📄 Website to Markdown Generator

Custom mode, which gives users 16 tick-box groups to select exactly what sections to include, was locked behind an administrator check that was never actually requested. The lock existed in three separate layers: a disabled, greyed-out button on the page; an email check in the page logic that only unlocked it for one specific administrator address; and a server-side route guard that refused non-administrators. Any one of the three layers would have been enough to block a normal user. The fix rewired the button to the same shared free-tool opener that every other tool on the page uses, removed the email check from the page logic, and dropped the administrator guard from the server endpoint. The end state is that any user can open Custom mode with all 16 tick-box groups, charged exactly once on Generate, through the same mechanism every other free tool already uses. No new routes were added or removed; the only change was the removal of the administrator restriction.

A test result appeared to show that Stripped mode was catastrophically broken: a WordPress page of 118 kilobytes came out as just a few hundred bytes of Markdown. A result like that demands to be taken seriously. Looking at what had actually happened revealed that the web address being tested was dead. It no longer pointed at a real article; it returned WordPress's standard page-not-found error, which has almost nothing on it. The 118 kilobytes were almost entirely the site's framework: its scripts, styles, navigation and footer. The actual readable content on that error page amounted to a couple of sentences. Stripped had done precisely the right thing. It stripped the framework and converted the tiny amount of real content that remained. The tool was never at fault; the test input was. The same broken test did incidentally reveal one genuine issue: a Google Tag Manager tracking frame leaking into the output, which is addressed in the next section.

A Google Tag Manager tracking frame was appearing right at the top of the Markdown output in all three modes. It is plumbing: a hidden frame whose only job is to let a tag manager run when JavaScript is switched off. The converters already removed scripts and styles but not iframes or noscript blocks. The tag manager hides its frame inside a noscript block, so both element types needed adding to the removal step. The fix adds noscript and iframe to the cleanup call in all three modes, before any conversion runs. Removing these is completely safe, because a tracking frame and a noscript fallback are never the page's real content. The removal runs before the three modes diverge, so all three produce output that starts with the first real content element.

On a number of sites, the header and navigation menus appeared in the Stripped output. The converters already removed the standard HTML navigation element, but many sites build their headers and menus from ordinary containers with names like mega-menu, site-header or header-nav, or mark them only with an accessibility role. Those do not announce themselves as navigation in the way the converter was looking for. The fix targets reliable patterns that identify navigation regardless of the specific site: the page-level header element (scoped to a direct child of body so in-article headers survive), containers whose names contain site-header, mega-menu or header-nav, and elements marked with the navigation accessibility role or with names like menu-drawer, mobile-menu or dropdown-menu. The word menu alone is deliberately not matched, because a restaurant's menu is genuine content. This applies to Stripped mode only; Full keeps all furniture, Custom makes it a tick group.

On one site, a phrase appeared doubled in the Markdown output. Duplicated text makes the output look careless and, if fed to an AI, distorts what the page actually says. The cause was a decorative scrolling banner: the kind of ticker that slides text across the screen in a continuous loop. To make the loop seamless, sites duplicate the text in the underlying markup. The converter, reading the markup rather than watching the animation, printed both copies. The first two attempts at a fix guessed the cause was a drop-down menu and removed that instead. Both failed. The real culprit was found only by inspecting the live page: the text sat inside an element explicitly marked as scrolling-text. The fix removes decorative scrolling banners by pattern: names containing scrolling-text, marquee, ticker or scrolling-banner. These elements are always decorative and always duplicate their text as part of how the scroll works. Two wrong guesses cost two deploys; a single inspection found the correct class immediately.

Some sites, shops in particular, leaked pop-up dialogs into the Stripped output: slide-out shopping carts, cookie or location-consent boxes. Many of these overlays are built with utility-style CSS that carries no obvious pop-up class name. The reliable signal is the accessibility marking: even when a pop-up has no useful class name, it must declare itself as a dialog for screen-readers to handle it correctly. So these overlays carry a dialog role or an is-modal attribute even when their class names are meaningless. The fix targets those accessibility markers plus the structural names that shops use for cart and consent drawers. The combination catches the overlays whether or not they were given a helpful class name: the accessibility markers catch the generically-styled ones, and the structural names catch the shop-specific drawers. A dialog role marks an overlay by definition and is never used for page content, so removing it is safe.

