AI Website Editing
Edit your website by asking your AI assistant, in drafts you approve
AI website editing lets your team update the website by asking an AI assistant in plain language, without waiting for a developer. We connect your site to the assistant your team works in, such as Claude or ChatGPT, or build a chat editor into the site itself. Drupal, WordPress or a custom stack: we limit what the AI can touch and keep changes in drafts wherever your site supports it, until a person on your team publishes them. We also build other AI, such as a booking assistant for a client's website, content analysis for large websites and semantic search, with examples further down this page.
What AI website editing looks like for your team
Your editor opens their AI assistant, connects to your site once through a secure login, and from then on asks for changes in plain language. The assistant reads the page, makes the change through tools we set up for your site, and saves it as a draft. Someone on your team looks it over and publishes.
Nobody has to learn the admin screens, and a text change no longer waits in a developer's queue.
Illustration of a typical request
From a sentence to a published page
Connect once
Your editor links their AI assistant to your site through a login and consent screen on your own site.
Ask in plain language
Describe the change. The assistant reads the page and uses the tools we set up for your site.
Review the draft
The change is saved as a draft wherever your site allows it. Nothing goes live on its own.
Publish
Someone on your team looks it over and publishes. Publishing stays a human decision.
Not tied to one CMS or one AI vendor
Two parts have to fit: your site and the assistant your team works in.
Your site
Our setups in production today run on Drupal, on WordPress and on a static site with no CMS. For anything else, such as another CMS, a custom-built application, a Jamstack site or a headless setup with a separate front end, we build the same layer for your stack: editing tools that match how your content is stored, an access model and a review step before anything goes live.
Your AI assistant
We connect your site through the Model Context Protocol (MCP), an open standard for linking AI assistants to outside systems. Claude supports it, and so do ChatGPT (custom connectors on paid plans), Gemini Enterprise and Microsoft Copilot Studio. Our production setups run with Claude, and we confirm the connection with your team's assistant before go-live. If you would rather not use an outside assistant at all, we can build a chat editor into the site itself.
What the package includes
An MCP server for your site
The connection your team's AI assistant uses to read and edit the site. We install and configure it on your CMS, or build one for a custom, static or headless site.
OAuth login
Editors connect through a login and consent screen on your own site. No shared passwords or API keys pasted into chats, and each connection can be revoked.
A restricted role
The AI reaches your site through an account and a set of tools limited to content work, never through full administrator rights.
Per-tool policies
We start with page deletion switched off and the homepage and site settings locked, then open up only what you need, such as menus or opening hours.
Drafts before publishing
We configure the AI to work in drafts wherever your site allows it, so a person on your team decides what goes live.
Tools for your building blocks
Your pages are built from specific components, paragraphs, blocks or templates. We add the tools the AI needs to use them correctly, plus menus and translations where you need them.
Instructions written for your site
The AI is told which tools to use, what never to do, and to ignore instructions it finds inside page content or uploaded files.
Docs, training and support
A short connection guide for your team, a live session on real tasks from your backlog, and help while the first edits go out.
Where an outside assistant does not fit, the package can instead include a chat editor built into your site, with a preview and one-click publishing.
Built to be safe on a live site
Giving an AI write access to a production website is only a good idea when the account it uses can do very little damage. That principle shapes every setup we do.
We also check the parts outside your site's code. On one client site, the CDN's bot protection was blocking the AI assistant's calls while the connection looked healthy. We found it in the firewall logs and added an exception for the MCP endpoint only.
Drupal
The AI's access is the overlap of the editor's own permissions and a dedicated OAuth scope, so it can never do more than that editor could do by hand. Your team connects with editor accounts, not administrator ones.
WordPress
Some MCP plugins ship with code execution, file access and WP-CLI switched on. We take those off the list of allowed tools before the plugin goes live, so whoever connects, the AI can only call the content tools on that list.
Static site with a built-in editor
In the chat editor we built for a static site, the AI can only read and change the content source files and has no shell access.
Proven setups, and a path for every other stack
Drupal 10 and 11
We use the Drupal MCP Server and Tool API modules, Simple OAuth with OAuth 2.1 (dynamic client registration and PKCE), and Canvas Tools on sites built with Drupal Canvas. Where the standard tools fall short, for example with Paragraphs, Layout Builder pages, menus or translations, we write custom tools for your content model. Signing keys live outside the web root and are created on deploy.
WordPress
We connect your site through the WordPress Abilities API and an MCP plugin, behind an allow-list. Custom abilities cover your own blocks, media, menus, theme options and multilingual pages.
A static site with a built-in AI editor
A simple company site can skip the CMS entirely. Your team edits it in a chat window in the browser, previews the result and publishes with one click. The editor we built runs on the client's own AI provider account and costs money only when someone edits.
Other CMSs, custom and headless sites
We start with a short technical review of how your content is stored and published, then build the tools, the access rules and the review step for your stack. The review confirms the scope before we start.
We run it ourselves
This page was drafted and saved on itech4web.com through the same kind of MCP setup we offer, as a draft that our founder reviewed before it went live. The same approach runs on client sites.
itech4web.com
Our own site. Pages like this one are drafted through MCP and published by a person.
Triangu
The team can build and edit Canvas pages from Claude.
