How the AI Segment Builder Works
Klaviyo introduced a feature inside the Lists & Segments tab called "Define with AI". The concept is simple: instead of manually stringing together conditions, dropdowns, and timeframes, you write a natural language prompt describing who you want to target, click preview, and Klaviyo generates the logic for you.
On paper, this sounds like a massive time-saver. You describe what you want, Klaviyo interprets it, and you get a clean segment in seconds. But in practice, the tool introduces subtle logic errors that can quietly wreck your campaign targeting.
I tested the tool across standard ecommerce use cases to see whether it holds up. If I am being completely honest, I would not use it in its current state.
Where the Default Logic Falls Apart
The biggest issue with the AI generator is that it makes assumptions about what terms mean, and those assumptions rarely match how an experienced marketer actually builds an engaged segment.
When you ask for "engaged customers", the AI tool typically generates a definition based on opened emails or clicked SMS messages in the last 30 days, paired with an "Active on Site" metric of at least two visits in the last 45 days. That sounds reasonable at first glance, but it creates two glaring blind spots:
- It excludes brand new subscribers: If someone subscribed to your newsletter yesterday, they have not had time to open three campaigns or trigger multiple site visits yet. Under the AI logic, they are excluded from your engaged segment entirely.
- It relies heavily on site tracking: Site activity can be messy depending on cookie consent and browser tracking restrictions. Making on-site activity a mandatory "AND" condition unnecessarily filters out real, active buyers.
That is a mistake. In my books, an engaged segment should always include recent signups (such as "Subscribed to list at least once in the last 30 days") and look at a wider window, like 60 or 90 days, depending on your buying cycle.
Product and Category Matching Fails
Things get even shakier when you try to use natural language to segment by purchasing behaviour. I ran a test asking the tool to build a segment of subscribers who engaged in the last 90 days and bought posters before.
The AI generated a condition looking for "Ordered Product is greater than 0 overall time where Name equals Poster". That condition resulted in exactly zero profiles. Why? Because the client store does not have an individual SKU titled "Poster". They sell specific items like "Vintage Botanical Poster" or have product categories containing the word poster.
To capture those buyers, the condition needed to use "contains Poster" or filter by collections passed through from Shopify. The AI missed that distinction completely. If you do not catch that mistake yourself, you end up sending a dedicated campaign to a ghost segment containing zero people.
The 280-Character Bottleneck
To fix those mistakes, you might think you can just write a more detailed prompt. Here is the problem: Klaviyo caps your prompt input at 280 characters.
By the time you explain that you want email engagement over 90 days, recent subscribers included, and product names that *contain* a specific keyword rather than matching it exactly, you hit the character limit. You end up spending five minutes rewriting prompts and debugging the generated preview.
In the time it takes to prompt the AI and correct its assumptions, you could build the segment manually in 30 seconds. If a productivity feature takes longer than doing the work by hand, it fails at its primary job.
When Deep Segmentation Stops Making Commercial Sense
Beyond the AI tool itself, this highlights a broader debate around segmentation. There is a common belief in ecommerce that you must slice your audience into dozens of ultra-niche micro-segments for every single send.
Unless you have a large database (roughly over 50,000 active profiles), hyper-segmenting your business-as-usual campaigns rarely delivers a positive return on effort. Let's look at the numbers:
- The micro-segment approach: You spend two hours designing and writing a campaign tailored specifically to people who live in a specific city and bought a specific accessory. In a list under 50,000 subscribers, that segment might only hold 1,000 people.
- The engaged broadcast approach: You spend that same time polishing a strong, high-converting campaign sent to your full 90-day engaged segment of 35,000 subscribers.
The return on investment on your time is simply not there for tiny segments on regular sends. Keep your BAU campaigns focused on clean, reliable engaged segments, and save deep segmentation for automated trigger splits where the work pays off continuously.
Final Thoughts
Klaviyo's AI segmentation tool is an interesting experiment, but right now it is too rigid and prone to silent errors. It requires experienced oversight to spot the flawed logic, yet anyone with that experience can build the segment faster themselves. Stick to manual definitions until the prompt controls and data mapping improve.
Ready to Clean Up Your Klaviyo Setup?
If you want to ensure your deliverability is protected and your segments are capturing real buyers, have our team review your setup with a comprehensive Klaviyo account audit.