Skip to content links leaked into the output. These are keyboard-navigation shortcuts that are useful on a live page but are not content, so in a clean conversion they do not belong. Catching all of them took three layers, each added because the previous ones missed a real case. Some skip-links carry a helpful name like skip-link or skip-to-content and a rule keyed on that name catches them. Many have no useful name and can only be recognised by their text: any link whose words begin Skip to. The hardest case, found on Magento, wraps the skip text in a bare nameless element that is not even a link. It has no skip-link class, no link address, nothing the first two layers look for, just the words sitting in a plain span. The three layers together catch every variant found across all platforms: by name, by link text starting with Skip to, and by exact phrase match on a bare element anchored start to end so only the standard accessibility phrases match and real content is safe.

A test suggested numbered lists were being mishandled. The mismatch was mostly a false alarm. Several site builders, Squarespace prominent among them, use numbered-list markup to build things that are not visible numbered lists: post carousels, sliders, category blocks. These frequently duplicate their slide text for the carousel to loop. When the converter reported more numbered-list items than the eye expected, it was accurately counting real numbered-list markup that the site used to build a slider. The count was faithful to the markup. There is one genuine edge worth documenting: when a numbered list is nested inside a bullet point inside another list, the inner numbering can be absorbed into the parent bullet in Stripped mode. The content is all preserved; only the inner numbering formatting is affected in this uncommon nesting pattern. The change needed to fix it sits in the shared list-rendering logic that every conversion depends on, so it was documented honestly as a known edge rather than shipped as a regression risk for a narrow cosmetic gain.

Many real sites ship structured data in JSON-LD that is invalid: malformed, with a syntax error that makes it unreadable. The test used a genuinely broken BigCommerce product page whose JSON-LD contained a value that was not properly formatted. The generator did exactly the right thing. It attempted to read the structured data, recognised it could not be parsed, quietly skipped the summary for that one page, and produced all the rest of the Markdown correctly. The parse error is written to the server log only. It never appears in the user's output. The correct behaviour when meeting broken structured data is to tolerate it, take what can be taken, and never let one site's malformed markup spoil the conversion of its content.

Video embeds present a specific challenge. The raw iframe that embeds a YouTube or Vimeo player is infrastructure: it is removed in the base cleanup because iframes are removed from all modes. But the video itself may be the content of the section. Silently dropping a video embed leaves a gap in the converted output that would be confusing for anyone reading or processing it. The generator handles video embeds by extracting the poster image where present and the source URL, and rendering both as a Markdown link so the output acknowledges the video and links to it. This means video content is represented in the Markdown output without requiring an embed iframe that an AI model could not display anyway.

HTML tables appear in a wide variety of markup patterns across real sites: some use standard thead and tbody, some place header cells in the first row of tbody without a thead, some use colspan and rowspan. The converter handles the common cases, extracting column headers from the first row and converting subsequent rows to pipe-separated Markdown table rows. Tables using colspan or rowspan are flattened to their text content with a note indicating that the table structure could not be preserved, which is preferable to silently producing a malformed Markdown table. The conversion is correct for the standard two-dimensional table structures that appear most frequently in real site content.

The 12 categories above were found and fixed during systematic testing across 219 pages on 11 CMS platforms. After all fixes were in place, the full test suite was run again across all 11 platforms. Zero chrome leaks were found in the final release. All three modes produce output containing only what they are defined to contain. The Stripped mode output on every platform tested starts with the first real content element of the page, contains no navigation, no tracking frames, no pop-up modals, no skip-links and no scrolling banner duplicates. Full mode output on every platform contains the complete page with tracking plumbing removed. Custom mode operates correctly with all 16 tick groups available to all users. The transparency report is published here in full so that anyone using the tool can read exactly what was found, what was done, and why.