You picked a template. You had a live site before lunch. That is the whole pitch, and it works.
Then the plan changes. New host, a real store, or you just want a copy of the thing you have been paying for. You search for "download my website." There is no such button.
That is website platform lock-in. Launch got cheap. Leaving with the thing that compounds did not.
Pretty templates are cheap
A magazine on a canvas builder looks organized. Eight categories. Pages that appear to share a template.
Underneath, a lot of them share nothing. Free-placement canvases do not give you a content model. They give you decorations that happen to sit in similar places.
So the quote to leave is not a page-count quote. It is a count of distinct layouts. Independent shops in 2026 price that work like a rebuild, because that is what it is.
DIY is free if you have five pages and a weekend. Automated movers run about $50 to $500 and leave the design behind. A professional rebuild off a canvas builder commonly lands between $1,000 and $5,000. Mid-size jobs run $8,000 to $25,000. Large store rebuilds start at $25,000 and climb past $75,000.
You are not paying to move a site. You are paying to rebuild the templates the export never contained.
The help page already told you
The biggest canvas builder says it in plain language. Your site has to run on their servers. Content is yours. The site is not.
You can dump collections, contacts, physical products (capped at 5,000 rows), and orders as CSV. Digital products stay put.
The pretty-template host will give you an XML file another CMS can import. You get layout pages, one blog, and text and image blocks. You do not get the store, extra blogs, headers, footers, style settings, or custom CSS.
Trial sites cannot export. Images in a self-hosted import often still point at their live URLs. Those links die when you cancel.
On that export article, 1,185 of 2,015 people marked it useful. Forty-one percent did not.
The design tool that sells you "code" will sell you a ZIP. Paid workspace only. The workspace that includes export starts at $19 a month on the yearly plan, and you still pay a site plan to stay published.
The ZIP has HTML, CSS, JS, and assets. It does not have CMS content, collection pages, ecommerce, translated pages, forms, or search. Lists render empty.
The motion-first designer says data is never locked in. Independent reviews in 2026 agree with the community. You can take a CMS CSV, images, and copied overrides. You cannot take a working site.
W3Techs, 17 August 2026: the open-source CMS still runs 40.8% of all websites. Hosted builders plus the big store platform sit on about 12% of the known CMS market. Almost none of those hosted tools hand you a working site on the way out.
That is a volume-of-trapped-sites number, not a collapse story. Those companies have grown for a decade.
When the invoice changes
October 2025, the motion-first designer collapsed five plans into three. Basic dropped from $15 to $10 a month.
Pages went from 1,000 to 30. Monthly transfer went from 50 GB to 10 GB. CMS collections went from two to one. About 10% of Basic users with two collections got pushed onto Pro at $30.
May 2026, the design-to-code tool merged plans. Their own pricing FAQ includes a case where the "cheaper" new plan costs more because of how transfer is billed. Code export is still a workspace upsell.
On 26 August 2026, the store platform cuts off custom checkout scripts. Theme files stay yours. Checkout customization does not.
None of that is a scandal. It is what happens when they already have the sites.
History already ran this experiment. A popular builder got bought by a payments company. Editor support was slated to end in July 2025. There is no automated move onto the parent's store product. About 0.4% of all websites still sit there.
A desktop site builder from a major creative-software vendor ended support in March 2020. People were still mid-update years later. They rebuilt.
When the company owns the tool, the format, and the hosting, your work lives on their roadmap.
Own the code is the wrong test
People treat "own your website code" as the exam. For most marketing sites, it is the wrong one.
Most marketing sites get redesigned every two or three years. A ZIP of last year's hover states is worth less than structured content, a URL map, brand tokens stored outside the builder, and first-party contacts and orders you can actually export.
The paid ZIP is the proof. You can download "the code" and still lose the CMS, the store, the forms, and the memberships. That is the business.
The new lock-in is not the visual editor. It is the prompt plus the backend.
One of the big builders launched a prompt-to-site product on 21 January 2026. Those sites are native sites on the same platform. You cannot switch that output to another template. You start a new site.
Their marketing says you are never locked into one mode of editing. That is not the same as leaving.
The same company bought a full-stack app generator in June 2025. Reviewers keep flagging the opposite of the git-export tools: database, login, and hosting are bundled, so getting data out is hard. That is the product.
In July 2026 they opened a path where a portable frontend can sit on their payments, bookings, and CMS. The frontend can move. Commerce, members, and bookings stay.
The prompt-to-repo tools advertise no lock-in via a git repo. Soft lock-in is billing and the hosted database, not the React files. In June 2026 a paid user sat through a 20-plus hour outage and could not export without credits.
AI-generated sites are now a measurable CMS category. One generator already shows up at 0.1% of all websites on W3Techs.
Also, do not treat "just host it yourself" as a magic answer. The open-source CMS is portable. The page-builder layer people bolt on top of it produces the same rebuild.
What travels is the content database, not the canvas. A portable foundation is a content model, a design system, and data you can leave with. Not a CMS logo.
Plenty of businesses should stay on the all-in-one builder. Booking plus payments plus a site in one login is a real product. The point is knowing which layer you can move before the invoice changes.
| Built on | Walks out | Stays behind |
|---|---|---|
| Canvas builder | CSVs and a domain | Layout, design, members |
| Pretty-template host | One blog as XML, some pages | Style, store, extra blogs |
| Design-to-code tool | Static ZIP on a paid plan | Live CMS, store, forms |
| Store platform | Theme files, product CSVs | Checkout, apps |
| Prompt-to-site tool | A repo, if billing works | Hosted database and login |
| Self-hosted CMS | Files and database | Page-builder shortcodes |
This afternoon
Before you add another page, export whatever the builder will let you export.
site-asset/
content/
posts.csv
products.csv
contacts.csv
brand/
logo.svg
fonts/
tokens.txt
urls.txt
Put that folder somewhere you control. If you ever migrate off a website builder, the URL list is the redirect map so old links still work.
If the builder cannot produce the folder, the next page you add is more inventory someone will retype.
That folder is the asset. The template is the rental.
We build sites this way for clients: content you can export, a design system that is not trapped in a canvas, and hosting you can leave. First conversation is free. If you want the checklists we publish with these posts, the membership at kief.studio is free.