Self-Improving Email
Give your assistant past email results and it uses them to shape your next Stensul email: module order, copy and subject lines, with the evidence behind each change. Early access.
Self-Improving Email puts your campaign results to work in the next email you build. Your ESP knows how each email performed. Stensul knows what was in it: the modules, their order, the copy, the buttons. Put the two side by side in one conversation and your assistant can tell you what to change, and make the change for you.
What you get depends on your data
The more your data can tie a result to a part of the email, the more specific the advice.
| You give it | It can recommend |
|---|---|
| Open rates, with subject lines | Subject lines and preheaders |
| Opens and clicks, with a tracking value per module | Module order, what goes above the fold, button copy |
| Funnel results (landing page visits, registrations) joined to those tracking values | Which modules drive the outcome you care about |
Click rates are the steadier signal. Opens are inflated by mail privacy features, so a subject line recommendation built on opens alone is weaker than one that also looks at clicks.
Set it up
-
Connect Stensul to your AI host
Follow the quickstart for Claude, ChatGPT, Copilot or another host.
-
Bring your results
Either export them (CSV, Excel or a Google Sheet all work; large files are fine) and attach them to the conversation, or connect the MCP server where they already live, such as Databricks, Snowflake or Marketo, to the same host as Stensul.
-
Check the export has what the assistant needs to match
See the table below. The most common gap is a report with results but no subject lines.
-
Tag links per module for future sends
Give each module’s links its own value, for example
utm_content=hero-ctaorutm_content=webinar-block. One shared value for the whole email means clicks can’t be tied to a module. See tracking parameters.
What to include
| Column | Why |
|---|---|
| Email name or ID, and send date | So the assistant can find the same email in Stensul |
| Subject line and preheader | So open rates can be tied to wording |
| Sends or deliveries, opens, clicks | The basic rates |
| Clicks per link, with its tracking value | So clicks can be tied to a module |
| Conversions (optional) | Registrations, form fills or anything you track past the click |
Prompts to start with
Improve an email you’ve already written
Here's our performance data for every <newsletter name> we've sent. Review it
alongside the Stensul draft "<email name>" and our past emails in Stensul.
We've already refined this copy, so don't rewrite it. Suggest a few tweaks,
with the data behind each one.
Build from a brief
Build a Stensul email from the attached brief in the <library name> library.
Use the attached performance data from our last 20 sends to choose the module
order, subject line and button copy, and tell me which results you based
each choice on.
Use data from a connected source
Our email results are in <Snowflake / Databricks / Marketo>. Pull opens and
clicks for our last 6 months of <campaign type> emails, compare them with the
same emails in Stensul, and suggest changes for the draft "<email name>".
What the assistant does
- Reads your results from the file or the connected source.
- Finds the matching emails in Stensul and reads their structure and copy.
- Matches results to subject lines, modules and buttons.
- Recommends changes, each with the evidence behind it. If there isn’t enough comparable history, it follows the approved template and tells you the draft isn’t based on performance data.
- Applies the ones you accept to the draft, within the library’s guardrails, and shows you a preview.
A recommendation looks like this (example values):
Move the webinar module above the product update. In your last 12 sends,
the module in the second position got 2.4x the clicks of anything below it.
Change the main button from "Learn more" to "See open roles". Emails that
used "See open roles" averaged a 15% higher click rate.
Where it stops today
- It recommends; it can’t promise a lift. Check the evidence, and keep a subject line A/B test in your ESP if you run one.
- It doesn’t measure what happens next. Compare your next send against past ones, or give the assistant the new results and ask again.
- It can only use what it can match. Missing subject lines or one tracking value for the whole email limit what it can say.
- Recommendations come back as text in the conversation. There’s no dedicated chart or report view yet.
- It doesn’t send. Review and send from Stensul.
Your results are shared with your AI host when you attach them or connect a source, so share only what your company’s AI policy allows. See security and data.
Tell us how it went
This is early access because we want to hear what worked. Send what the assistant recommended, what you changed and how the email did to your CSM.
Related
- Tracking parameters
- Brief to draft
- Workflow setup skill, to build drafts from your team’s brief queue