Learn and master Clay with Clay university

A snippet is a blank section you leave in your email that the AI fills, based on the context of your business and the context of your lead.

What changed in this version of Sequencer is how that blank gets filled. There's a single agent that reads your entire sequence, looks at all the context about your company, and writes every snippet with all of it in view. It isn't filling each blank in isolation.

That matters because the snippet in message two knows what message one said, so it has the context of the whole campaign when it drafts.

Authoring one

1. Insert a snippet in the message and describe what you want it to generate
2. Read the preview it produces for the current lead
3. Page through several leads to see what it writes for each of them
4. Accept or discard it

In the example, the first snippet comes back written mostly from the business context set in settings: an introduction naming a software engineering manager in New York City who offers freelance web development support for startups and growing companies, ending with a request for a brief demo.

Adding an instruction not to include any CTAs and updating removes the closing ask from the generated copy. The snippet responds to the instructions you give it, so tighten the description until the output is the shape you want.

Page through leads before you accept

Switch leads and read what changes. In this campaign, a snippet written for one lead comes back meaningfully different from the one written for another.

The outputs here don't differ dramatically, since these are fake leads without much context on them, so the model can only work with what it has.

If your snippets aren't different enough from lead to lead, that's telling you the context is too thin. Go back and give it more: add to your AI context, add to your business settings, drop in more documentation, or put more detail in the campaign goal.

🎛️ Keep Your CTA Out of It

You could hand the whole email to the model, and plenty of tools work that way. But most of the time that's not what you want. Some parts of the email should be deterministic, and you don't want AI changing them.

Your call to action is the clearest case. You may want it to be exactly what you wrote, every single time. If the line is "I'd love to book 15 minutes to talk through whether the services are something you're interested in," followed by a booking link, that sentence should sit at the bottom of the email and read the same on every send. There is no upside to a model rephrasing it on the fly.

So the shape to aim for is:

• The opening: personalized, written by the model
• The middle: details you want identical on every send
• The CTA: your exact copy, hard-coded, every time

AI snippets and the copy you write by hand are meant to work together like this. Be non-deterministic where variation helps you, and deterministic where it doesn't.

💳 What a Snippet Costs

Keep an eye on the cost chip at the top of the editor: one action to sequence a lead, and one more when the message uses AI. In the same place you can track the action credits and data credits you're spending, so the more AI snippets you use, the more credits you'll be spending.

🔀 Spintax

Spintax rotates a phrase across a few pre-selected options, and each send picks one at random. You can give it between two and five variations. The syntax is curly braces with pipes between the options:

{Hi|Hello|Bonjour}


Save that and you have three different greetings getting selected per send.

Why bother

If you send 10,000 messages and all 10,000 have the identical greeting, that pattern is itself a signal. Providers can see that 10,000 people received the same text, character for character, and that's going to get you dinged on your domain reputation. Varying a few phrases breaks the pattern for almost no effort.

Good places for it: your greeting, your transition sentences, your sign-off.

Bad place for it: anywhere the exact wording matters. That's the same rule we just applied to the CTA, so you probably don't want spintax there either.

🚀 What's Next

You now know how to author an AI snippet, check it lead by lead before accepting it, keep the parts of your email that must stay exact out of the model's hands, and use spintax where variation is cheap insurance.

Next we'll cover A/B testing, so you can find out whether the personalized version actually beats the plain one.

‍