One sentence in. Reviewed before it goes live.
A marketer typed one line to Personyze AI: “Win back everyone who has been quiet for 30 days with what they used to browse.” The agent compiled real targeting, built the content in the store’s own style and staged it — and a person published it. Switch the visitor below to see who gets what, and the request itself.
Built with AI Personalization Agent — see the feature →
The knitwear you were looking at is still here
The sweaters you viewed in August, and what is new in knitwear since. Everything below is in stock today.
I had forgotten which sweater I liked. The page remembered, and it was still there in my size.Verified buyer — Columbus, OH
Describe it. Review it. Publish it.
Say what you want
In the Personyze panel, or in Claude or ChatGPT through the Personyze MCP server. Plain words: who it is for, and what they should see.
It builds the real thing
Targeting compiled into actual rules and read back to you, content made in your own fonts and colors and placed on the page, with a control group if you want one.
A person publishes it
Everything lands in staging mode, visible through a preview link and from staging IPs only. Real visitors see nothing new until someone publishes it in the panel.
Free AI tool · Inspire me
See ideas like this for your own site.
How Personyze AI built this win-back campaign.
One sentence became real targeting
“Quiet for 30 days” compiled to a time-since-last-visit rule, and “what they used to browse” to products viewed in past sessions. The agent reads the rules back in the chat, because a rule that says something slightly different looks identical from the outside.
Content in the store’s own style
The welcome-back hero and the row under it use the store’s fonts and colors, in the place the normal hero sits. There is one hero for knitwear browsers and one for outerwear browsers — switch between the two quiet shoppers to see both.
What they used to browse, in stock only
The row is each visitor’s recently viewed items, filtered to what is in stock, with bestsellers as the fallback — so nobody is welcomed back with a sweater they cannot buy.
Nothing went live on its own
The agent created the campaign in staging mode, and the preview link shows it on the real site to the reviewer only — that is the request state. It reached the two quiet shoppers because a person checked it and published it.
An active shopper sees the page as it was
The visitor who was here two days ago does not match the rules, so the banner, hero and bestsellers are untouched. The campaign changes the page only for the audience it compiled.
A control group keeps the win-back honest
At the marketer’s yes, the agent held back 10% of the audience, split by visitor, so they see the normal homepage. The campaign is measured against people who came back anyway, not against nothing.
See how it is built, and where else it fits
Questions about this example
Where can I ask the Personyze AI agent to build a campaign?
In the Personyze panel, or from Claude, ChatGPT or another MCP assistant through the Personyze MCP server. It is the same agent working on the same account either way.
Does the agent publish campaigns by itself?
No. A campaign it builds is created in staging mode, which only a reviewer sees, through the preview link it returns or from a staging IP. Real visitors see it only after a person publishes it in the panel.
What does it build from one sentence?
The targeting, compiled into real rules and read back to you so you can check it says what you meant; the content, designed in your site's own fonts and colors and placed on the page; and, where it fits, a traffic split that holds back a control group.
Describe your next campaign.
Ask in the panel, or from Claude or ChatGPT. Personyze AI compiles the targeting, builds the content in your style and stages it — you review it and decide what goes live.