What the extraction actually does
Four steps, all of them mechanical. The text is split on anything that is not a letter, digit or apostrophe. A possessive "'s" is trimmed. Words on a list of English function words — articles, prepositions, pronouns, auxiliaries, the ordinary connective tissue of a sentence — are discarded, along with anything shorter than the minimum length and anything purely numeric. What survives is ranked by how many times it appeared and the top of that list becomes the tags.
There is no stemming, so "workbench" and "workbenches" are separate words and only one may make the cut. There is no phrase detection, so a two-word concept becomes two single-word tags or neither. There is no knowledge of your subject. If your post uses an unusual word four times, that word will be near the top of the list whether or not it is a term anybody else would use.
Why no tool can suggest tags from your text alone
A useful hashtag suggestion needs information that is not present in your post: how many people follow the tag, how much is posted under it, whether it is dominated by bots, whether it means something else in another community, and whether your audience uses that word at all. None of that can be inferred from a caption. Tools that present extracted words as "recommended" or "trending" hashtags are relabelling word frequency as market data.
The only reliable check is looking. Type the tag into the platform's own search and see what comes up, how much of it resembles your post, and how recent the top results are. That takes a minute per tag and it is the step no tool substitutes for.
Formatting rules that trip people up
A hashtag runs until the first character that is not a letter, digit or underscore. So "#3d printing" tags "#3d" and leaves "printing" as plain text, and "#half-lap" tags only "#half". Spaces have to be closed up or the phrase abandoned. Accented and non-Latin letters work on all the major platforms. Purely numeric tags behave inconsistently and often do not link at all.
Casing does not affect matching — #ShopBuild and #shopbuild reach the same place — but it does affect whether a long tag is readable, and it matters for screen readers, which handle capitalised word boundaries far better than a run of lowercase. If a tag is more than two words long, capitalise each one.
Placing the tags
The dotted-line format exists to push a block of tags below the caption fold on Instagram, so the caption reads cleanly and the tags sit out of the way. It works, and it is purely cosmetic. Instagram has said that tags in the caption and tags in the first comment function identically for reach, so choose on the basis of how the post reads, not on a theory about the algorithm. On X, a block of tags is actively counterproductive and one or two placed inside a sentence is the norm.
A word on how many
The thirty-tag cap on Instagram is a documented limit, not a recommendation, and treating it as a target is a habit left over from an era when it worked. Instagram's own guidance has pointed toward a handful of closely relevant tags. Relevance is the part that is hard and the part no generator can do for you: a tag that describes your post accurately is worth more than ten that describe a bigger audience you are not in.
Questions people ask
Why are some of the extracted tags nonsense?
Because the selection is frequency, filtered by a stopword list, and nothing more. A word can be frequent in your post and useless as a tag — a proper noun, a verb that happened to repeat, a piece of jargon nobody searches. The table showing the count for each extracted word is there so you can see the basis and delete the ones that do not earn their place. The tool is a starting list, not a decision.
Does it handle plurals and different tenses?
No. Matching is on the exact word as written, so "sled" and "sleds" are counted separately, as are "build" and "building". This can push a concept out of the top of the list because its mentions were split across forms. If you notice that happening, add the form you want to the manual field and it will be placed first.
Are these tags going to get my post more reach?
Unknowable from here, and be suspicious of anything that says otherwise. Reach depends on the post, the account, the audience and the platform's ranking at that moment, none of which this page can see. What tags reliably do is make a post findable by someone browsing that tag, which is a smaller and more honest claim than the one usually made for them.
Can I use it on text that is not English?
The formatting and deduplication work on any script, and accented and non-Latin characters survive intact. The extraction will be poorer, because the stopword list is English only, so the most frequent words in another language will mostly be that language's function words. Raise the minimum length, or paste your own tags into the manual field and set the extraction count to zero.