Blocking the three AI assistants everyone has heard of takes a minute. The hard part is the thousands of smaller AI tools, many of which launch on new domains every week.
This page covers the well-known AI tool domains, the 18 subcategories behind our AI list, and how to run a policy that allows approved tools and holds back the rest.
These are the main public domains of widely used AI tools. Several run on subdomains of very large companies, which changes how you block them, as explained further down.
| Tool | Main domains | Supporting domains | Type | Blocking note |
|---|---|---|---|---|
| ChatGPT | chatgpt.com chat.openai.com | openai.com oaistatic.com oaiusercontent.com | Chat assistant | Block ChatGPT |
| Claude | claude.ai | Vendor's own domains | Chat assistant | Own domain, simple to block |
| Gemini | gemini.google.com | Shared Google hosts | Chat assistant | Hostname only, never the parent |
| Microsoft Copilot | copilot.microsoft.com | Shared Microsoft hosts | Chat assistant | Hostname only, never the parent |
| Perplexity | perplexity.ai | Vendor's own domains | AI answer engine | Own domain, simple to block |
| DeepSeek | chat.deepseek.com | deepseek.com | Chat assistant | Block DeepSeek |
| Grok | grok.com | Also inside x.com | Chat assistant | In-app use needs the X policy |
| Meta AI | meta.ai | Also inside Meta apps | Chat assistant | In-app use needs the social policy |
| Mistral Le Chat | chat.mistral.ai | mistral.ai | Chat assistant | Own domain, simple to block |
| Poe | poe.com | Vendor's own domains | Multi-model chat | Own domain, simple to block |
| Character.AI | character.ai | Vendor's own domains | AI companion | Block AI companions |
The famous tools are a tiny share of the AI category. Most of the 11,800+ domains are small writing tools, image generators, homework helpers and chat wrappers that nobody would think to add by hand.
New AI products launch on fresh domains every week, which makes a typed-in list stale almost at once. Our database labels them AI / LLM Tools, so an approved-tools policy can sit on a category lookup.
The AI / LLM Tools category is split into 18 subcategories, so a policy can block image generators for students while allowing a coding assistant for developers. Full definitions are on the AI and LLM tools category page.
Each AI domain also carries its regular web filtering categories. An AI homework helper, for example, carries AI / LLM Tools with an education secondary, so school policies can treat it differently from a general assistant.
Subcategories also make reports readable. Instead of a long list of unfamiliar AI domains, the monthly report shows how much traffic went to writing tools, image tools, coding tools and companions, which is the view leadership and school boards ask for.
Each of these returns the AI / LLM Tools label, and one also carries a second tag for web apps. Policies can key on the first label and ignore the second.
| AI service | Categories in our data |
|---|---|
| openai.com | AI / LLM Tools |
| claude.ai | AI / LLM Tools, Web-based Applications |
| perplexity.ai | AI / LLM Tools |
| midjourney.com | AI / LLM Tools |
| huggingface.co | AI / LLM Tools |
Few organisations want all AI blocked or all AI open. These three patterns cover most real decisions, and each one maps cleanly onto the category data.
Allow the AI assistant the company pays for and has a data agreement with. Block or warn on the rest, so staff do not paste customer data into unknown tools.
Teachers use AI for planning. Students get the school's chosen tool, while image generators and companions stay blocked.
Younger children get no AI chat at all. Teens get educational AI with companions and image tools off.
For school context, see how to block AI chatbots and how to block AI image generators.
A starting map many organisations adapt. Each cell is an action you can set per subcategory, rather than one switch for all of AI.
| Subcategory | Office staff | Developers | Students | Young children |
|---|---|---|---|---|
| Text & Language | Approved tool, warn others | Approved tool, warn others | Approved tutor only | Block |
| Image & Visual | Warn | Warn | Block | Block |
| Code & Development | Warn | Approved tools allowed | Allow in coding classes | Block |
| Models & Infrastructure | Block | Allow approved platforms | Block | Block |
| Agents & Automation | Block | Warn | Block | Block |
| Education & Learning | Allow | Allow | Approved tools allowed | Block or parent choice |
| Lifestyle & Entertainment | Block | Block | Block | Block |
| Security & Detection | Allow | Allow | Allow for teachers | Block |
Keep the map in your written AI policy as well as in the filter. When someone asks why a tool is blocked, the answer should be one line in that document.