MuSe in Vienna
The team edits pages, menus, images and opening hours in German and Russian from the Claude app.
LWG CPAs & Advisors
A site in Indianapolis with a built-in AI editor and one-click publishing, so a small accounting firm keeps its own site current.
A fixed setup fee per site
Setup prices are starting points. We send a fixed quote after a short review of your site, and the monthly support plan is optional.
Drupal or WordPress
Everything in the package above, on a platform we already run in production.
Other stacks and built-in editors
Other CMSs, custom, static or headless sites, and built-in chat editors.
Monthly support plan
This optional plan keeps AI editing working after launch: updates to the site and its AI tooling, fixes when something breaks, changes to what the assistant may do, new tools as your needs grow, and help when an edit needs a person. Hours and price are agreed per site.
Your AI subscription is separate and paid directly to the provider: for example a paid Claude or ChatGPT plan per editor, or, for a built-in editor, an AI provider API account billed per use.
AI we build beyond website editing
AI website editing is what clients ask us for most often. We also build AI for other jobs, such as a booking assistant for a client's website, content analysis for large websites and semantic search. Each card says where the work runs and what stage it is at.
A booking assistant for MuSe in Vienna
For MuSe, a creative space and kids cafe for families in Vienna, we built Mia, a chat assistant that helps parents book a children's birthday party. It answers in German, Russian or English from MuSe's own pages, shows free slots straight from the booking calendar and saves the parent's contact details as soon as they share them. It mentions add-ons such as an entertainer when they fit, then sends the request through the site's usual booking form, so staff get the normal calendar entry and emails. Complaints, large groups and anything it does not know go to the team, and it tells parents up front that it is an AI. It is not live yet.
Content analysis for large websites
On a site with thousands of pages, some problems only show up when you read pages side by side. Our content analysis finds near-duplicate pages, pages that give different fees, dates or eligibility rules for the same thing, and related pages that should link to each other but do not. Semantic search narrows thousands of pages down to the likely problems, an AI model reviews each one, and your content editors get a report with a short list of the fixes that matter most.
Semantic search across emails, meetings and files
Semantic search finds text by meaning, not by exact words. In our CRM it covers every email, recorded meeting, chat and note with a client, plus the text inside attached PDF, Word and Excel files. Search results show only records the person has permission to read. Our content analysis uses the same approach to find pages that say the same thing in different words.
Also running today
An AI assistant on every client record
Ask what we promised a client or when we first met, and it answers from that client's emails, meetings and documents. It searches them before it answers instead of guessing.
Meeting notes and emails turned into tasks
After a recorded call or an incoming email, AI picks out the next steps, files them as tasks on the right client and sends development work to the right project as tickets.
Alt text drafted for images
When an editor uploads an image, AI drafts the alt text that screen readers read aloud. The editor can change it before saving. It runs on the Drupal AI module.
A summary written for AI assistants
An llms.txt file gives assistants such as ChatGPT, Perplexity and Claude a short, plain summary of who you are, what you offer and which pages matter most.
If you have a different job in mind, tell us what takes your team too long, and we will tell you whether AI is a good fit for it.
Common questions
What is AI website editing?
AI website editing means your team changes the website by asking an AI assistant, such as Claude or ChatGPT, in plain language. The assistant connects to your site through the Model Context Protocol (MCP) with a restricted account, makes the change with tools set up for your site and saves it as a draft. A person on your team reviews and publishes it.
Our site does not run on Drupal or WordPress. Can you still do this?
Yes. Drupal and WordPress are where we have ready-made setups. For another CMS, a custom application, a static site or a headless front end, we build the editing tools and access rules for your stack, after a short technical review.
Which AI assistants work?
Any assistant that can connect to a remote MCP server with OAuth. Today that includes Claude, ChatGPT (custom connectors on paid plans), Gemini Enterprise and Microsoft Copilot Studio. Our production setups run with Claude. Plans and features change often, so we check what your team's subscription allows before we start.
How much does it cost?
Setup on Drupal or WordPress starts at $2,000. Other stacks and built-in chat editors start at $3,000. We send a fixed quote after a short review of your site. The monthly support plan is optional, and its hours and price are agreed per site. Your AI subscription is paid separately to the provider.
Do we still need a developer?
For content, no. New features, design changes and new page components are still development work, and our website support team handles those.
Can the AI break the site?
Not through the tools we give it: it cannot delete pages, change settings, install plugins or modules, or manage users unless you ask us to allow that. On a draft-based setup, the worst case is a draft you did not want, and you discard it.
Can you add other AI features to our website?
Yes, when there is a specific job for AI to do. For MuSe in Vienna we built a booking assistant that answers parents' questions, checks free slots and sends birthday party requests. It is a working demo and not live yet. For large websites we offer content analysis that finds near-duplicate pages, conflicting numbers or rules, and related pages that do not link to each other. In our own CRM we run semantic search and an AI assistant across client emails, meetings and documents, and AI that turns meetings and emails into tasks. On websites we have also shipped AI-drafted alt text and an llms.txt summary for AI assistants. For anything else, we look at the problem first and tell you whether AI is the right fix before we quote.
Let your team edit the site by asking
Tell us which site you want to edit with AI, what it runs on, and which assistant your team uses. We will review your content model, list what the AI should and should not be allowed to do, and send you a fixed quote.
