A/B test your sequence copy

A campaign can hold two copy variants at once, and each lead that enrolls is assigned to one of them. Everything else stays shared — the same leads, sender accounts, schedule, and settings — so the only thing that differs between the two sequences is the writing. A/B testing lives in the campaign's Sequence tab, alongside the rest of the editor.

For writing and personalizing the copy itself, see Build your sequence.

Creating a variant

1. Open the campaign's Sequence tab and click A/B test. Clay adds a second sequence called Variant B, splits enrollment 50/50 between the two, and opens the new sequence's name for editing with the generated name selected — so you can call it something descriptive straight away, or leave it as Variant B.
2. The new sequence starts as an exact copy of the first, so nothing is under test yet. On the message you want to vary, click Test subject or Test body.
3. Write the alternative copy in the field that unlocks. It's seeded with the original wording, so you're editing rather than starting from a blank message.
4. Repeat for any other subject or body you want to vary. A test can cover one field of one message or every field of every message, with one exception: a message that replies on an existing thread inherits that thread's subject, so only its body can be tested.

Once a second sequence exists, the same choice also lives in each message's 3-dot menu, as A/B test subject and A/B test body checkboxes. They're the same toggles, and they show you which of that message's fields are currently under test. Unchecking one hands the field back to the shared copy, and whatever you typed into it while it was under test is dropped.

Where the window is wide enough, both sequences sit side by side. Otherwise the editor shows one at a time — use the sequence name above it to switch between them, or to Rename a sequence to something more descriptive than Variant A and Variant B.

The shared sequence shape

Both sequences always have the same number of messages, the same delay before each one, and the same reply-or-new-thread setting per message. Only subject and body text can differ, and Clay keeps that true for you rather than leaving it to be remembered: the structural controls in the editor write to both sequences at once, so adding a message adds it to both.

That constraint is also what keeps the comparison meaningful. If one sequence had an extra message or a longer gap before it, a difference in replies would be telling you about timing as much as about copy.

‍

‍

Reading the results

Open the Analytics tab. While a test is running, two filters scope what you're looking at:

• Sequence scopes the whole view to the leads one sequence enrolled.
• Compare puts two sequences in columns side by side over the same date range.

Each sequence reports the same figures a campaign does, counted over its own leads.

‍

‍

Underneath those sits the number a test is usually decided on — reply rate, with the counts behind it: how many of the delivered emails drew a reply, and how many of those replies were positive. It's the share of delivered emails rather than of emails sent, so it reads a little higher here than the campaign-wide reply rate does on the same campaign. For that view and for reply categories, see Campaign analytics and replies.

Give the data time to mature

Replies lag sends by days, so a sequence that only started enrolling recently reads low for reasons that have nothing to do with its copy. Clay shows what share of a sequence's sends are at least three days old, along with the reply rate among just those sends.

A difference is worth acting on once most of the sends in both columns are mature and the counts are big enough to mean something. Each column also names the window its sequence was running in — Live since a date, or Ran between two — because two sequences that never enrolled leads at the same time differ by when they ran as much as by how they were written.

Converting the winner

When one sequence is clearly ahead, end the test and promote it.

1. Save any outstanding copy edits. Until you do, End test is disabled and hovering it reads Save copy changes before choosing a winning sequence., so the sequences you decide between are the ones on file.
2. Click End test.
3. In the Complete A/B test dialog, choose the Winning sequence. It opens with nothing selected, and confirming without a choice asks you to Pick a sequence to continue with.
4. Confirm with Select winning sequence.

Keep testing closes the dialog and leaves the split as it was.

Promoting a winner is an allocation change rather than a copy edit, and it applies to future enrollments only:

• Every lead that enrolls from then on gets the winner's copy, and the other sequence leaves the editor.
• Leads already enrolled keep the copy they were sent, and their sequence plays out as written.
• The sequence that lost keeps its results too: it stays available in the Sequence and Compare filters, marked Ended.

There's nothing to configure about statistical significance, and no threshold Clay is waiting to cross. You have the rates, the counts behind them, and how mature the data is — the call is yours to make.

Cancelling on a draft campaign

While the campaign is still Draft, the same button reads Cancel test and opens Revert to single sequence instead. Pick the Sequence to keep, and the other is dropped when you save. Nothing has enrolled, so there are no results to preserve.

If neither sequence has any copy written yet, there is nothing to lose whichever one you keep, so Clay skips the dialog — Cancel test drops the second sequence straight away.

FAQs

Can I change the copy once a test is running?

It depends on the campaign's status. Copy edits, adding or removing a sequence, and changing the split each unlock at different points:

‍

‍

Renaming a sequence is available in every status.

Is spintax the same as an A/B test?

No, and the two work well together. Spintax randomizes individual words and phrases inside one sequence's copy so that sends don't read identically, and Clay doesn't track which wording a given lead received. An A/B variant is a tracked arm with results of its own, and a sequence under test can still use spintax throughout.

How does Clay decide which sequence a lead gets?

Each lead is assigned when it enrolls and keeps that sequence for the rest of the campaign. The assignment converges on the split you configured across the full lead set, so a single day's batch of enrollments can look uneven — a 60/40 result on a few hundred leads is normal, and it evens out as the rest of the segment enrolls.