Some of the most used assistants live on subdomains of companies whose main domain you cannot block. Getting this wrong takes out search, mail or office tools for everyone.
# blocks search, mail, maps and more ||google.com^ ||microsoft.com^
Every service under the parent goes down, and the help desk hears about it within minutes.
# blocks only the assistant ||gemini.google.com^ ||copilot.microsoft.com^
Use a format that supports full hostnames, such as adblock-style rules or RPZ entries. Hosts files work too, but match only the exact name.
When you add exceptions or extra rules for these platforms, always write the assistant's host rather than the parent. See the blocklist formats guide for how each format matches subdomains.
AI is spreading into products people already use. A domain list is strong for standalone tools and weaker where AI is a feature of something else.
Assistants built into mail and document tools run on the same domains as the suite. Turn them off in the suite's admin console.
Chat assistants inside messaging and social apps use the app's own domains. Your social media policy covers them.
Some browsers ship their own AI side panels. Manage them with browser policies on managed devices.
Apps that run AI locally make no network call to an AI service. Only device management can restrict them.
Developers may call AI APIs from code. Block the Models & Infrastructure subcategory on networks where that is not allowed.
Some tools proxy AI access around filters. Those carry both AI and VPN / Proxy categories, so the VPN and proxy list helps here.
An AI policy is rarely about the tools themselves. It is about four worries, and each one points to a different part of the list.
Staff paste contracts, code and customer records into tools with unknown terms. Focus on Text & Language, Search, Knowledge & Docs and Data, Analytics & Research.
Students submit generated essays and homework. Focus on Text & Language and Education & Learning, with an approved tutor allowed.
Image and companion tools can produce content unsuited to children. Focus on Image & Visual, Video and Lifestyle & Entertainment.
Teams sign up for paid AI tools on cards, outside procurement. Logging the whole category shows the spread before renewal time.
Interest in AI filtering comes from almost every sector, but the default action differs a lot between them.
| Buyer | Typical default | What they allow | Why the list matters |
|---|---|---|---|
| K-12 schools | Block for students | One approved tutor, staff tools | New homework helpers appear weekly |
| Universities | Mostly open | Nearly everything | Log and report, block companions on shared machines |
| Banks and insurers | Block with exceptions | The licensed enterprise assistant | Customer data and regulator questions |
| Healthcare | Block with exceptions | Approved clinical scribing tools | Patient data protection |
| MSPs | Per client | What each client signs off | One category, many tenant policies |
| Security vendors | Product setting | Customer's choice | An AI toggle customers now expect |
| Parental control apps | Off for young children | Learning tools for teens | Companion and image tools |
MSPs can read MSP web filtering, and vendors building the toggle into a product can start with OEM licensing.
No category moves faster. New AI tools launch every day, existing ones rebrand, and wrappers appear around every new model release.
The AI tools set is updated every day, so new tools reach your list within a day of being categorized.
About 300,000 newly registered domains are categorized each day across all categories.
New domains are checked again after a week and a month, catching tools that launch with a placeholder page.
All other domains are refreshed monthly, which catches sites that pivot to AI or away from it.
Because tools come and go so quickly, a hand-kept AI list is out of date almost as soon as it is written. That is the main reason teams license this set rather than build one.
New AI domains reach the database as part of the roughly 300,000 added each day. You can load them as files, or call the API when a request hits a host your copy lacks.
The AI set ships in the same formats as every other category: CSV, Parquet and JSON source files, plus RPZ, hosts, adblock style, firewall EDL and PAC exports for licensed customers.
Run the category in log-only mode for two weeks to see which AI tools are actually in use, grouped by subcategory.
Agree on approved tools with legal and data protection, then map the rest to allow, warn or block.
