Original study · 271,550 posts · 14,780 accounts

Do captions affect Instagram reach? We tested AI, long and copy-paste captions.

If you've scrolled Reels lately you've seen it: a cat video, a gaming clip, an anime edit, all captioned “Tonight, V stepped into the crowd, taking in live performances at Vogue World: Hollywood.” People paste it because they think it boosts reach. Others swear by long captions, or by letting ChatGPT write them.

So we checked. We took 271,550 public posts and compared each one with the same account's other posts from the same few months. Same account, same audience, different caption.

The short answer

  • AI captions don't help. Captions that read like AI copy got about 25% fewer likes and 34% fewer comments than the same account's other captions.
  • Long captions do help, slowly. 1,200+ characters: about 9% more likes a month after posting, about 29% more after six months, and 20 to 25% more comments.
  • The viral copy-paste caption didn't work. Posts with “Tonight, V stepped into the crowd” and similar captions got 13% fewer likes than the same accounts' other posts, and won only 48% of the time.
  • A few captions do go with more likes. “Tag a friend” (+43%), giveaways (+42%), a first line that starts with a number (+7%). Selling (−14%), “link in bio” (−16%) and no caption at all (−16%) go with fewer.
  • More than 5 hashtags now costs you. Since Instagram's 5-hashtag limit in December 2025, posts that still go over it get about 9% fewer likes. Before the limit they didn't.

How we measured it

Every post against its own account.

Most caption studies compare accounts with each other: all the posts with long captions against all the posts with short ones. That mostly measures which accounts write long captions. Here, every post is compared with the same account's other posts from the same calendar quarter. A post that got 20% more likes than its account usually gets, that quarter, scores +20%. Account size, niche and anything Instagram changed that quarter cancel out.

271,550 posts from 14,780 public accounts went in: businesses, creators, meme pages and brands in many languages, mostly posted in 2025 and 2026. We left out pinned posts, paid partnerships, posts with hidden like counts and posts under a week old.

One honest limit up front: Instagram doesn't show reach for other people's posts. Likes are the closest public number. A post shown to more people collects more likes, so likes track reach, but they aren't the same thing.

Question 1: AI captions

Captions that read like ChatGPT got fewer likes, not more.

You can't prove who wrote a caption. You can judge how it reads. We had a language model read 5,014 captions from 394 accounts and answer one question: does this read like generic, polished AI-assistant copy? Think “Elevate your self-care routine with our signature treatment”, against “ok so i burnt the first batch lol”.

Likes vs the account's usual, by how the caption reads

5,014 captions of 80+ characters from 394 accounts. Each post is compared with the same account's other posts from the same quarter.

← fewer likesmore likes →

That's a 13-point gap between the account's AI-sounding captions and its human-sounding ones. AI-style copy is often selling copy too, and selling posts get fewer likes on their own, so we held that fixed, along with caption length and format. Going from “clearly a person” to “clearly AI copy” still meant about 25% fewer likes and 34% fewer comments. The range is wide (somewhere between 9% and 38% fewer likes) because it rests on 394 accounts.

The studies that come up when you search this (a 24-post test on one food account and a lab experiment) found AI captions did as well as human ones or a bit better. Real feeds look different. The cost isn't that Instagram detects AI. It's that the same smooth sentence could sit under any salon, gym or jewellery shop, and people scroll past it.

What about the em dash?

Share of captions over 100 characters with an em dash, by year posted.

3.1%

2021

1.5%

2022

3.7%

2023

4.0%

2024

12.3%

2025

11.9%

2026

It tripled when AI writing tools took off, and captions with one read more like AI copy on average. But the dash on its own barely moves likes (2 to 4% fewer, too small to be sure of). Stock AI words like “elevate” and “game-changer” made no difference at all. Deleting the dashes won't fix a generic caption.

Question 2: long captions

Long captions win, but it takes months to show.

Instagram's Adam Mosseri has said longer captions don't affect reach, and the big caption-length studies contradict each other. Our data explains part of why.

Caption lengthLikes vs usualCommentsLike rate, all accounts
No caption−6%−9%1.53%
1 to 50+2%−2%2.11%
51 to 150same−1%1.46%
151 to 300−2%−2%1.11%
301 to 600−1%+1%0.89%
601 to 1,200+2%+6%0.87%
1,200++6%+10%0.99%

Length in characters; hashtags, @mentions and links don't count. The first two columns compare each post with the same account's other posts from the same quarter. The last column is the plain median of likes divided by followers across all accounts, the way most caption studies measure it.

Look at the last column first. Across accounts, short captions get a 2.1% like rate and long ones under 1%. That's the comparison most studies publish, and it says long captions hurt. But the accounts writing long captions are different: more of them are business accounts, and they post fewer Reels. Compare each account with itself and the picture flips. Its posts with 1,200+ characters beat its usual, and comments rise from about 300 characters up.

