Your freshly generated OG image looks perfect in the browser, but Facebook still shows yesterday's version — or nothing at all. Platform crawlers read raw HTML, ignore JavaScript, cache aggressively, and fail silently.
Client-side-injected meta tags are invisible to them. If your React app or single-page framework injects og:image after hydration, the crawler sees nothing.
Use curl with the exact user agent that Meta's crawler sends:
curl -H "User-Agent: facebookexternalhit/1.1" \
-H "Range: bytes=0-524288" \
-s https://your-site.com/page | grep -i "og:\|twitter:"
Meta's crawler requests may include a Range header for a byte span, which a server should honor precisely or ignore altogether [1]. Meta's documentation says you can simulate its crawler with a curl request using the facebookexternalhit/1.1 user agent and a Range header of bytes=0-524288 [2]. Do not use DevTools or view-source for this check — both run JavaScript and show you the hydrated DOM, not the crawler's raw HTML.
Open Graph properties must appear within the first megabyte of the document or the crawler cuts them off [1]. If your HTML payload starts with a massive inline script, stylesheet, or JSON blob, your meta tags may fall past that threshold. Move them into <head> and keep the document lean above the fold.
Missing, duplicate, or wrongly-attributed tags. Open Graph uses property="og:image", not name="og:image". Twitter uses name="twitter:image", not property. The deprecated og:image:url still works but og:image is preferred. If you have multiple og:image tags, most crawlers take the first and ignore the rest.
Relative instead of absolute URLs. og:image must be an absolute https:// URL. Crawlers resolve relative URLs against the page URL, and the resolution breaks in surprising ways across platforms. Always use https://your-cdn.com/image.png, never /image.png or image.png.
Wrong size or format. The 1200×630 ratio is the safest default. Facebook enforces minimum dimensions and a maximum file size. LinkedIn prefers a slightly different ratio and a smaller file size, accepts JPG/PNG/GIF, and does not reliably render WebP. X needs twitter:card set to summary_large_image or falls back to a small thumbnail. WhatsApp has much tighter practical limits on file size and aspect ratio — exceed them and it silently drops the card entirely.
Bot-blocking rules. Cloudflare challenges, WAF rules, hotlink protection, and authentication walls all stop crawlers cold. Apple's iMessage link previews send a user agent containing both facebookexternalhit/1.1 and Twitterbot/1.0 [1], so blocking either of those strings breaks iMessage previews too.
Redirect chains and wrong Content-Type. A 200 status with an HTML error page inside, text/plain on a PNG URL, or infinite redirects all read as failures to the crawler. The image itself must return image/jpeg or image/png with a proper 200.
Slow or cold-start image endpoints. Crawlers time out quickly. If your image generation starts on first request, the crawler may cache the absence of an image. Your own page cache or CDN can also serve stale HTML before the crawler even reaches the image URL.
| Platform | Typical cache duration | How to force refresh |
|---|---|---|
| Long-lived (weeks) | Sharing Debugger "Scrape Again"; Graph API ?scrape=true; HTTP/HTTPS cached separately | |
| About a week | Post Inspector | |
| X | Roughly weekly | Card Validator retired — DM the link to yourself |
| Slack | Short (under an hour) | Natural expiry only |
| Discord | Short (under an hour) | Natural expiry only |
| No fixed expiry | Per-exact-URL cache; new chat, add ?v=2, or wait | |
| iMessage | Per-device | On-device cache with no invalidation |
| Telegram | Varies | Message @WebpageBot with your URL |
Facebook keys its cache on og:url as well as the page URL, so changing og:url without changing the image URL may not refresh the image. Existing shares keep their cached data even after you scrape again — only new shares pick up fresh data.
The universal fix across every platform is changing the og:image URL itself. Append ?v=2, ?v=3, or use a content-derived filename. Platforms treat this as a completely new image and fetch it fresh. Do not trust your own browser or social feed as verification — the platform's own debugger preview is the only reliable source.
For programmatically generated images, make the image URL content-derived so any change auto-busts every platform cache. MarkupGo's Magic Template URL feature lets you generate images from existing templates by passing data and options as query parameters, with no code [3].
The Magic URL structure is https://render.markupgo.com/template/{templateId}.{extension}?{query}, with supported output formats png, jpeg, webp, and pdf [3]. The query string itself becomes the cache key — change any parameter and every platform sees a new URL.
For a blog post template, the URL might look like:
https://render.markupgo.com/template/blog_og.png?title=Debugging+OG+Images&author=Sarah+Chen&date=2026-11-16
Reusing the same Magic URL within 15 days serves the cached image without consuming credits; after 15 days the cache clears and regeneration consumes credits [3]. Every unique Magic URL generates a new image and consumes credits [3].
Warm the URL before publish to dodge scraper timeouts. Hit it once yourself, then let the 15-day server-side caching handle repeat crawler hits from Facebook, LinkedIn, and others. Magic URLs are public and accessible to anyone with the URL, so treat template IDs accordingly — don't embed sensitive data in parameters [3]. MarkupGo lists OG images as a specific use case for Magic URL [3].
facebookexternalhit user agent and grep for og:image — confirm absolute https URL, 1200×630 target, and position within first 1MBimage/jpeg or image/png Content-Type, loads quicklyog:image URL if refreshing — add new ?v= parameter or filename, never rely on debugger alone