How to Measure Instagram Keyword Trigger Performance

Manish 18 October 2026 8 min read

Which words to teach is keyword triggers that get commented. This page is performance: did LINK beat INFO on this shop this month, on comparable posts, with sends and tagged outcomes — not vibes.

A keyword can match a lot and still be a bad word. YES on a giveaway matches tourists. PRICE on a SKU Reel matches people who might buy. If you only rank by comment volume you will keep the giveaway and fire the SKU word. Measure matches, DMs actually sent, replies that need a human, and clicks you can tag. Then kill words that only attract spam.

A simple keyword scorecard

Comments that matched this keyword (honour exact versus contains — they are different rules).
DMs actually sent for that match, minus the ones still queued or errored.
Replies in the thread that were not just “thanks” if you care about qualification.
Tagged clicks or collected emails attributed to that keyword (utm_content=price vs utm_content=link).

Write one row per keyword per comparable post family. Do not average a launch Reel with a meme sound. Match types change the first column: keyword match types belong in the rule setup. This scorecard assumes the rule is already the one you meant.

CommentLink is not an A/B lab. A second Reel is not a control group; it is a different video. You can still learn that LINK outperforms INFO for your shop on lookbook edits. You cannot publish a universal ranking of keywords.

Do not compare across unlike posts

A giveaway YES is not a SKU PRICE. A “comment your city” sticker is not a “comment DATES” wedding Reel. Compare keywords on similar creative, similar offer, similar distribution. If one post went to non-followers and the other sat in follower-only reach, the Request-folder effect will masquerade as a keyword insight.

Purchase-intent words versus research words are a different split — purchase versus research intent. Performance measurement here is narrower: same intent class, two words, which one earned the next step you actually wanted.

Kill words that only attract spam

If a word matches bots, giveaway hunters, or people who type it because a duet told them to, the send count will look healthy and the tagged outcomes will not. Retire the word, tighten to exact match, or add excludes. Do not “fix” spam matches by making the DM longer.

FREE that delivers a real file is a product word. FREE that you taught because a growth thread said so is often a magnet for people who will never open checkout.
EMOJI-only catch-alls: you will spend the hourly cap on hearts. Measure that once, then stop.
Competitor brand names as keywords: you will match arguments, not buyers.

A shop example: LINK versus PRICE versus SIZE

On two similar garment Reels in the same week, teach LINK on one and PRICE on the other, with SIZE as a third rule scoped to a fit close-up. Scorecard after seven days: LINK matched 60, sent 55, 8 tagged PDP clicks, 1 size-related reply. PRICE matched 22, sent 22, 9 tagged PDP clicks, 4 “is this on sale” replies you answered by hand. SIZE matched 15, sent 15, 12 chart taps, 3 fit questions you took human.

If your goal was checkout sessions, PRICE won on this fabric despite fewer matches. If your goal was unloading the inbox, LINK “won” volume and still left fit questions on SIZE. That is performance. It is not a law that PRICE always wins. Next month’s drop might reverse it. Keep the sheet. Do not tattoo a ranking on the brand.

Spoken CTA has to match the rule

If the Reel says “drop a fire emoji” and the rule listens for LINK, your match column will look like a failed keyword. That is a teach problem, not a performance problem. Fix the caption and the on-screen word before you retire LINK.

When to split the word instead of retiring it

Sometimes LINK is fine on lookbooks and rotten on restocks because restock commenters wanted WAITLIST. Split the caption: teach WAITLIST on sold-out colourways, keep LINK on in-stock lookbooks. Performance then has two rows. Merging them is how you conclude “LINK is dead” when only one media family was dead.

If you use contains-match, a short word inside longer comments will inflate matches. Exact match will look “worse” on volume and cleaner on intent. Measure both only if you actually ran both. Do not compare this month’s exact PRICE to last month’s contains PRICE and call it a conversion story — you changed the rule.

A 14-day keyword bake-off you can defend

Pick two words in the same intent class on this shop. LINK versus PRICE on two lookbook Reels is a bake-off. LINK versus YES on a giveaway is not. Write the start date, the media permalinks, and the tagged next step you will honour — a PDP click with utm_content, a collected email, or a booking tag. Fourteen days or two comparable posts, whichever comes second. A quiet Tuesday is not a bake-off.

