Personalize with AI

Create one email whose message changes with the customer’s circumstances. In this guide, Loopwell uses Adaptyle to suggest a first project to three new signups. The design and call to action stay the same.

What you will build

Maya is coordinating a launch, Leila has interview notes to organize, and Hugo is preparing a client handoff. Each has zero projects of their own. They need different advice from the same onboarding email.

These saved, unedited outputs were generated with fictional customer data. Maya gets a launch sign-off project, Leila gets an interview-synthesis project, and Hugo gets a client-handoff project. Choose Template to read the shared instruction.

Before you begin

You need a draft template, broadcast or workflow email, a sender, a published brand kit and AI credits for previews. Use your own customer goal and progress fields; Loopwell is an example, not a required data model.

For this example, the email needs each customer’s first name, goal, own-project count, work already owned by teammates and any unfinished task. Stored fields arrive with the record. Select Evaluated attributes and related records in Data when you need them.

Mark a passage for maxinja

1. Choose the paragraph that should change

   Open the email in the visual composer. Keep the greeting, button and required campaign facts outside the paragraph you want to personalize. For Loopwell, only the project recommendation changes.

2. Add the Adaptyle instruction

   Select that paragraph and choose Adaptyle from the text menu. In the block’s instruction, describe the writing job:

   Suggest a useful first project from their goal, progress and team context. Give one next step. Avoid work their teammates have already started.

   Add writing rules that fit your campaign: use only supplied facts, keep the paragraph short and do not invent product features. Type / in the instruction for suggestions based on the data the email carries.

3. Select the data behind the message

   Open Data, choose a sample customer and inspect the field values. For Maya, the relevant difference is that Daniel owns launch assets while sales and design sign-off is still missing. Her own-project count is zero; her teammate’s project does not make it one.

   Choose your published brand kit for writing guidance. If a required field is absent, supply it or give the instruction an explicit fallback, such as “If their goal is missing, describe how to name a project and add one next step without assuming a goal.”

4. Generate the recipient preview

   Under Personalize → Runs with, choose Model and Thinking. Both are required; the workspace’s default model does not fill them in. Open Preview, choose Personalized, and click Generate preview.

   Base shows the email without the AI rewrite. Personalized shows the generated email for the selected data. The Adaptyle instruction and editing outline do not appear in the recipient’s email.

5. Compare another customer

   Return to Data, select a different contact and regenerate. Compare Leila’s interview notes with Hugo’s handoff: the paragraph should reflect their different work. Check the greeting, unchanged call to action and its destination as well as the generated paragraph.

Find the controls in the composer

App appearanceMatch pageLightDarkView full size

To stop personalizing a passage, select it and choose Remove Adaptyle. The authored content remains. To refine the saved draft with the assistant, ask maxinja to change that paragraph; to change future recipient wording, edit its Adaptyle instruction.

Compare a representative set

Check these cases before using the email in a campaign:

Open Checks in the editor’s right panel. Resolve its issues, inspect representative personalized samples and send an inbox test. A clean draft and a successful sample do not establish that every future output will be correct or that a campaign has been delivered.

For a broadcast that later stops on a quality hold, inspect the actual recipient output in Quality review. That review concerns a particular generated email; it is separate from approving the draft or looking at a sample. See review a quality hold.

If a render rejects a protected link, markup or fixed text, narrow the instruction. Make layout changes in the draft and keep recipient-time instructions focused on writing. See Adaptyle validation.

What it costs

Previews, regenerations and each recipient render that needs Adaptyle use AI credits. Select only the passages where context changes the message. Liquid substitutions and fixed copy do not themselves need a generative call.

Use Personalize → Estimate before scaling. Subject and body are generated together; a correction or campaign review can add calls, so one prompt does not equal one billable call. Sending allowances, overages and other operation charges are separate. The estimate marks unknown components, and one selected sample is not an audience-wide average.

Getting good results

• Give the instruction a concrete writing job, such as recommending a first project from the customer’s recorded goal.
• Select the required customer context and state what to do when it is missing.
• Keep links, offer terms and required wording outside AI regions.
• Compare different customers. A later generation can produce different wording.
• Keep the email focused. More than 10 instructions triggers a warning.

Related

Use the same email in a broadcast or a workflow. For code-editor syntax, rendering rules and failure recovery, see the Adaptyle reference.