How to Collect Instagram Hashtags From Captions and First Comments

Collection Steps

1. Select one known post and keep its URL or short code. Copy the current caption from the public source into a separate caption field with observation time. The comments Actor does not supply captions. Do not use the old endpoint from past research.
2. Run a small default-order sample with `maxCommentsPerPost` set to `3` for the pilot. Set `postUrls` to your known public post or Reel URL, `collectAllAvailable` to `false`, and `commentSortMode` to `default`. Inspect target identity, text, and `OUTPUT`. `savedRows` and `cappedTargets` are run totals. Target entries hold per-target `resultState`, `sampleHealth`, and `stopReason`. The checked current three-row capped control verifies source identity, text, and dates only. It does not prove owner-comment recovery or the chronological first comment. Do not promise an owner comment from a higher cap. Increase the cap only for a justified known post. If the public sample hides the needed comment, mark the first-comment tag evidence as unknown and inspect the public post manually.
3. Identify the owner account from the source post. Match `ownerId` where present because a username alone can change. A comment that tags the creator is not written by the creator. Keep ambiguous author matches unresolved.
4. Keep each candidate owner-written comment's `commentId`, `text`, `createdAt`, `ownerId`, `ownerUsername`, `ownerVerified`, `likesCount`, `repliesCount`, `collectedAt`, and source URL. Note that `repliesCount` is a displayed count, not a reply export. Manually check if it is the intended initial hashtag comment. Oldest-first only sorts the collected public sample. It cannot establish the true first comment. Do not label an uncertain candidate first.
5. Extract visible hashtag tokens, keep raw spelling, and make a separate casefolded comparison key. Use a clear token rule. For English, accept the hash symbol followed by letters, digits, or an underscore. Punctuation ends the token. Other scripts need a matching rule and manual checks. Exclude user mentions, email addresses, and lone hash symbols.
6. Build a provenance table. Include the post key, comment key or caption source, author verification, raw tag, normalized key, `collectedAt`, and status. Do not mix another user's tag with the owner's tags.
7. Compare caption tag keys with verified owner-comment tag keys. Report caption-only, comment-only, and shared sets. Keep unavailable owner-comment evidence outside the empty-set claim. This shows observed tag placement, not proof of reach.

Fictional Worked Example

Here is an original fictional example to show how the table works. Fictional post P1 has the caption "Garden notes #LeafCare #HomeGarden." A verified owner comment C1 says "More tags: #HomeGarden #SoilMix #LeafCare #SoilMix." A different account C2 says "Nice #PlantSale."

Owner identity matches C1, and its initial placement is checked on the public post. The caption set has leafcare and homegarden. The owner-comment set has homegarden, soilmix, and leafcare. Shared tags are leafcare and homegarden. Comment-only is soilmix. The caption-only set is empty.

Six hashtag occurrences across the caption and comment become three distinct owner-used keys. The duplicate soilmix occurrence stays in the provenance data. The plant sale tag from C2 is excluded from the owner-used set but stays in the source data. If C1 cannot be seen on another run, say owner-comment tags are unavailable instead of saying the owner used no comment hashtags.

In the table below, `Manual post-owner match` is a reader check that comment `ownerId` matches the account owning the post. Native `ownerVerified` is the public badge field, not proof of post ownership. Badge values were not supplied in fictional data, so they are kept as unknown. Table `Active` and `Excluded` values represent reader review decisions, not native dataset status, because the Actor has no dataset status field. Native `postUrl`, `postId`, and `postShortCode` are retained with every actual provenance row; reader post key P1 is used for fiction, and unsupplied real URLs and identifiers are marked as unknown rather than fabricated.