Key takeaways
- Beginners should run one comment keyword per post, a short real word of four to eight letters that names the offer, with matching set to “contains”.
- Short keywords can hide inside other words under “contains” matching, so ART fires on “started” and “smart”, and two-letter words like ME match almost any sentence.
- Keyword matching ignores capitalisation in every serious tool, but phone autocorrect can rewrite words one letter away from commoner ones, so test the word in your own comment box first.
- STOP and START are reserved opt-out and opt-in words, so never use them or any word containing them, such as RESTART or STOPWATCH, as a comment keyword.
In this article
The comment keyword is the only part of your funnel that your audience has to type from memory, on a phone, one-handed, while half-watching your video. Everything else you control. That word you do not: autocorrect gets a vote, and so do thumbs, and so does whether the person was really listening when you said it. Most automations that "do not fire" were never broken in the first place. They were listening for a word nobody typed correctly. The fix is boring and takes about thirty seconds at build time: pick a short, real, unambiguous word, say it out loud in the video, put it on screen, and set the match mode so near-misses still count. Below are six tests to run a candidate word through, a table of words that work against words that quietly fail, and a straight answer on how many to run at once.
Short answer: Use one keyword per post for your first automations. Pick a short, real, easy-to-spell word that describes the offer, four to eight letters, no numbers or punctuation, and set matching to "contains" so "GUIDE please" still counts. Add a second word only when real comments show you need it.
How many keywords should a beginner use?
One per post, and one automation per offer. The reason is diagnostic as much as strategic: with one word on one post, a failure has exactly one possible cause, and you learn something from every test instead of guessing which of five triggers misbehaved.
Most builders let you put up to five comma-separated keywords on a single trigger, which sounds like insurance and usually is not. Every extra word widens the net, and a wide net catches comments you did not want: people quoting the word back at each other, people typing it in a sentence that means the opposite, people replying to someone else's comment. Start narrow. After a week of real comments you will have evidence rather than guesses, and the second keyword you add will be a misspelling you actually watched happen.
To run a word through these tests in one go, use the free comment keyword generator: it checks length, accents, reserved words and the everyday words a keyword hides inside, and tests it against sample comments.
Six tests for a comment keyword that people actually type
Run every candidate word through these in order. A word that fails test two or three is not worth defending, however clever it is. To watch which words are pulling and which are dead weight before you commit to a plan, HyperDM's keyword triggers run on the free tier, so you can prove a word on one post first.
Step 1: Start from the offer, not from your vocabulary
The word should name the thing you are sending. If the DM carries a pricing sheet, the word is PRICE. If it carries a free checklist, the word is CHECKLIST. People do not remember clever words, they remember what they wanted. A branded word like SUNDAYCLUB asks a stranger to learn your vocabulary before they can have the thing, and the drop-off happens in that gap. Save the branding for the DM, where you have their attention and they have nothing to type.
Step 2: Keep it to one real word, four to eight letters
One word, not a phrase. Phrases invite word order mistakes, extra spaces and autocorrect capitalisation. Four to eight letters is the sweet spot: long enough to be deliberate, short enough to type without a typo. Real dictionary words beat invented ones because phone keyboards actively help with them, suggesting the right spelling instead of fighting it. Avoid numbers, hyphens and apostrophes entirely, since each one is a place where a thumb can miss and a mobile keyboard can be unhelpful.
Step 3: Survive autocorrect and the mobile keyboard
Type the candidate word into your own phone's comment box before you commit. Words that sit one letter away from a commoner word get rewritten: FLOW becomes FLOG or FLOOR, PLAN becomes PLANE, LINK is safe but LNK is not. Invented spellings are the worst offenders, because the keyboard treats them as errors to correct. Matching ignores capitalisation in every serious tool, so GUIDE and guide trigger the same automation and you do not need to tell people to shout.
Step 4: Check the word does not hide inside other words
This is the test nobody runs, and it produces the strangest failures. With "contains" matching, a keyword fires whenever the letters appear anywhere in the comment, including inside another word. ART matches "started", "smart" and "party". LIST matches "listen" and "realistic". Two-letter words like IN and ME match almost any sentence a person could write. Before you set a short word live, type it into the search box of your own comments and see what it already hits. If the word turns out to be a common substring, either lengthen it or switch the match mode in Step 5.
Step 5: Choose "contains" or "exactly", deliberately
Two match modes, two different bets. "Contains the keyword" fires if the word appears anywhere in the comment, so "GUIDE please, thanks!" counts. "Is exactly the keyword" requires the comment to be nothing but the word, which stops substring accidents dead but also throws away every comment with a word of politeness attached. For most creators, "contains" plus a word that is not a substring of anything is the right combination. Choose "exactly" only when your word is short and common and you cannot find a better one.
Step 6: Keep clear of the reserved words
Opt-out and opt-in words sit above everything you build and cannot be outranked: a message containing STOP revokes consent for automated replies, and START restores it. Never choose those as a comment keyword, and never choose a word that contains them as a substring, which rules out START, RESTART and STOPWATCH as triggers. Check your tool's channel settings for any custom opt-out words your workspace has added before you go live.
Good and bad comment keywords, side by side
| Keyword | Verdict | Why |
|---|---|---|
| GUIDE | Good | Real word, names the offer, no common substring |
| PRICE | Good | Exactly what the person wants, impossible to mistype |
| LINK | Good | Short, safe spelling, universally understood |
| RECIPE | Good | Names the offer, no substring risk |
| ART | Bad | Hides inside "started", "smart", "party" |
| ME | Bad | Appears in "me too", "some", "comment" |
| FLOW | Risky | Autocorrects to FLOG or FLOOR on some keyboards |
| SUNDAY CLUB | Bad | Two words, invites word order and spacing errors |
| GET-IT-NOW | Bad | Punctuation, three words, hard on a phone |
| STOP | Forbidden | Reserved opt-out word, always wins |
Caps, emoji, typos and what people really send
Treat the comment you asked for as the best case and plan for everything around it.
Capitalisation is a non-issue: matching ignores case, so put the word in capitals in your caption purely because it stands out visually. Emoji attached to the word are harmless under "contains" matching and fatal under "exactly", which is another reason "contains" is the safer default for a comment word to get link offers. Misspellings are the real cost, and the honest fix is measurement rather than prediction. After your first few hundred comments, read the ones that did not fire and add the top one or two misspellings as extra keywords. That is the point at which an Instagram keyword automation starts to feel tuned instead of guessed. Our comment-to-DM setup guide covers where those unmatched comments show up, and DM templates that convert covers what to do with the ones that do fire.
Worth knowing: Meta Business Suite has its own built-in keyword automations, including a comment to message rule, and the same word-choice rules apply there. The underlying mechanism is identical in every tool, because the mechanism is Meta's: one private reply per comment, within 7 days of the comment being created.
Picking a comment keyword you will not have to fix later
A good comment keyword is unglamorous: a short real word that names the offer, that a phone keyboard will help with rather than correct, and that does not live inside a commoner word. Say it aloud in the video, write it in the caption, set matching to "contains", and keep one word per post until you have real data to argue with. The payoff for thirty seconds of care is that every failed test afterwards means something, because the word has stopped being a variable. When you are ready to run several offers at once and want to see which words are actually pulling, HyperDM's plans start free with 50 conversations a month.
Frequently asked questions
HyperDMRun the automations. Add AI when you’re ready.
Comment-to-DM, keyword replies and follow gates — set them up, test them, then add optional AI answers from your real catalog. Free on one DM channel, 50 conversations a month, no card.



