AI variant suggestions
Once you have a few completed campaigns, liftstack can generate new variant ideas based on what has actually worked for your audience. This is not a generic AI copywriter; suggestions are grounded in your actual test data.
Requires an LLM connection configured in Settings > AI. Available on all plans with a configured API key.
How it works
On any snippet detail page, look for the “Suggest variants from winning patterns” panel. When you request suggestions, liftstack:
- Analyses your historical test results to identify patterns that consistently win (tone, structure, length, framing)
- Injects the top winning patterns as context for the AI
- Generates new variant content that follows those patterns while remaining distinct from what you have already tested
Each suggestion shows why it was generated (e.g., “urgency framing has won 5 of your last 7 hero tests”) so you can evaluate whether the reasoning makes sense.
You can accept a suggestion to create a new variant directly, edit it first, or dismiss it.
Brand guidelines
If you have configured brand guidelines in your workspace LLM settings (Settings > AI), the suggestions respect your brand voice and style constraints. This prevents the AI from generating content that feels off-brand, even when following data-driven patterns.
Winning pattern analysis
Behind the scenes, liftstack identifies winning patterns by looking at attribute-value combinations (from content attributes) that correlate with winning variants. A pattern qualifies as “winning” when:
- The variant has appeared in at least 3 completed campaigns
- The variant has won at least once
- The variant’s win rate is 40% or higher
Patterns are sorted by frequency, with the top 5 by win rate included in the AI generation prompt.
Without an LLM connection
AI variant suggestions require a configured LLM provider. Without one, the suggestion panel is not available. You can still use Content Intelligence insights (limited to automatic and rule-based attributes) and the Content Playbook to inform your variant creation manually.
Template scanning
liftstack can also scan your existing ESP templates to identify content areas suitable for A/B testing. This runs automatically in the background.
For each template in your connected platform, liftstack detects:
| Block type | How it is detected |
|---|---|
| Subject line | Always present |
| Headings | H1-H3 HTML tag patterns |
| CTA text | Button and link elements with button styling |
| Preheader | Elements with preheader class or HTML comment patterns |
Template scanning is entirely rule-based (no AI required) and works on all plans. Use it to spot testing opportunities in templates you have not experimented with yet.