og:image size
Every guide to this opens with 1200 by 630. It is a fine size. It also appears in neither of the primary sources that could be read for this page.
The Open Graph specification gives no size at all, and the one platform whose own documentation states that pair calls it a minimum, not a recommendation.
The specification does not give a size
og:image — An image URL which should represent your object within the graph.
That is the whole of what it says about the image itself. It goes on to define optional structured properties — width, height, type, alt — which let you DECLARE the size you chose. Declaring a size is not the same as being told what size to use, and the protocol only does the first.
Searched for directly: 1200 by 630 does not appear on that page. Neither does 1200 by 627, nor 1200 by 675. The only dimension anywhere on it is 300 by 300, inside an example showing three images of different sizes.
A minimum, read as a recommendation
LinkedIn does publish numbers, and this is the one the field repeats back with the wrong word attached:
Minimum image dimensions: 1200 (w) x 627 (h) pixels
Recommended ratio: 1.91:1
A minimum is a floor you clear. A recommendation is a target you aim at. Treating the first as the second is how people end up shipping exactly 1200 by 627 and believing they have hit a mark, when what they have done is scrape a limit.
The RATIO is the part worth keeping, because it survives whatever the platforms do to the pixel counts. 1.91:1 at 1200 wide is 628. That is where the folk number comes from and why 630 looks so close to right.
LinkedIn also publishes a hard floor nobody quotes: an image under 401 pixels wide is shown as a thumbnail rather than a card. If a share preview has silently gone small, that is the first number to check.
What two of the sources could not tell us
Meta’s sharing documentation refused an automated request, and X moved its card documentation, so neither was read on the day this was written. No number here is attributed to either of them.
Every other page ranking for this question states figures for both, confidently. Some of those figures may well be right. None of them is traceable from the page you are reading it on, and a number you cannot trace is a number somebody copied from somebody who copied it.
Defines og:image, og:image:width, og:image:height and og:image:alt. Gives no recommended size anywhere. The only dimension on the page is 300x300, inside an example showing three images.
Minimum image dimensions 1200 by 627, a recommended ratio of 1.91:1, a 5 MB file limit, and a hard floor: an image under 401 pixels wide is shown as a thumbnail.
Meta and X
Not read. Meta returns an error to an automated fetch and X moved its card documentation. Nothing on this page is attributed to either, which is why you will not find their numbers here.
The quotations, the searches run against them, and the date they were fetched are kept in the repository so this can be re-checked rather than trusted.
So what should you actually make
Three sources, one usable instruction: aim at the ratio, clear the floor, and stop optimising.
1200 by 630 is a perfectly good answer. It clears LinkedIn’s minimum, it is within two pixels of 1.91:1, and it is what this site’s own share card uses. The point is not that the number is wrong. It is that nobody can tell you where it came from, and a recommendation nobody can source is worth less than a ratio you can.
If you are generating the tags rather than hand-writing them, the tool below writes og:image alongside the title and description from the same fields, so the card and the result page cannot drift apart.