The surprise was timing. The longer we waited after posting to count the likes, the bigger the long-caption lead got:

  • +9%

    likes when counted 7 to 30 days after posting

  • +10%

    likes when counted 30 to 90 days after posting

  • +12%

    likes when counted 90 to 180 days after posting

  • +29%

    likes when counted 180+ days after posting

1,200+ character captions against the same account's 51 to 150 character captions, with format and hashtags held fixed.

Our best reading: long-caption posts keep picking up likes for months, possibly through search, long after a short post has gone quiet. A study that counts likes after a week or two will undersell them. We'll check this again with repeat readings of the same posts once we have a few months of them.

One length clearly loses, though: none.

Question 3: captions that “boost reach”

“Tonight, V stepped into the crowd” didn't boost anything.

The caption is Vogue's, from October 2025, about BTS's V. The theory on Reddit is that Instagram reads the famous name and shows the video to his fans as well, so the post reaches two audiences. Meme pages have done the same with “Japan is turning footsteps into electricity” and a Portuguese caption about Singapore turning rainwater into drinking water. Our data had 148 of these posts from 24 meme and repost accounts, so we compared each one with the same account's other posts that month.

−13%

likes vs the same account's other posts that month

48%

of the time the copy-paste post did better. A coin flip.

−32%

comments

That's a small sample (148 posts over 31 account-months), so the range is wide, from roughly half the usual likes to about 20% more. A small benefit isn't ruled out. A big one is: the “+30% reach” people report doesn't show up anywhere in it.

And if the trick does put the video in front of V's fans, it didn't turn into likes. Instagram's Mosseri has said ranking looks at likes and sends per reach, so an audience that watches and scrolls on is the last thing you want.

So which captions do get more likes?

The ones that ask people to do something. Unless it's buying.

Same method: each kind of caption against the same account's other posts from the same quarter.

The captionLikesComments
Asks people to tag a friend+43%3.6×
A giveaway or contest+42%5.4×
Asks people to save or share+11%+25%
Tags other accounts in the caption+7%+13%
First line starts with a number+7%+10%
Ends on a question+3%+4%
Asks for comments (“comment below”)+3%2.4×
First line starts with an emoji−6%−2%
“DM me”−13%−9%
Selling words (prices, sale, book now)−14%−4%
“Link in bio”−16%−16%
No caption at all−16%−20%

The pattern is simple. Captions that give people something to do (tag someone, enter, save it, answer a question) do better. Captions that send people somewhere else (your DMs, your bio link, your shop) do worse. “Comment below” more than doubles comments but leaves likes where they were, so it inflates one number without the post reaching more people.

For scale: none of these comes close to what the post itself does. A collab post, shared with another account, got 33% more likes than the same account's solo posts. A carousel with 7 or more slides got 29% more than one with 2 to 6. Spend the effort there first, then on the caption.

And hashtags?

Hashtags never helped much. Going over 5 now hurts.

Instagram announced a limit of 5 hashtags per post on 18 December 2025. Most accounts switched within weeks.

What people did after the cap

Share of posts with 6 or more hashtags, and with exactly 5, by month posted.

6 or more hashtagsExactly 5
Jan 2025
Apr 2025
Jul 2025
Oct 2025
Nov 2025
Dec 2025
Jan 2026
Feb 2026
Apr 2026
Jul 2026
Oct 2026

Instagram announced the 5-hashtag limit on 18 December 2025. The July 2025 dip comes from which accounts happened to post that month, not a trend.

Before the limit, an account's posts with 6 or more hashtags did as well as its posts with 1 to 5 (+2%, within noise). Since then, the posts that still go over 5 get about 9% fewer likes, and they're below the usual in every quarter of 2026. Of 585 accounts that used 6+ hashtags on most posts in late 2025, 414 never did again.

Within the limit, the count barely matters. Exactly 5 against 1 to 4 was +2.5%. No hashtags against 1 to 5 was no different. Use a few that say what the post is about, and stop there.

What to do with this

Three rules for your next caption.

1. Write it like you, not like a brochure.

If you use AI, use it for the first draft and then put back the details only you know: the client's name, what went wrong, what you'd tell a friend. Generic copy was the single biggest caption cost we measured.

2. Go long when you have something to say, and judge it later.

A long caption's lead grows for months. Don't drop the format because the first week looked flat.

3. Skip the tricks. Give people a reason to act on the post.

Pasted viral captions didn't help and 6+ hashtags now hurt. “Tag a friend who needs this” and a first line that starts with a number did more. Keep the sales pitch for posts that are meant to sell, and expect fewer likes on them.

Writing captions for a business?

See which posts and captions are beating their usual for the accounts you compete with.

These are averages over 14,780 accounts. Your niche will have its own version. Add an Instagram handle and SocialToolsAI finds the competing accounts, then shows the posts doing better than their usual, captions included, so you can see what's working in your space this week.

See your competitors' top posts

Free to start. No card, no Instagram login.

Methodology

How the numbers were computed, and where they're weak.

