A caption is the text underneath your post, and its job is different from the one most people give it: it does not have to sell your product in every post, it extends the attention the image or the reel has just won. That is why it gets built back to front. The opening carries everything, because Instagram folds the rest away behind "more" in the feed, and whoever burns it on a run-up never gets the third sentence read. Technically you have plenty of room: Meta documents a maximum of 2,200 characters, 30 hashtags and 20 @ tags for the caption field of its publishing API. In practice you rarely need it, because a caption has to add exactly one thing, namely the thing there was no room for in the video. At the end stands a single next step, not three. And the hashtags belong in the caption itself, not in the first comment, because Instagram writes about its own search in plain words: "For a post to be found in Search, put keywords and hashtags in the caption, not the comments." The point everything hangs on, though, comes before the first word: a caption written after the reel is finished reads like an afterthought, and you can hear it.
I no longer write my own captions, but I supply the raw material for them. What I collect about AI at work sits on the AI page; the full reel workflow at my place, from idea to approval, is in Can AI Create Instagram Reels?. One distinction up front so we do not talk past each other: the hook is the opening inside the video, the first second and a half of frame, on-screen line and first spoken sentence. That is the subject of How to Write Strong Instagram Reel Hooks. The caption is the text under the post. Both sit on the same screen at the same time, both get roughly the same half second, and that is exactly why they must not say the same thing.
What a caption has to do
Three things, in this order.
It holds the attention the image created. Nobody reads a caption before seeing the post. The attention is already there when your text starts, and your only job is not to let it drop.
It delivers the part that did not fit in the post. A 35-second reel inevitably leaves something out: the exception, the number, the second case, the sentence a client said word for word. That leftover is your caption material, and it is already sitting on the table the moment the script is done.
It makes the post findable. Instagram describes its own search as matching what you type against "relevant usernames, bios, captions, hashtags and places". The caption is therefore one of the few places where you can say in words what this is about. A reel without text leaves that route unused.
What a caption is not: a product description, a summary of the video for people who have just watched it, or a storage shed for thirty hashtags. Instagram names the viewer's own activity, meaning what they like, share, save and comment on, and the information about the post itself as important groups of feed signals. A caption therefore works on two routes, both of which can be evidenced: it gives search the words your post can be found by, and it gives the reader a reason to save or to reply. That is a sober expectation and a good reason to take the text seriously.
The first two lines
How much stays visible before the "more" depends on device, format, type size and feed layout. Count on one or two lines and you will rarely be far off. Those lines are not an introduction, they are the whole post in miniature. Three rules are enough.
No run-up. "Today I want to tell you about something that happened to me this week" uses both lines and says nothing. Start with what happened.
Do not repeat the hook. Hook and caption opening sit on the screen at the same time. Writing the same sentence twice throws away one of two chances. The hook stops the thumb, the caption opening answers the question the hook opened, or asks the next one.
One thought, not a paragraph. Two lines are roughly a sentence and a half. Squeeze three thoughts in and you lose all three.
On the left there is an announcement, a promise of text that only arrives after the tap. Nobody taps for a promise. On the right there is already content, and the reason to keep reading appears as a by-product.
How long may a caption be?
The documented limit comes from Meta itself: a maximum of 2,200 characters, 30 hashtags and 20 @ tags. The figure sits in Meta's developer documentation and applies to publishing through the API described there; go over it and you get an error instead of a post. Whether the app draws exactly the same line in its own input field, Meta does not say in that place. For practical purposes it makes no difference: it is a ceiling, not a recommendation.
In practice there are three lengths, and you choose between them not by feel but by how much your image has already told.
The same craft applies to all three: use line breaks, do not write a block. The most common length error is not too long anyway, it is undecided: a caption that starts on one thought, jumps to a second halfway through and squeezes in three offers at the end. Then length is not the problem, it is that three captions are stuck inside one.
What goes at the end
Exactly one next step, and it has to match what was just in the video.
- A question, when you want to know whether the topic is a topic for your people at all. It has to be answerable: "Which of the three is the one that trips you up?" works, "What do you think?" does not.
- A pointer to another post, when the thought continues.
- A concrete offer, when the post pays into a service. Not every time: if every caption ends that way, soon nobody reads that far.
- Nothing at all, when the post stands on its own. That is right more often than people think.
There is one limit here and it is not a matter of taste: Meta lists "engagement bait" as its own category in its content distribution guidelines, meaning posts that explicitly ask for likes, comments, shares or tags without a real occasion for it. Accounts that do this repeatedly are demoted in distribution, according to the same source. "Tag three friends" is therefore not merely worn out, it is named. A genuine question about the subject of the post is something else.
Hashtags: what they do and what they do not
More half-knowledge collects here than around any other part of a caption, so only what Instagram and Meta say themselves.
Thirty is the ceiling, from the same Meta specification as the character limit. It is not a target.
They belong in the caption, not in the first comment. Instagram writes about its own search word for word: "For a post to be found in Search, put keywords and hashtags in the caption, not the comments." That sits in Instagram's search explainer of 25 August 2021 and is therefore an older recommendation from the provider; I found no newer version of it. The widespread habit of parking the hashtag block in the first comment costs you precisely the findability the hashtags were there for.