CommentLink is not an A/B lab and a second Reel is not a control group. You are answering: on this fabric, this month, which taught word earned the step we already named. If the spoken CTA said PRICE and you taught LINK, throw the week out. That is a teach miss, not a keyword insight.

Name two comparable Reels and one outcome column — not a blended “leads” cell.
Teach one word per Reel. Do not overlap contains-match so both rules fire on the same comment.
Export matches, sends, errors, and tagged outcomes. Delivery failures are not quality.
Kill or split the loser only after two similar posts, or when the word only attracts spam.

Tagged outcomes per word — not a conversion-rate essay

Comment-to-DM conversion rate is a stage definition owned elsewhere. This page is the keyword row. PRICE should produce PDP sessions or “is this on sale” threads you chose to count. LINK should produce lookbook taps. SIZE should produce chart taps and fit questions. If you dump all three into one cell, you will keep LINK because it is loud and fire PRICE because it looks small.

Jewellery Reel. Comment: “PRICE on the 18k hoop.” First DM: “Current price is on this PDP (utm_content=price-hoop-12). Custom diameter: reply CUSTOM — I will not invent a workshop quote.” After seven days: PRICE matched 22, sent 22, 9 tagged PDP clicks, 4 custom threads you took human. LINK on the same week’s lookbook: 55 sends, 8 lookbook taps, 0 hoop checkouts. PRICE won the hoop. LINK won volume. Write both sentences. Do not average them into a fake ranking.

Mistakes that invent a winning keyword

Comparing this month’s exact PRICE to last month’s contains PRICE. You changed the rule.
Ranking on matches while half the sends errored on a dead token. Fix delivery first.
Folding a 24-hour Story sticker GUIDE into a week-old Reel GUIDE. Different window, different row.
Keeping FREE because send count is high when the file only attracts people who never open checkout — unless the file is the product.
Teaching competitor brand names and calling argument-matches “engagement.”

CommentLink will not allocate traffic, freeze a control Reel, or draw Mixpanel funnels per keyword. Tags, UTMs, and a sheet you already trust are the join. Native CRM sync is not a keyword scorecard. No HubSpot deal stage named after LINK.

When MENU beats ORDER and you still keep both

Bakery. Spoken on Reel A: “Comment MENU for tonight’s PDF.” Reel B: “Comment ORDER for the 48-hour cake form.” MENU matched 80, sent 74, 40 PDF taps, 2 allergy threads you took human. ORDER matched 11, sent 11, 6 real cake holds. If you retire ORDER because it “lost,” you just fired the word that makes Saturday. Split the scorecard. Allergy, “the bill was wrong,” and named-person drama never get MENU or ORDER.

Hardware restock Reel. PART: “Comment PART plus the size stamped on the bit. This chat will not diagnose a gas leak — call the number we staff.” HOURS belongs on a different post. Measuring PART against HOURS is unlike posts again. Measure this month’s PART against last month’s PART on similar bins.

FAQ

How do I know which Instagram keywords convert?

Count matches, sends, and tagged outcomes per keyword on comparable posts. Kill words that only attract spam.

Should I rank keywords by comment volume?

No. Volume is the match column. Conversion lives in sends that become clicks, emails, or bookings you defined.

Can I A/B test two keywords on one post?

You can run two rules on one media if the words do not overlap. You still do not have a lab. Treat it as directional.

What if a keyword matches but DMs do not send?

That is delivery (cap, token, permissions), not keyword quality. Fix the send path before you retire the word.

Do Story reply keywords use the same scorecard?

Same columns, shorter window. When the Story expires, stop the clock. Do not compare a 24-hour sticker to a week-old Reel without saying so.

How long should I wait before killing a word?

Wait until you have comparable creative, not a single quiet Tuesday. If the word only brings bots after two similar posts, kill it.

Is keyword performance the same as comment-to-DM conversion rate?

No. Conversion rate is a stage definition. This page ranks words on comparable posts using matches, sends, and tagged outcomes.

Can I bake-off two words on one Reel?

Only if they do not overlap. Treat it as directional. A second Reel is not a control group.

What if LINK wins volume and PRICE wins checkout?

Keep both rows. Volume is not the kill criterion unless the volume is spam.

Send the link in a private DM

Connect Instagram with Meta login and send a private DM when someone comments a keyword.

See comment-to-DM