Do likes and comments pick the same top Instagram post?

“Send us your top post” sounds like a straightforward request from a business to a creator. But a post with the most likes can differ from the post with the most comments. Before choosing the screenshot for the report, both sides need to agree on what top means.

Picture a fictional creator reviewing two campaign posts. Post A has 12 likes and one comment. Post B has seven likes and four comments. Both comparisons have an unambiguous leader, but they point to different posts. Calling either one the best-performing post would hide which measure made the choice.

Our question was narrower than which post was best: within an account’s included posts, did likes and comments each identify one leader, and was it the same post? That distinction leaves room for two other answers. Several posts can tie, and a measure can be recorded as zero for every included post.

These are different reporting situations. A tie is not permission to choose whichever thumbnail appears first. A column of zeros cannot distinguish the posts by that measure. Keeping those outcomes visible makes the report easier to interpret.

We examined the two files in Tyler Horan’s public Instagram dataset, collected between April 15 and May 15, 2020. One came from the #sponsored hashtag. The other used related hashtags found in those posts. These selected lists are historical source material, not a survey of today’s Instagram accounts. ^[1][2]

We included accounts with at least two distinct posts in the same file. This gave us 3,966 eligible accounts in the #sponsored-derived archive and 2,167 in the comparison archive. Each account counted once within its file, however many included posts it had. Accounts represented by only one post were outside the comparison because they offered no choice between posts.

A tie above zero means that both measures had at least one recorded response, but likes, comments or both had several posts sharing their highest count. We kept that state separate from an all-zero measure. These rules let every eligible account fit one of the four outcomes shown below.

In the #sponsored-derived archive, likes and comments selected the same unique leader for 1,903 accounts, or 48.0%. They selected different unique leaders for 1,200 accounts, or 30.3%. Another 408 accounts had a tie above zero, and 455 had at least one measure recorded as zero on every included post. ^[4]

In the comparison archive, the same unique leader appeared for 526 accounts, or 24.3%. Different unique leaders appeared for 306 accounts, or 14.1%. Ties above zero accounted for 236 accounts; a measure recorded as zero throughout accounted for 1,099. All four outcomes remain in the denominator.

The two tables describe the files separately. They do not show that sponsorship causes better agreement, more discussion or stronger performance. The files were collected differently, and the available fields do not tell us each post’s age or how long people had to respond.

The reporting lesson comes from seeing the separate outcomes. “The measures did not pick the same unique post” can mean a different leader, a tie or an all-zero measure. Those answers should not be folded into one claim that the metrics disagreed.

A tie can be easy to miss in a sorted table. The first row looks like a winner, even when the next row has exactly the same count. Our calculation kept every post at the maximum rather than using row order or a post identifier to select one.

Consider another fictional pair. Post A has 12 likes and four comments. Post B has 12 likes and one comment. Post A is the unique leader for comments, but likes leave A and B tied. There is a post that belongs to both sets of leaders, yet neither likes alone nor the archive supplies a reason to break that tie.

This situation occurred within the tie-above-zero category. The sets of leaders overlapped for 320 of those 408 accounts in the #sponsored-derived archive, or 78.4%. They overlapped for 192 of 236 in the comparison archive, or 81.4%. These percentages use only accounts with a tie above zero as their bases. They are not extra agreement rates for all eligible accounts. ^[4]

If a report needs a single example, the creator and business can choose one for an explicit editorial reason, such as the product detail it illustrates. That is a separate choice. Write “selected example” and explain it, rather than presenting the chosen post as an undisputed numerical winner.

For a recurring report, keep the tie rule stable. Listing all tied leaders is straightforward when the list is short. If a team uses another display rule, preserve the fact that a tie exists so the layout does not silently turn it into a unique result.

When every included post has zero comments, the comment measure cannot identify one unique leader. That remains true if the posts have different like counts. In a fictional pair with 12 and seven likes but zero comments on both, likes choose the first post; comments do not choose between them.

The archive contains this state, not just a hypothetical example. In the #sponsored-derived file, 443 eligible accounts had some recorded likes but zero comments on every included post. In the comparison file, 1,015 accounts had that combination. Accounts with both measures at zero, or likes alone at zero, are separately retained in the downloadable results. ^[4]

Write what the field shows: “zero comments recorded across these included posts.” Do not translate it into “the audience was not interested.” The archive does not establish why a count is zero, how many people saw the post or whether a person responded somewhere else.

In a team’s own report, keep unavailable data separate from a measured zero. A missing value cannot be used as evidence that nothing happened. This is a reporting recommendation, not a correction we made to the source. The released count fields passed our specified checks, and their zeros remained zeros.

