How do the most popular websites set up their link previews? On 2026-11-17 we fetched the homepage of each of the top 10,000 domains of the Tranco list, read its Open Graph and Twitter Card tags with the parser behind our OG Checker, downloaded each og:image and measured it against the limits Facebook, LinkedIn, X and WhatsApp document. 5,927 homepages could be read; every share below says what it is a share of.
Share of the 5,927 readable homepages that declare each tag (first value counts; empty content does not).
| Finding | Share | Count |
|---|---|---|
| og:title | 65.5% | 3,885 of 5,927 homepages |
| og:description | 62.1% | 3,682 of 5,927 homepages |
| og:image | 58.0% | 3,438 of 5,927 homepages |
| og:type | 56.9% | 3,372 of 5,927 homepages |
| og:url | 57.2% | 3,390 of 5,927 homepages |
| og:site_name | 48.0% | 2,846 of 5,927 homepages |
| og:image:alt | 7.8% | 461 of 5,927 homepages |
| twitter:image | 31.4% | 1,863 of 5,927 homepages |
| twitter:card | 49.4% | 2,930 of 5,927 homepages |
The top 1,000 domains against the rest of the list: 34.7% of the top homepages and 42.7% of the others have no og:image.
| Finding | Share | Count |
|---|---|---|
| Top 1,000 | 34.7% | 180 of 519 homepages |
| The rest of the top 10,000 | 42.7% | 2,309 of 5,408 homepages |
Of the 3,438 homepages with an og:image. A relative or protocol-relative address works in a browser but crawlers are not required to resolve it; several og:image tags leave the choice to each platform.
| Finding | Share | Count |
|---|---|---|
| Does not load as an image | 4.4% | 150 of 3,438 og:images |
| Relative URL (/img/og.png) | 3.8% | 132 of 3,438 og:images |
| Protocol-relative URL (//cdn…) | 0.8% | 28 of 3,438 og:images |
| More than one og:image tag | 3.8% | 130 of 3,438 og:images |
Share of all 5,927 readable homepages whose og:image loads, has a known pixel size and is within every limit the platform documents (as on our OG image size page, checked 2026-11-02). No image or a broken one counts as a fail.
| Finding | Share | Count |
|---|---|---|
| Facebook: within every documented limit | 43.0% | 2,549 of 5,927 homepages |
| LinkedIn: within every documented limit | 34.1% | 2,022 of 5,927 homepages |
| X (summary_large_image): within every documented limit | 45.1% | 2,673 of 5,927 homepages |
| WhatsApp: within every documented limit | 45.1% | 2,676 of 5,927 homepages |
| Facebook, LinkedIn and X at once | 31.8% | 1,883 of 5,927 homepages |
Size limits as a share of the 3,266 og:images with a known pixel size; file-size limits as a share of the images whose length we know. Below 200×200 Facebook shows no image at all; below 600×315 it shows a small one instead of the large card.
| Finding | Share | Count |
|---|---|---|
| Smaller than LinkedIn's 1200×627 | 38.0% | 1,242 of 3,266 measured images |
| Smaller than Facebook's 600×315 (no large card) | 22.0% | 717 of 3,266 measured images |
| Smaller than Facebook's 200×200 (no image at all) | 7.8% | 254 of 3,266 measured images |
| Larger than WhatsApp's 600 KB | 5.9% | 193 of 3,287 images of known length |
| Larger than LinkedIn's and X's 5 MB | 0.1% | 2 of 3,287 images of known length |
| Larger than Facebook's 8 MB | 0.0% | 0 of 3,287 images of known length |
| A format X does not list (JPG, PNG, WebP, GIF) | 7.2% | 237 of 3,288 loaded images |
Format as read from the file header. The median og:image weighs 42 KB (one in ten is over 370 KB) and is 1,200 pixels wide.
| Finding | Share | Count |
|---|---|---|
| image/png | 48.4% | 1,590 of 3,288 loaded images |
| image/jpeg | 29.6% | 973 of 3,288 loaded images |
| image/webp | 14.2% | 467 of 3,288 loaded images |
| image/avif | 5.5% | 182 of 3,288 loaded images |
| image/svg+xml | 1.6% | 52 of 3,288 loaded images |
| image/gif | 0.6% | 21 of 3,288 loaded images |
Of the 3,266 measured images. Facebook and LinkedIn recommend 1.91:1; the X large card is 2:1; a square image is cropped or shown small.
| Finding | Share | Count |
|---|---|---|
| 1.91:1 | 47.4% | 1,547 of 3,266 measured images |
| 2:1 | 3.8% | 125 of 3,266 measured images |
| 1:1 | 18.9% | 616 of 3,266 measured images |
| Any other ratio | 29.9% | 978 of 3,266 measured images |
The twitter:card value of the 5,927 readable homepages. Without summary_large_image, X and several chat apps show a small thumbnail beside the text.
| Finding | Share | Count |
|---|---|---|
| No twitter:card | 50.6% | 2,997 of 5,927 homepages |
| summary_large_image | 38.5% | 2,284 of 5,927 homepages |
| summary | 10.3% | 608 of 5,927 homepages |
| player | 0.0% | 1 of 5,927 homepages |
| app | 0.3% | 16 of 5,927 homepages |
| other | 0.4% | 21 of 5,927 homepages |
| Homepage read (analysed) | 59.3% | 5,927 |
| robots.txt disallows us (not fetched) | 2.7% | 269 |
| robots.txt answered with a server error (treated as a disallow, per RFC 9309) | 0.6% | 63 |
| Blocked as a bot (unauthorised, forbidden, rate-limited or a challenge page) | 14.2% | 1,420 |
| Another HTTP error (not found, server error…) | 2.7% | 271 |
| Not an HTML page (an API, a file, a redirect to an app) | 0.4% | 41 |
| Did not answer (timeout, TLS or connection error) | 6.9% | 692 |
| No public web address (a DNS-only or infrastructure domain) | 13.2% | 1,317 |
One row per domain (rank, outcome, final address, every tag flag, the image's status, type, bytes and pixels, and each platform verdict), and the aggregate every number on this page comes from. Free to reuse under CC BY 4.0 with a link to this page.
Shareglint, "State of Open Graph images 2026", census of the top 10,000 Tranco domains (list K97GW), 2026-11-17. https://shareglint.com/en/research/state-of-open-graph-2026
The OG Checker runs the same parser and limits on any page; the size page lists every platform's documented limits.
In our census of 2026-11-17, 42.0% of the 5,927 homepages we could read among the top 10,000 Tranco domains had no og:image tag. Shared on social media or in chat, a link to them shows no picture or one the platform picks itself.
38.0% of the og:images we could measure were smaller than 1200×627, the minimum LinkedIn documents. Facebook's large card needs 600×315; 22.0% were smaller than that.
4.4% of the og:images that homepages declare did not load as an image when we requested them. A further 3.8% use a relative address and 0.8% a protocol-relative one, which crawlers are not required to resolve.
40.7% of the 10,000 domains (4,073) could not be analysed: robots.txt disallowed us (2.7%), the site blocked us as a bot (14.2%), or the domain has no website (CDNs, DNS and ad infrastructure are high on the list). Each reason is counted in the methodology.
Yes. The per-domain CSV and the aggregate JSON are free to reuse under CC BY 4.0; link to this page as the source.