You wrote a good sentence, put it on the graphic, and it does not land the way it did in your head. It might be three sentences where it should be one. It might be a clever line that takes a beat too long to decode. On a page, a reader gives words time. On a feed, they give a graphic a fraction of a second, and copy that works on the page can die on the post for reasons that have nothing to do with the writing being bad.
Copy on a graphic is not the same craft as writing prose, because it is read under different conditions and it is part of a design. The words and the layout are not two jobs handed to two people; they are one object, and the copy has to be written for the space, the speed and the size it will actually appear at. Treat the text as design and it gets shorter, sharper and clearer, which is the whole game. This is the copy half of the design principles, and it leans on everything in typography for content about how those words get set.
One message per graphic
The single most useful rule is also the hardest to obey: one graphic, one message. A viewer glancing at a post can take away exactly one idea, so decide which idea that is and cut everything that is not it. Two messages on one graphic do not give you two chances to land; they give you nothing, because the eye cannot hold both in the half-second it has and takes away neither.
This is the copy version of guiding the eye to one thing. If your post is trying to announce a launch and explain a feature and share a testimonial, it is three posts wearing one square. Split them. When you are tempted to add "and also", that is the signal you have started a second graphic. Make it a second graphic. The discipline is subtractive: write down everything you want to say, pick the one thing that matters most, and delete the rest without mourning it.
Headline and support line
Most graphics need two levels of copy and no more: a headline that stops the scroll, and a support line that pays off the stop. They do different jobs and should not be confused.
The headline is the hook. It carries the one message and it has to work at thumbnail size, glanced at, before anyone has decided to care. It is the largest, boldest text on the canvas, and it is short enough to be read in one look. Everything known about what makes a scroll stop applies here: how hooks work is the deeper treatment, and the same instincts drive a headline on a graphic as drive the first line of a video.
The support line is the smaller second level that adds the one piece of context the headline earned the right to deliver: the specific, the proof, the where-and-when. It is set smaller and quieter, it comes second in the reading order on purpose, and it is optional. If the headline says everything, do not add a support line for the sake of symmetry. Silence is cheaper than a redundant sentence, and the layout breathes better without it.
Brevity is not optional
On a graphic, every extra word costs you twice. It steals the half-second of attention you were given, and it forces the type smaller so all the words fit, which makes each one harder to read. Short copy is therefore both easier to take in and physically more legible, because fewer words can be set larger. Length works against you on both axes at once.
So write it long, then cut it hard. Say the thing in a full sentence, then remove every word the sentence survives without. Drop the throat-clearing opener. Cut "in order to" to "to", "the fact that" to nothing, the adjective that is not doing real work. A headline that started at fifteen words is almost always stronger at six. The test is not "is this a good sentence" but "can someone read this in a glance and get the one message". That test is failed by length long before it is failed by wording.
Verbs over adjectives
Weak copy piles up adjectives (amazing, powerful, revolutionary, seamless) and hopes the adjectives do the persuading. They do not, because they are claims the reader has heard a thousand times and discounts on sight. Strong copy uses verbs and concrete nouns, and shows the thing instead of praising it.
The difference is trust. "An amazing, game-changing new tool" tells the reader you think highly of yourself. "Cut your edit time in half" tells them what happens to them, in a verb they can picture. The second is shorter, more specific and more believable, and it does not ask the reader to take your word for how impressive it is. When you catch an adjective doing the heavy lifting, replace it with the action or the fact underneath it. "Beautiful results" becomes the result shown; "effortless setup" becomes "set up in two minutes". Prefer the concrete every time, because the concrete is what the reader can check, and checkable copy is trusted copy.
One call to action, if any
If the graphic wants the viewer to do something, ask for one thing, plainly, and only one. A post with three asks (follow, and visit the site, and comment below, and book a call) gets none of them done, because a divided instruction is easier to ignore than a single clear one. Decide the single action that matters most for this post and make that the only thing you request.
Keep the ask short and literal. "Book on Saturday", "Link in bio", "Save this for later". The reader should know exactly what to do without decoding it. And be honest about whether the graphic needs a call to action at all. Plenty of the best posts ask for nothing; they earn attention, deliver the one message, and leave. A call to action bolted onto a post that had no reason for one just adds a line of clutter and dilutes the message you actually came to deliver. When you do include one, it is a level of copy in its own right: smaller than the headline, clearly the last thing read, and never competing with the message for the eye.
Copy and design are one object
Everything above only works because the words and the design are decided together. The headline's length is a typographic decision, because it sets how large the type can be. The support line's presence is a layout decision, because it changes what the composition has to hold. The call to action's placement is a hierarchy decision, because it has to come last. Write the copy as if you are also setting it, and set the type as if you also wrote it. That is why the same person, or at least the same pass, should do both.
The practical habit is to draft the copy and rough the layout in the same sitting, not to write a paragraph and hand it off to be "designed". When you see the words at the size they will really appear, you cut differently and better, because the space itself tells you what is too long. If you are running variants to see which line lands, treat it the way you would any test: testing hook variants applies as much to a line on a graphic as to a video hook. And wherever you assemble the post, whether that is Canva, Figma or Studio, write the copy to the space rather than pouring finished prose into it after the fact.
Do this today
Take the post you are working on and count the messages on the graphic. If there is more than one, cut it to the single most important, and split the rest into their own posts. Then rewrite the headline in six words or fewer, swap any adjective for the action or fact underneath it, and decide whether the post needs a call to action at all, and if it does, make it one. Read what is left at thumbnail size. If the one message lands in a glance, the copy is doing its job.