Why won't my WordPress export file import into Cursor?

Your WordPress export file will not import into Cursor for a simple reason: Cursor is an AI code editor, not a content management system, and the WordPress XML is a backup format that means nothing outside WordPress. There is nothing in a code editor built to parse it. The fix is not a different importer, it is a different format. Export your content as clean Markdown for the writing and structured JSON or CSV for the data, which an AI coding tool can actually read and build from. WPBuildAI produces that clean, code-ready export of your whole site, so the content arrives in a form the tool can use.

Note which core version produced the file, because WordPress 7.0 shipped on 20 May 2026 and was patched twice by 17 July, and an export generated either side of an upgrade can differ in ways that look like a parser bug when they are really a schema difference.

Why the XML is a dead end here

The mismatch is between what the file is and what the tool does. A code editor opens, edits, and generates code; it has no concept of “import this site.” The WordPress export is not code, it is a serialised description of a WordPress site, so the editor sees a large XML document with no instructions for what to do with it. The 2024 Web Almanac shows how much of a WordPress site is generated at runtime from the database rather than stored as plain content, which is exactly why the export is full of WordPress-specific structure a code editor cannot interpret. The content is in there, but wrapped in a format only WordPress decodes, so handing it to Cursor is like handing a database backup to a word processor: the right data, the wrong tool to read it.

What WordPress XML actually is

The export, often called WXR, is a WordPress eXtended RSS file: an XML document describing your posts, pages, custom post types, taxonomies, comments, and serialised metadata in WordPress’s own vocabulary. It exists to move a site between WordPress installations, where another WordPress on the other end knows how to interpret every tag. Outside that context, the structure is opaque: shortcodes sit unrendered as raw text, metadata is PHP-serialised rather than plain, and the content is interleaved with platform bookkeeping. So the file is not broken and you did nothing wrong; it is simply designed for a WordPress-to-WordPress journey, not a WordPress-to-code one. Understanding that it is a backup-and-transfer format for WordPress specifically explains both why Cursor cannot use it and why the answer is to re-export in a portable form.

What a code editor expects instead

A code editor and the AI inside it work with content as clean, readable text and data, not as a platform’s backup. They read Markdown as prose, JSON and CSV as structured data, and generate code from clear instructions. So the question is not “how do I make Cursor import WXR” but “what form of my content can Cursor actually use.” The answer is the plain, portable forms below. This is the same realisation behind why WordPress code does not port via GitHub: the tool wants portable inputs, not WordPress internals. Reframing the task from importing a backup to supplying clean content is what unblocks the whole rebuild, because it points you at an export the tool can read rather than a format it never will.

What to give a code editor: clean formats

Feed an AI coding tool formats it reads natively:

• Clean Markdown for page and post content, headings and text with the markup stripped.
• Structured JSON or CSV for anything tabular, products, metadata, redirect lists.

The Generative Engine Optimization study found models work best with clean, well-structured input, and that holds whether the model is answering a question or generating a site. The llms.txt convention captures the same point: stripped, plain-text content is what models parse best. Given clean content, a tool like Cursor can build the front end around your real material instead of inventing filler. The export approach is the same as turning a site into clean Markdown, and the format is the whole fix here: the content was never the problem, the wrapper was.

A worked example: the XML that became Markdown and JSON

Picture a developer who exports their blog as WXR, drags the file into Cursor, and asks it to build the site. Cursor treats the file as text, has no idea it describes a website, and produces nothing useful; the shortcodes and serialised metadata are noise. The developer then re-exports differently: each post becomes a clean Markdown file with its title and date in front matter, and a JSON file lists every post’s URL, categories, and meta description. Now Cursor reads the Markdown as content and the JSON as structure, and it can scaffold pages, wire up a blog index, and generate components around the real text. Same content, two formats, opposite outcomes. The XML was a wall; the Markdown-plus-JSON was a doorway, and nothing about the content changed except the form it arrived in.

How to produce the export

Getting clean output takes a deliberate export rather than the default backup. The goal is one Markdown file per page or post, carrying its title, date, and key metadata, plus a structured file (JSON or CSV) holding the site-wide data: the URL list, categories, and the metadata you need for redirects and SEO. Where the WordPress content contains shortcodes or builder markup, those need resolving into plain content or noting for a rebuild, since they will not mean anything to the code tool. The output you want is “content as Markdown, data as JSON/CSV, nothing WordPress-specific left in,” which is exactly the code-ready export WPBuildAI generates. Producing it is the step that turns an impossible import into a straightforward build, because it hands the tool material in its native language.

What Cursor does well once content is clean

The point is not that Cursor is the wrong tool, but that it is the right tool fed the wrong input. Once the content arrives clean, a code editor with AI is genuinely strong at the build: give it the Markdown and structured data, describe the components and layout you want, and it can generate the front end, scaffold the pages, wire the navigation, and produce clean, version-controllable code around your real content. The content arriving clean is what keeps it building your site rather than improvising one, which is the same content-quality principle as in will an AI builder import my content. So the workflow is a clean division of labour: the export supplies the content in a readable form, and the code tool does what it is good at, turning that content into a site.

The SEO step a code editor skips

A code editor builds the site, but it does nothing to protect your rankings, and that is a separate, essential step. Rankings attach to URLs and content, so the migration has to carry your metadata across and map every old URL to a 301, the core of Google’s site move guidance. Cursor will happily build a beautiful site at new URLs with no redirects, and you will lose traffic, not because the build was bad but because the migration never happened. So treat the code-ready export and the redirect map as two parts of one job: the export feeds the build, the redirects preserve the rankings. WPBuildAI exports the content code-ready, rebuilds, and maps every URL, so the rebuild keeps its traffic, the same end-to-end as converting your WordPress site to an AI website.

Common mistakes feeding content to a code tool

The recurring errors all come from handing the tool the wrong form of the content. Dragging in the WXR file expects a code editor to parse a WordPress backup, which it cannot. Pasting raw HTML or shortcode-laden text imports noise the tool then has to fight. Exporting only the body and forgetting the metadata and URL list leaves you unable to set up redirects later. And treating “Cursor built the site” as the finish line skips the migration, so the rebuilt site looks right and ranks worse. Each is avoided by the same move: export clean Markdown plus structured JSON or CSV with the metadata and URLs, let the tool build from that, and run the redirect mapping as its own step, the discipline that turns a code-tool rebuild into one that keeps its traffic.

Key points to remember

WordPress XML will not import into Cursor because a code editor is not a CMS and WXR is a WordPress-only backup format full of structure the tool cannot interpret. The fix is the format, not a different importer: export clean Markdown for content and structured JSON or CSV for data and metadata, which an AI coding tool reads natively and can build a real site from. Produce a deliberate code-ready export with the URL list and metadata included, let the tool do the build it is good at, and handle the redirects and metadata as a separate migration step, because a code editor does not protect rankings. WPBuildAI produces the code-ready export, rebuilds, and maps every URL to a 301, so the rebuild keeps its traffic; send your site URL for a fixed quote.

Not affiliated with WordPress, Cursor, or Lovable.