTTP and a clear NXDOMAIN or sinkhole address for apps.
- Block known DNS over HTTPS hosts and set browser policy to stop private DNS.
Microsoft Intune
- Use Intune to deploy your endpoint web filter with the AI category set to warn or block, scoped to device groups.
- Add domains for tools the category does not yet cover to the filter's block list.
- Use the Edge URL block list in the Intune settings catalog for browser-level control.
- Control Copilot features inside Microsoft 365 through the Microsoft admin settings, not URL rules.
Google Admin console
- Under Generative AI in the Admin console, turn the Gemini app on or off per organisational unit.
- Under Devices, Chrome, Settings, URL blocking, add other chatbot domains to the block list.
- Restrict Chrome extensions to an allow list so AI sidebars cannot be installed.
- Combine with a network or DNS filter for the long tail, which a manual list cannot track.
MDM for iOS and Android
- Restrict app installs to an approved catalogue so chatbot apps cannot be added.
- On supervised iOS, add a web content filter payload with your blocked domains.
- Push a DNS or per-app VPN profile so mobile data goes through your filter.
- Review the list of installed apps monthly for new AI apps.
Major chatbot domains, for platforms that only accept lists
These are the main public addresses of the best-known assistants. They are a starting point only, since each service also uses support domains that change.
chatgpt.comopenai.comgemini.google.comclaude.aicopilot.microsoft.comperplexity.aichat.deepseek.comgrok.commeta.aichat.mistral.aipoe.comcharacter.aiCheck any domain against the database with the website category checker, or a whole log export with bulk categorization, to see which of your users' AI sites are already covered.
Section 6Apps, extensions and bypass routes
Users route around AI blocks in a few predictable ways. Each pair below shows the route and the control that closes it.
Section 7Test plan for an AI category block
Run this once after setup and again every quarter. AI tools move quickly, so the second run often finds gaps the first did not.
- Pick ten AI sites: three famous, four mid-sized, three you found this month. Visit each from a managed laptop on the office network.
- Confirm the approved tool still works for its group and is blocked for everyone else.
- Try a file upload to a warned site. It should fail if your upload rule is in place.
- Repeat from guest Wi-Fi with a personal phone, then from a managed laptop at home.
- Open the phone apps of two assistants on mobile data on a managed phone.
- Turn on a consumer VPN and repeat step 1. Every site should still fail.
- Open a news article about AI and a university AI research page. Both should load; if not, the category is too broad.
- Read the logs by subcategory. The busiest subcategory tells you where to focus your AI policy next.
Section 8Keeping the AI category current
AI tools are among the fastest-moving parts of the web. A block is only as good as the data behind it, so the update cycle matters as much as the rule.
| What happens | How the data handles it |
|---|---|
| A new AI tool registers a domain | About 300,000 newly registered domains are categorized every day and shipped as daily delta files |
| A parked domain later becomes an AI tool | New domains are checked again after a week and after a month |
| An established site adds AI features | Multi-category ranking adds AI / LLM Tools as a secondary category without changing its primary |
| An AI tool closes and the domain is reused | Monthly refresh of all other domains picks up the change |
| You need the data in another format | CSV, Parquet or JSON, with exports such as DNS RPZ, EDL, hosts and PAC files; see blocklist formats |
Building a filter product? OEM licensing covers embedding the categories in your own platform, and the web filtering database overview lists everything in the file.
FAQQuestions admins ask about blocking AI chatbots
How many AI chatbot websites are there?
Far more than most admins expect. Our AI tools set holds 16,000+ AI-tool domains across 18 subcategories, and more are added every day.
Can I block all AI chatbots but allow one approved tool?
Yes. Block the AI category, then add an allow rule above it for the approved tool's domains, scoped to the users who should have it.
Will blocking the AI category break normal websites?
Not if the category is tightly defined. Ours covers sites whose main purpose is using an AI tool, not news about AI or software with a minor AI feature.
Do AI features inside Microsoft 365 or Google Workspace get blocked by a URL category?
Usually not, because they run inside the suite's own domains. Control them in the suite's admin settings.
How fast are new AI chatbot domains added?
New domains are categorized daily and checked again after a week and after a month, delivered as daily delta files.
Should schools block every AI chatbot?
Many block general chat and companions while allowing a chosen education tool. See K-12 school web filtering for how districts set this up.
Block AI by type, not by brand
License the AI tools category set with 18 subcategories and daily updates for your firewall, DNS filter or security product.
ChatGPT is a trademark of OpenAI. Gemini, Google, Chrome and Google Workspace are trademarks of Google LLC. Microsoft, Copilot, Intune, Edge and Microsoft 365 are trademarks of Microsoft Corporation. Claude is a trademark of Anthropic. Perplexity, DeepSeek, Grok, Meta AI, Mistral, Poe and Character.AI are trademarks of their respective owners.
These names are used here only to identify the products discussed. Web Filtering Database is not affiliated with, sponsored or endorsed by any of these companies.