Return to the fictional creator with two campaign posts: A has 12 likes and one comment; B has seven likes and four comments. Suppose the agreed task is to choose examples for a discussion about the next campaign. The report can present both leaders without pretending that either count answers every question.

A useful report sentence is: “At our stated review cutoff, Post A led these two posts in likes and Post B led them in comment count. We will review both before choosing what to reuse.” This keeps the observed result separate from the next creative decision.

The creator can bring Post A to discuss its presentation and Post B to read the comments in context. A comment count alone cannot tell the business whether the responses were favorable, detailed questions, repeated messages or something else. The study does not measure sentiment, unique commenters or the value of that discussion.

If the partner’s real question is about sales, neither leader establishes the answer. The team needs evidence that addresses the agreed sales question, with the relevant measurement and attribution limits. Do not rename a like count “purchase intent” or treat a comment as a confirmed customer.

The comparison also needs a stated checkpoint. Counts collected at different post ages can reflect different opportunities to accumulate responses. Our archive does not let us standardize that opportunity. In a new report, record when the observations were taken and decide which comparison window fits the question.

This approach does not require a single combined engagement score. If a team uses one, name the formula and explain the decision it serves. Adding likes and comments answers a different selection question from the two separate rankings examined here.

Use these prompts in a creator handoff or campaign report. The template is a suggested reporting aid, not a form tested in the historical study. It works whether the answer is one leader, different leaders, a tie or no unique choice on a recorded-zero measure.

Posts included: [account and post list]. Explain why these posts belong together. Name any unavailable posts or fields, and keep the comparison within the scope the report actually covers.

Review cutoff: [date, time and time zone]. Record when each count was observed. If posts had different observation periods, describe that difference before interpreting their counts.

Likes: [leading post or tied posts, recorded like count]. Comments: [leading post or tied posts, recorded comment count]. Keep the entries separate. Write “tied” where needed and keep unavailable values distinct from zero.

Selection status: [same unique leader, different unique leaders, tie above zero, or a measure recorded as zero throughout]. If a measure is zero throughout, say which one. This entry explains why the report does or does not name one common leader.

Interpretation: [the narrow statement supported by the result]. For example, “Post A had the most likes among these two posts at this checkpoint.” Avoid turning a count into a claim about content quality, favorable sentiment or sales.

Next review: [what needs inspecting, who will do it and when]. Link the actual posts or evidence within the team’s own reporting workflow. For the fictional example, the action is to review both leaders before deciding what to reuse.

Keep the report concise enough that the business and creator can read it together. The point is to make the selection understandable, not to surround an ordinary observation with technical language. A reader should be able to tell which measure picked the post and what decision is still open.

The source release contains 29,425 posts in the #sponsored-derived file and 28,779 in the other-hashtag file. We used the released account identifiers, post identifiers, likes and comments. The follower field was not used to choose accounts, weight them or normalize results. This is a reanalysis of the released files, not a reproduction of the producer paper’s combined-engagement analysis. ^[1][2]

After validating the source structure and identities, we retained every account with at least two distinct posts within a file. The eligible accounts contributed 16,765 and 7,373 posts respectively. We excluded 12,660 and 21,406 accounts represented by a single post. These were not treated as failed agreement cases.

The included post sets were usually small. The median was two posts per eligible account in each file. The lower quartile was two in both; the upper quartile was four in the #sponsored-derived file and three in the comparison file. These are the posts available in each archive list, not complete account histories. The full files shared 29 account identifiers, so their account counts should not be added as distinct people.

For each eligible account, we found all posts with the highest recorded like count and all with the highest recorded comment count. An all-zero measure was classified first. Among the remaining accounts, we distinguished the same unique leader, different unique leaders and at least one tied leader. Tied cases were then separated by whether their sets of leaders overlapped.

Every primary percentage divides its category count by all eligible accounts in that file, including zero and tied cases. The tie-detail percentages explicitly use the group with a tie above zero instead. We used exact counts and fractions, then rounded percentages to one decimal place. Rounded category percentages need not sum to exactly 100.0%.

The eligibility rule, categories and outputs were fixed before computing these results. Two independently written calculations agreed on all reported aggregates. The analysis describes two finite selected archives, so it supplies no population confidence interval or causal effect. Account identifiers do not establish unique people, professional creators or paid sponsorship contracts. ^[4]

The dataset is released under CC BY 4.0. The downloads provide aggregate results and the method needed to reproduce them from the original source. They include no account or post identifiers. The source files remain available from Zenodo, and attribution does not imply that the producer endorses this interpretation. ^[1][3]