A/B testing for TikTok DMs — know which message closes

How does A/B testing work for TikTok DMs?

Every step of a Hooka campaign accepts up to 20 message variants. Each time a DM goes out, one variant is picked at random — a uniform split, so every variant gets a fair share of sends — and the variant index is recorded on that exact DM.

That per-DM record is what makes the analytics trustworthy. Reply rates aren't estimated from campaign totals after the fact; they're computed from which variant each recipient actually received.

One thing Hooka deliberately does not do: auto-optimize. There's no bandit algorithm quietly shifting traffic toward an early "winner" mid-flight. The split stays random and uniform from the first send to the last, your sample stays clean, and you make the call. Variants support personalization too — handle, username and custom per-recipient fields like first name — so you're testing the message, not the merge tags.

How is cold DM reply rate measured?

A variant's reply rate is the share of unique recipients who sent an inbound reply in that conversation after the variant went out. That's the whole definition — no modeled attribution, no fuzzy windows.

Unique matters: one lead replying six times counts once. "After" matters too: a conversation where someone messaged you before your DM doesn't inflate the number. Replies are matched against real inbox messages, checked every 60 seconds — the same signal that automatically stops a follow-up sequence the moment a lead answers.

What about open rates? TikTok DMs have no open tracking — not in Hooka, not anywhere. Reply rate is the honest metric, and it's the one that correlates with revenue anyway. Every reply lands in the unified inbox, where you can mark leads interested or not interested.

Once a variant reaches at least 5 sends, the best performer in its step gets a star. Treat 5 as a floor, not a verdict — let real volume accumulate before you kill a variant.

Which sender account converts better?

Hooka also breaks reply rate down per sender account, because two accounts sending the identical message rarely convert identically.

Profile photo, bio, follower count and posting history all shape whether a stranger accepts a message request from you. Per-account reply rates surface that: if one account pulls 11% and another pulls 4% on the same variant, the message isn't the problem — the messenger is. Rework the weak profile, or shift volume to the accounts that convert.

A/B analytics are included on every plan. Pricing is per connected TikTok account — $89/account/month with volume discounts down to $79, and no monthly DM cap (as of June 2026).

Can you A/B test follow-ups too?

Yes. A Hooka sequence runs up to 10 follow-up steps, each with its own delay (1–60 days) and its own set of up to 20 variants — and the analytics are per step, per variant.

Follow-ups only go to leads who haven't replied; a reply always stops the sequence automatically. That keeps step-level data meaningful: step 2's reply rate measures how well step 2 revives the people step 1 didn't convert — not a blend of everything.

In practice, that's where the wins hide. Openers get tested to death; almost nobody tests the bump message. Our cold DM templates guide has follow-up examples worth putting head-to-head.

How do you run a clean A/B test?

1. Change one thingWrite 2–3 variants that differ in a single element — the hook, the offer, or the CTA. If everything changes between variants, the data tells you nothing.
2. Let the random split runWait until every variant has at least 5 sends — ideally far more — before reading the numbers. Early leads flip constantly.
3. Keep the starred winner, cut the restCheck per-account reply rate too; sometimes the "losing" message is really a losing profile.
4. Duplicate and go againCampaign duplicate clones the full config — variants, follow-ups, settings — without the recipients. Load a fresh list, keep the winner as control, test a new challenger.

That duplicate step is the quiet workhorse: because recipients never carry over, each re-run starts from a clean sample instead of a half-contacted list. Cross-campaign dedup is on by default, so a lead you already messaged elsewhere gets skipped, not spammed.

What does the dashboard show?

The dashboard gives you the rolled-up view: DMs sent this month, a 30-day conversation-based reply rate, and a 30-day activity chart.

Interested-lead movement comes from a real audit log — every time a conversation is marked interested in the inbox, that change is recorded, and the dashboard shows the delta over time. No vanity counters reconstructed after the fact.

Frequently asked questions

Keep reading