It is not only the hashtags that count, it is the words. Instagram recommends putting relevant keywords into your content, your caption, your bio and your hashtags. Whoever writes the caption in the words their clients actually use has done the larger part already.
Off-topic hashtags hurt. Instagram writes that topic-relevant hashtags can help reach more interested people, but warns in the same breath against trying to game the system: if content is shown to audiences who are not interested in it, engagement and distribution may suffer.
What follows is an unspectacular practice: few hashtags that genuinely name the subject, at the end of the caption, with the same terms appearing as ordinary words in the text above. If you want clients in your own language, use hashtags in that language.
The caption does not come out of the reel
Most people work like this: film the reel, cut it, watch it, write a caption to go with it. Anyone who looks at a finished video and then goes looking for text has only one source left, the video, and therefore writes a summary of it. Nobody reads a summary of the thing they have just watched. Good captions come from one stage earlier, from the same observation the reel came from. The workflow is simple:
Step 1: Write the observation down before anything else exists. Five to eight raw lines, with a date and what somebody actually said. A constructed example, not a case from my own business, so you can transfer the form:
Tuesday: third enquiry this week for application photos, all three
asked about the price first and not about the date.
One wrote, word for word: "I actually need this by Friday."
For two the date was the real problem, not the price.
For the third it was not, she really was only comparing.
Step 2: The reel gets part of it. Thirty-five seconds hold the observation, the most likely reason and one sign the viewer can spot for themselves. No more.
Step 3: The caption gets the rest. Specifically the part that would get in the way in the video, which is also the most interesting part: the third enquiry, the one where it was not the date after all. The exception never fits into a reel, because it makes the reel complicated. In a caption it makes the text credible, because it shows you are observing rather than selling.
Step 4: Only now build the first two lines. Out of step 3, not out of step 2.
The difference is visible from outside. A caption written out of the reel says: "As shown in the video, it is usually the date." A caption written out of the observation says: "With the third one it really was the price. Here is how I could tell." The second one carries information that was not in the video, and that is exactly what somebody taps "more" for.
Who writes the captions at my place
In my operation an AI employee drafts them. Conny takes a content assignment, reel, hook, script, carousel or caption, and delivers a draft ready for approval in my voice, checked against audience, content pillars and the house rules. Behind her hook rulebooks sit 15,536 analysed Instagram hooks, and from the analysis of my own account she knows which formats work here and which do not. For captions something else counts, and it is the point of the next paragraph: she works from the same raw note as for hook and script.
Conny does not get the finished reel and write a caption to go with it. She gets the observation note, the same five to eight raw lines the hook and the script come from. Whoever sees only the video can at best summarise what is in it. Whoever sees the raw note can pick out the part that did not make it in, and that is what a caption is for.
The second half matters just as much: Conny invents no personal facts about me. No anecdote, no number, no client detail. If my note does not say how the third enquiry sounded, the draft has a gap in that spot rather than a plausible invention. I fill it in thirty seconds, because I was there. She drafts, I approve.
How to set up a role like this with a rulebook instead of a pile of prompts is in Hiring an AI Employee: The Process, the craft behind it in CLAUDE.md: The Memory of Claude Code, and the way the roles are divided at my place in My AI Workforce. Which work suits a handover like this at all, I sorted through in Which Tasks You Can Hand Off to AI.
Frequently asked questions
How many characters can an Instagram caption have?
A maximum of 2,200 characters, plus no more than 30 hashtags and 20 @ tags. Meta states these values in its own developer documentation for the caption field of its publishing API; go over them there and you get an error. For typing into the app itself Meta gives no figure in that place. For everyday use it makes no difference: good captions sit well below it.
Where do hashtags belong, in the caption or the first comment?
In the caption. Instagram writes explicitly about its own search that keywords and hashtags have to be in the caption and not in the comments for a post to be found. The tidy hashtag block in the first comment therefore costs you exactly the findability the hashtags were meant to provide.
How many hashtags should I use?
Fewer than are allowed, and only ones that genuinely name the subject of the post. Instagram warns against trying to game the system with off-topic hashtags, because engagement and distribution may suffer when content is served to people with no interest in it. More important than the count is that your topic words also appear in the body of the caption.
Is an Instagram caption generator any good?
For the form yes, for the content no. A generator that only knows your topic hands you a caption that fits a thousand posts, and it shows. The same technology becomes useful the moment it gets your raw note: the five lines with a date, a quote and the exception. Then it phrases your case instead of inventing a generic one.
What to do next
Take the next post you are planning anyway and reverse the order. Write the observation down first, five to eight raw lines with a date and a quote, before you touch the camera. Then mark what goes into the video and deliberately leave the rest standing: the exception, the number, the second case. Build the caption out of that, before the reel is cut. If you then want to stop explaining that on every single post, the next step is a rulebook instead of a pile of prompts. How I set that up, and what Conny's template looks like, is in my community.