Sample

271,550 public posts from 14,780 accounts with 100+ followers, in 24,691 account-quarters with at least 4 posts each. Mostly posted in 2025 and 2026, in many languages (64% English). Left out: pinned posts, paid partnerships, hidden like counts, and posts under 7 days old when their likes were counted. The accounts are ones our product follows (businesses, their competitors, creators and meme pages), not a random slice of Instagram.

The comparison

For each post, the log of its likes minus the average for the same account in the same calendar quarter. Results are shown as a percentage against that usual. Ranges come from resampling whole accounts. Where we say “held fixed”, it's a regression on the same within-account numbers with format, caption length bands, hashtags, emoji, calls to action, post age and the other caption features together. The tables of single caption types compare posts with and without the feature inside the same account-quarter.

AI-sounding captions

A language model scored 5,014 captions (80+ characters, every such post in one random quarter per account, 394 accounts) on how much they read like generic, polished AI-assistant copy, and separately on selling, giveaways, specific details and personal stories. Asking it directly whether a caption was written by AI gave near coin-flip answers, so we didn't use that. This measures style, not authorship: some people write like that by hand.

Long captions over time

We count each post's likes once. Older posts were counted much later after posting than recent ones, and the long-caption lead grows with that delay. An earlier version of this analysis read that as long captions working less well in 2026. At matched delays there's no clear decline. Posts counted long after posting are partly a different set of accounts, so the slow-burn reading is our best explanation, not a measured curve.

Copy-paste captions

148 posts with the Vogue World caption about V, “Japan is turning footsteps into electricity” (English, Portuguese and Chinese versions) or the Singapore NEWater caption, posted November 2025 to October 2026. Each compared with the same account's other posts in the same month (24 accounts, 31 account-months). Small sample; read it as “no sign of a boost”, not as proof of harm.

Cause and effect

Comparing an account with itself removes a lot, but it isn't an experiment. An account may save its long captions, collabs or giveaways for its biggest moments. We also saw one result that is almost certainly backwards: posts with comments turned off got more likes, because people turn comments off on posts that blew up. We left it out of the advice.

Questions people ask

Frequently asked.

Do AI-generated captions get more engagement on Instagram?+
Not in our data. We had a language model read 5,014 captions from 394 accounts and say how much each one read like generic, polished AI-assistant copy. Compared with the same account's other posts, captions that clearly read like AI copy got about 25% fewer likes and 34% fewer comments, with selling and caption length held fixed. We can't prove who wrote a caption, so this is about how it reads: smooth, generic copy that could describe any business.
Does caption length matter on Instagram?+
A little, and it takes time to show. When the same account wrote a caption of 1,200+ characters, that post got about 9% more likes than its 51 to 150 character posts when counted a month after posting, and about 29% more when counted six months or more after posting. Comments were 20 to 25% higher from the start. Across accounts, long captions look much worse (a 0.87% like rate against 2.1% for short ones), but that's because of the kind of accounts that write them, not the captions.
Why does every reel have the caption “Tonight, V stepped into the crowd”?+
It's Vogue's October 2025 caption about BTS's V at Vogue World: Hollywood. Since mid-2026, meme and repost accounts have pasted it under unrelated reels, on the theory that a famous name in the caption pushes the video to his fans too. We found 148 posts with it or similar viral captions (“Japan is turning footsteps into electricity”, “Singapore NEWater”) from 24 accounts. Those posts got 13% fewer likes than the same accounts' other posts that month, did better only 48% of the time, and got about a third fewer comments.
Does copying a viral caption boost reach?+
We found no sign of it. Compared with the same account's other posts in the same month, posts with pasted viral captions got 0.87 times the likes. The range is wide (it's 148 posts) so a small benefit isn't ruled out, but the claims of +30% reach we've seen don't show up. If a caption does bring in a second audience, those people don't seem to like the post.
How many hashtags should I use on Instagram in 2026?+
Five at most. Instagram announced a 5-hashtag limit on 18 December 2025. Before that, posts with 6 or more hashtags did about as well as the same account's posts with 1 to 5. Since then, the ones that still go over 5 get about 9% fewer likes. Within the limit, the count doesn't matter much: 5 hashtags against 1 to 4 was +2.5%, and none against 1 to 5 was no different.
Is it bad to post on Instagram with no caption?+
It goes with fewer likes. When an account posted with no caption, those posts got about 16% fewer likes and 20% fewer comments than its captioned posts from the same quarter. The no-caption posts weren't a different format: about half were reels, the same as the rest.
Do em dashes make a caption look AI-written, and does it matter?+
Em dashes did spread at the same time as AI writing tools: they were in 1.5 to 4% of longer captions until 2024 and about 12% in 2025 and 2026, and captions with them read more like AI copy on average. But on their own they barely move likes (about 2 to 4% fewer, not clearly different from zero). It's the generic writing that costs you, not the punctuation.
Is this reach or likes?+
Likes. Instagram doesn't show reach or views for other people's posts, so likes are the closest public number we can compare across 14,780 accounts. A post shown to more people collects more likes, but a caption can also change how many viewers tap the heart, so read the results as engagement first and reach second.