Instagram Auto DM keywords: choose, match and test them

A keyword rule matches text, not what somebody meant

vkit's standard comment matcher looks for the configured keyword within the comment, ignoring letter case. For GUIDE, ‘Guide please’ qualifies. So does ‘guidebook,’ because it contains the same sequence. The table describes the text check; a matching comment still needs the correct active campaign and eligible delivery conditions.

This is not a promise that AI interprets every comment. ‘I need the resource’ does not contain GUIDE. ‘gudie’ has the letters in a different order. If you want an alternative spelling or phrase to count, configure and test it explicitly instead of assuming the system repairs it.

Short words create the most surprising accidental matches. YES appears within ‘yesterday.’ FIT appears in ‘outfit.’ Those may be perfectly ordinary comments rather than requests. Read your proposed keyword inside longer words before you put it in a caption.

The objective is not to trigger the greatest number of replies. It is to deliver the requested resource to people who understand the invitation. For the complete follower journey, start with how Instagram Auto DM works.

Choose a word that belongs to the resource

Write the resource first: a packing checklist, a lighting setup or a weekly planning sheet. Then choose a recognizable word that points to it. PACKLIST, LIGHTING and WEEKPLAN are more specific than YES or WANT. Test them against your audience's normal comments rather than treating any example as universally safe.

Use one visible instruction: ‘Comment PACKLIST and I'll DM the carry-on checklist from this Reel.’ Keep the keyword identical in the artwork, spoken instruction and caption. If the graphic says PACK and the caption says LIST, people can follow your directions correctly and still miss the trigger you configured.

Do not make people solve a code. A complicated string may avoid accidental matches while becoming impossible to remember. Choose the shortest word that remains distinctive in context, then read the caption aloud. Somebody watching on a phone should understand the request without pausing to study it.

• Connect the word to the promise

  LIGHTING makes sense for a desk-lighting diagram. A random word can work technically but gives the follower less help remembering what they requested.

• Read it inside longer comments

  Try ordinary praise, questions and longer words. A substring match can be correct in software and still feel unexpected to the person receiving it.

• Keep the campaign consistent

  Check the post's text, caption and configured keyword together. Update all three if the resource or instruction changes before publication.

Use exclusions carefully instead of writing a fragile rule

vkit checks excluded terms before accepting a comment rule. If an excluded term is present, that rule is skipped even when its keyword also appears. Exclusions use containment too, so a broad excluded fragment can block more comments than you intend.

For example, with GUIDE as the keyword and NO as an exclusion, ‘guide please, I know this will help’ contains NO inside ‘know.’ That is probably not the rejection you meant to filter. A word can look harmless in a settings field and behave very differently inside a real sentence.

Begin with a clear invitation and a small, understandable rule. If you add an exclusion, write one intended blocked example and several legitimate requests that should still pass. Prefer revising an ambiguous campaign instruction over building an ever-growing list of exceptions.

Treat an explicit request to stop as something the team must handle, not a language puzzle a keyword can completely solve. Text matching cannot reliably infer sarcasm, every language variation or the full history of a conversation. Read the actual replies as part of normal campaign maintenance.

Make a test sheet with successes and non-matches

Before launch, record the post, keyword, exclusions, reply and destination in one row of your campaign notes. Under it, list the exact comments you want to test. Include the plain keyword, lowercase, the keyword inside a sentence, a nearby emoji, a typo and an unrelated compliment.

Add at least two negative cases: a comment that contains a misleading longer word and a comment that includes an exclusion. Write the expected result before running the test. Otherwise it is too easy to call every observed behavior ‘working’ because a message arrived somewhere.

Ask a consenting tester to verify the full journey on the intended account and post. Text matching is only one stage. Check the actual sending handle, the reply content and the destination on mobile. Use appropriate fresh test conditions when repeat-comment behavior would make the result unclear.

If the expected and observed outcomes differ, change one setting at a time. Save the old and new values in your notes. The Auto DM troubleshooting checklist covers other reasons a correctly matched request may not produce the delivery you expected.

Add campaigns without making the keyword map a mess

Keep a small register: resource name, post URL, keyword, destination, owner and active status. Before adding another rule, search that register for similar words and overlapping post scope. A rule attached to the wrong post can send a perfectly functioning reply with the wrong resource.

Review campaigns when an offer expires or a destination moves. A keyword is easy to remember, but the link behind it can become stale. When followers ask for an old guide, decide whether to keep that resource available, update the promise or pause the campaign.

vkit is a Meta Tech Provider. Its Instagram Auto DM gives this content-to-reply workflow a place alongside creation and publishing. Pro includes unlimited contacts without per-contact fees, so your keyword decisions can focus on relevance rather than trimming the contact list to stay below a billing tier.

Choose one resource and follow the setup guide. A clear keyword plus a tested destination is a stronger first campaign than ten clever triggers nobody has checked. Improve the copy after reading what real people ask next.

Frequently asked questions