How to Migrate Your Website From Webflow to a Custom Claude Build
By Constantinos Yenis, VP of GTM Engineering at Chili Piper ·
Chili Piper just rebuilt its entire website as a custom, Claude-coded site and moved it off Webflow. In this session, the person who led the migration walked through exactly how it was done, why they did it, and the parts that are harder than a demo makes them look. If you are weighing a move off Webflow or WordPress to a custom AI-built site, this is the playbook, including the traps that cost real time. This recap covers the whole talk: the strategy, the stack decisions, the pre-launch checklist, and the honest tradeoffs.
Why move off Webflow at all?
The core argument is that the interface is changing, not the output. Building a website has always meant carrying the vision of the business and the designer to the end user through code. The traditional path was designer to Figma to developer to CMS. What is new is that the mouse is becoming language and the handoff to a developer is becoming a conversation with an LLM. What gets replaced is the interface, not design skill.
That distinction matters, because AI is an amplifier. It multiplies the skill of whoever is behind it. Hand the new interface to someone without design fundamentals and you get the sloppy AI look that most models default to. Hand it to a strong designer and you get their vision, faster. Nobody becomes a great designer just because the tool changed.
The second shift is where the bottleneck lives. On a traditional stack the bottleneck is execution: developers carry a product backlog while marketers wait on landing pages for a campaign shipping tomorrow. On a custom AI build, the bottleneck moves to specification, meaning knowing what you actually want, having a clear message, and having the designs. That is a better bottleneck to have.
Should you actually do this?
Not every site should move. The honest answer is: do not move just to move.
| Stay where you are if... | Consider moving if... |
|---|---|
| Your site is a small portfolio and you are happy with it | You are locked out of editing because a developer built it and you have no access |
| You run few campaigns and create little new content | You run many campaigns and want to create content and pages fast |
| The current cost and flexibility work for you | Cost or platform rigidity is holding you back |
What do you gain?
- Creativity gets cheap. Want a game on your 404 page? You can have it. The only real cost is tokens, and it is minimal. Things that used to require learning to code and putting in the hours are now available for a fraction of the time and cost.
- No built-in ceiling. Anyone on the team can create their own pages.
- Full control over performance and code. Optimizing speed on WordPress or Webflow is hard because you do not control the whole codebase, and drag-and-drop builders add technical overhead. Full code access also means real control over SEO details like meta tags and schemas, which some platforms keep rigid.
- The dev queue can disappear, though this depends on how technical your team is.
What do you lose or risk?
- Speed is double-edged. Anyone can build anything fast, which means people get carried away, add extra icons and sections, and create confusion. If you confuse, you lose. Less is more.
- Design can drift into a "Frankenstein" if you have no design system. A useful tip from the talk: literally telling Claude "do not make the website a Frankenstein, create a component system and a design system and reuse whatever you can" works better than a huge, detailed prompt.
- Governance. Because anyone can ship a page, you need process, training, and a review step.
How long does a migration really take?
A demo makes it look like an hour. A real migration can take a quarter. A first look at an existing site recreated in Claude can be ready in a day, but that is the easy part.
The time goes to everything around the code:
- Cross-functional alignment and stakeholder buy-in on whether the migration is even a priority
- A website that keeps moving, since campaigns keep shipping during the migration
- Dependencies you did not know about, such as product features living on the same domain
- Blockers that force a restart. On an earlier migration, a multi-language setup got tangled deep in the code and the work had to be reverted several commits and partly rebuilt.
The clear warning: do not tell your dev team to move the site in a week without a proper plan.
Recreate first, or redesign from scratch?
There are two paths, and the right one depends on what you already have.
| Path | Best for | Watch out for |
|---|---|---|
| Recreate like-for-like, then improve after | Big sites that rank and have ads and campaigns running | Protecting rankings and campaigns during the move |
| Redesign from scratch | New sites, or when you are unhappy with the current design and want to drop unmaintained legacy pages | More risk if the current site drives real traffic |
For a large, ranking site the safe choice is to recreate what you have, keep it like-for-like so you do not lose rankings or break campaigns, and then go improve it. That was the approach on both large sites in this project: recreate the existing design, then move into redesign and ongoing improvements. You can also mix the two, scrapping a stale blog category and redirecting the old posts while keeping high-value pages intact. Analytics helps here: pages with no traffic that never ranked are usually safe to drop.