Load the list on the resolver and firewall, with the approved tools on an allow list checked first.
Use a warning page that names the approved tool, so staff know where to go instead.
Warn pages work well for AI. Most staff who open an unapproved AI tool are trying to get work done. A page pointing them to the approved tool changes behaviour better than a hard block.
AI tools are mostly web apps with mobile versions, and several run on shared platforms. Test each route once before calling the rollout done.
Open each well-known assistant in a fresh private window and confirm the block or warning page appears.
Open the assistant apps on a phone on Wi-Fi. Cached sessions can hide a failed block, so sign out first.
Confirm search, mail and documents still work after blocking the assistant hosts on big platforms.
Sign in to the approved assistant as a normal user and run a full conversation with a file upload.
The AI category is about using AI tools, not reading about AI. Keeping that boundary tight stops a block rule from taking out the sites people learn from.
Stay in News/Media or Blogs/Personal Sites. Students researching AI ethics can still read about it.
Sit in Education/Reference or Science, so research pages remain open.
Keep their main category. A note-taking app with an AI button is still a productivity tool.
Careers and press pages can sit in Business. Check the specific host with the domain lookup.
If a site is wrongly in the AI category, allow it locally and report it so it goes back into review. Your allow entry keeps the site open for your users in the meantime, so nobody waits on a ticket.
Keep a short note with each allow entry saying who asked for it, which tool it covers and why it is needed.
AI tooling changes more quickly than most categories, which tilts the comparison.
| Question | Hand-kept AI spreadsheet | Our category database |
|---|---|---|
| How many AI sites | The ones someone noticed | 11,800+ AI-tool domains |
| How a site is judged | Opinion of whoever edits the sheet | One AI / LLM Tools label returned per lookup |
| Launch week coverage | Behind until the next review | New domains added every day |
| Beyond AI | Another sheet for each policy | 58 categories in the same data |
| Use in a product | Copy and paste into rules | Files, real-time API or offline database |
The AI tools category set holds more than 11,800 AI-tool domains in 18 subcategories, and it is updated daily as new tools launch.
Yes, but you must block the exact hostname, not the parent domain, or you will block search and mail too. Use a format that supports hostnames, such as RPZ entries or adblock-style rules like ||gemini.google.com^.
Block the AI / LLM Tools category and put the approved tool's domains on an allow list that is checked first. The approved tool keeps working while new and unapproved tools are blocked as they appear.
Not when the AI feature runs on the same domain as the main app, such as an assistant inside an office suite. Those features are controlled in the app's admin settings, not with a domain list.
No. News, blogs and research about AI sit in News/Media, Blogs/Personal Sites, Education/Reference or Science. The AI category covers sites where you use an AI tool, not sites that write about AI.
The AI tools set is updated daily, and newly registered domains across all categories are categorized every day, then checked again after a week and a month.
Yes. Run the AI category in log-only mode, then group the logged lookups by subcategory and domain. That shows which tools are in use before you set a policy.
For staff, a warning page that points to the approved tool usually works better than a hard block, because most people opening an AI tool are trying to get work done. For young students and children's devices, blocking is the common choice.
Pricing depends on the data you license and how it is used. Starter database tiers start from $999; enterprise and OEM licences are quoted per deployment.
Database plans begin at $999 for the 1 million most popular domains. If you want the AI category inside your own product, OEM licences start at $10,000 per year.
License 11,800+ AI-tool domains in 18 subcategories, updated daily, together with 120 million domains in 58 web filtering categories.
ChatGPT is a trademark of OpenAI. Claude is a trademark of Anthropic, PBC. Gemini is a trademark of Google LLC. Microsoft Copilot is a trademark of Microsoft Corporation. Perplexity is a trademark of Perplexity AI, Inc. DeepSeek is a trademark of its owner. Grok and X are trademarks of their owners.
Meta AI is a trademark of Meta Platforms, Inc. Mistral and Le Chat are trademarks of Mistral AI. Poe is a trademark of Quora, Inc. Character.AI is a trademark of Character Technologies, Inc.
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.