How do you prepare and export the old site?
Do not cancel or delete your existing CMS. Keep it as a backup for a while.
- Export what the platform gives you: XML from WordPress, CSV from Webflow.
- Scrape the site with a tool like Screaming Frog. Be warned that you will keep discovering URLs that were not transferred for weeks afterward. Thank-you pages and stray pages that never got moved were a recurring source of delay.
- Use Google Search Console to find the pages that actually rank and the pages Google has indexed, then set up redirects so those do not turn into 404s.
- You can connect a Webflow MCP so that while building, Claude can check the old collection for elements and content that might be missing.
Do not skip the fundamentals of building a site
The same discipline that applies to designing a website applies here. Watch a short, high-level tutorial on how a site is built with code so you understand components and design systems, even if you never hand-code it yourself.
Then ask Claude to build a design system and reusable components. Instead of ten product pages with ten different heroes and ten different social-proof blocks, you want one of each with parameters you change. Set your colors and logos once for consistency. Left alone, LLMs will happily recreate components as brand-new code on every page, which makes the codebase unmanageable and more expensive over time.
CMS or no CMS?
This was called the million-dollar question. A CMS is where your content lives, not just posts but client logos, case studies, pricing, and features. WordPress and Webflow are CMSes; so are headless options like Sanity, Contentful, and Strapi, which can plug into custom code.
The decision comes down to who edits the site and how content gets managed. There are three broad models:
| Content model | How it works | Fits |
|---|---|---|
| Markdown in the repo | Writers put markdown files directly into GitHub | Technical teams comfortable in a repo |
| Keep a CMS | Writers log into a CMS and publish, no repo required | Big content teams and freelancers who want nothing to do with the code. A separate WordPress-to-custom project used Sanity for exactly this reason, so a team of lawyers could keep publishing without touching Claude |
| Agent as editor | Writers talk to an agent and the agent writes to GitHub | Teams that want LLM speed and flexibility |
A CMS is the safer choice when a large content team is used to logging in and editing, but it keeps the mouse-and-desktop interface and gives up the flexibility an LLM offers, such as interactive elements or a quiz embedded inside a blog post. A hybrid works too: a CMS for some content, with other pages living in markdown or code.
Chili Piper went with the agent-as-editor model. Writers, marketers, and even interested salespeople interact through Claude, and the approval process runs through GitHub pull requests with automated checks in place. Markdown-in-repo is still used for fast manual changes. This is the choice "so far," and a CMS could be added later if customers or the community start contributing content.
What breaks if you are not careful?
Do not break the lead flow
This is the highest-stakes part. Custom forms, WordPress form plugins, and Webflow forms carry a lot of hidden infrastructure: field validation, email notifications, and connections to lead-routing tools like Chili Piper Concierge that get leads to reps fast. Claude telling you it recreated a form does not mean the form actually works.
The cautionary tale: on one project a contact and demo form looked perfect and showed users a success animation, but for two days every submission went nowhere. Verify every critical customer touchpoint, and make sure all your critical scripts, forms, and pixels actually moved.
Build in automated checks
One real advantage of a custom site is that you can run automated checks on every deploy to production, plus daily and weekly checks, so critical infrastructure cannot silently break. That is something Webflow and WordPress sites generally do not have.
Protect your SEO
- Make sure content is at parity and that URLs are transferred or redirected properly.
- Export your redirects from your server or CMS and confirm they actually work.
- After go-live, set up a 404 monitor. You will have 404s, often from images or CMS files, so find and redirect them quickly.
Handle languages carefully
Translation is more than swapping text. You need proper hreflang so Google understands the English, Spanish, and German versions are the same page, and you need to verify translations visually, because Claude or Codex can claim a translation happened when it did not.
Lock down security
Check for exposed keys, stray files uploaded to production, and unguarded paths. Have your dev team take an extra look, but do not pile on so many rules that the site itself becomes a bottleneck.
"Done is a claim, not a result"
The most quotable line of the session doubles as its verification philosophy. When Claude says it finished something, check again. Better still, have agents check the work of other agents. Claude builds, then Codex runs an extra pass and often catches improvements Claude missed. Tools like Gemini-based Antigravity are strong at visual parity and responsiveness across breakpoints. You still need real automated checks that can catch things out of place, but agents reviewing agents is a genuinely useful layer.
Do you still need a translation tool?
A common question is whether a custom build removes the need for a localization platform. Claude and Codex can translate directly in code. If you use a CMS, translation is limited to what the CMS supports, so you may need a localization tool that connects by API or MCP so translators can edit.
The real value of a dedicated tool is translation memory, meaning consistent word choice so a chosen translation of a term like "scheduling" is always used. You can approximate this in a custom build with a dictionary file referenced from your CLAUDE.md so translations stay consistent. For teams with many languages and dedicated translators, keep the translation-management software. For basic needs, Claude alone can cover it.
When do you handle SEO, GEO, and AEO?
Recreate first, changing only the stack while keeping parity, then optimize and re-optimize. Recreating first also lets the team see whether the migration itself helped. If you mishandle URLs, your blog can end up performing worse than before, so the priority is safeguarding the rankings you already have. Once the site is live, SEO MCPs like Semrush, Ahrefs, and DataForSEO help you improve on-page SEO, H1s, and schemas systematically.
Does the new site actually perform better?
Here is where the session was refreshingly honest. Go-live was delayed and rolled back, so hard results are coming in a follow-up session that will share the numbers and the mistakes, not just the wins.
The bigger point on performance: the migration by itself does not make wonders. A speed comparison showed a dramatic drop in first-load time, but the old Webflow instance was not well optimized to begin with, carrying legacy scripts that could not be deferred, so it is not a fair benchmark. The new site actually has more elements above the fold with hard-coded animations rather than Lottie, plus optimized image loading and Next.js image handling that beats the old CDN. Full code control is what makes that level of optimization possible.
The warning is just as important as the win: a custom build can be worse if you do it wrong. Point a basic model at a poorly optimized site and the result can regress. The interface change only pays off when you hand it to the right people, migrate carefully, and then do the improvement work. If you do not do the improvements, nothing improves.
Key Takeaways for GTM Teams
- The interface changed, not the output. AI amplifies the skill of whoever is behind it. It does not replace design judgment.
- The bottleneck moves from execution to specification. Your new constraint is knowing exactly what you want.
- Do not move just to move. The move makes sense when you are locked out of editing, run heavy campaigns, or need speed and code control.
- Recreate like-for-like first, then improve. For a ranking site, protecting rankings and campaigns comes before redesign.
- Plan for a quarter, not an hour. Most of the time goes to alignment, dependencies, and URLs you keep discovering.
- Protect the lead flow above everything. A form that shows a success message can still be sending submissions nowhere. Verify it.
- Done is a claim, not a result. Build automated checks, and use agents to review the work of other agents.
Want to Attend the Next Session?
GTM Clawllective is a community for go-to-market teams building real workflows with AI. Sessions are hands-on and practitioner-led. The live discussion stays off the record, and every recap includes the speaker's demo video, so showing up live is the only way to join the conversation and ask your own questions. Join the community to get the invite.
About the Speaker
Constantinos "Costas" Yenis is VP of GTM Engineering at Chili Piper, where he works across sales, marketing, and product to help the company and its customers build more automation into their go-to-market motion. He joined Chili Piper in April 2026.
He started as a full-stack developer and later moved into marketing, also managing sales and customer success teams along the way. Over roughly two decades he has built sites across PHP, WordPress, Webflow, Framer, and custom code, and has led several platform migrations, including WordPress-to-custom and Webflow-to-custom builds. That range is why he approaches an AI website migration as a website project first and an AI project second.
Watch the session replay: How Chili Piper Built an AI-Native Website With Claude