The agent makes testing cheap to set up. It does not make a weak hypothesis worth testing, and it cannot create traffic you do not have.
What Framer agent A/B testing is and why designers should care
Framer agent A/B testing is the ability, released on 3 November 2026, for the built-in Framer Agent to create, edit and analyze A/B tests and funnels in Framer Convert, and to add or edit triggers. According to Framer's Convert update, the same release added country-based triggers, so a site can show an overlay to visitors from a specific country or region (Framer's own example is a banner for visitors in Spain promoting an event in Barcelona).
The short answer for designers: you can now describe an experiment in plain language and let the agent build the variant, wire up the conversion step and read the result back to you. That removes most of the setup friction that kept small teams from testing at all. It does not remove the need for a good hypothesis, enough traffic, or a designer who can tell a real win from noise.
What actually shipped on 3 November
Convert itself is not new. Framer introduced it as an add-on earlier in 2026, bundling Funnels, A/B Tests and Triggers. What changed is who operates it. Until now you set up a test by hand: duplicate the page into a variant, add tracking IDs, configure the conversion step, start the test, publish. The Convert product page now says you can set these tools up "manually or with agents".
The day before, on 2 November, Framer shipped Harness 2, a rebuild of the agent's underlying harness that lets it handle multi-step workflows in one pass and show progress sooner. Framer says those capabilities also extend to external agents through MCP. That matters because an A/B test is exactly the kind of multi-step job (variant, tracking, goal, start, publish) where a slow, step-by-step agent used to fall apart.
Here is what the official A/B testing help article says about the mechanics, whoever sets the test up:
A page can have up to five variants, and traffic is split evenly across them.
Every variant needs the same tracking ID on the link or form you are measuring.
Framer uses a Bayesian method and reports a "Probability of Being Best". A winner can be declared once that passes 90 percent and the test has enough visits.
Tests do not use cookies. Visitor identifiers reset daily, so a conversion only counts if it happens on the same day as the visit. A cookie-based option for longer windows is planned but not available yet.
Convert is an add-on for sites on the Pro or Enterprise plan, billed on monthly analytics events (page views, clicks on tracked links and form submissions).
Manual testing vs agent-run testing in Framer
Criteria | Manual setup in Convert | Framer Agent runs the test |
|---|---|---|
Creating the variant | You duplicate and redesign the page | You describe the change, the agent builds it |
Tracking IDs and goal | You add them to each variant | The agent adds and configures them |
Reading results | You open the Analytics tab | You ask the agent to analyze the test or funnel |
Country or region targeting | Triggers set by hand | Agent can add or edit triggers, including country-based ones |
Risk of mistakes | Missed tracking IDs, mismatched goals | Variants that drift beyond the hypothesis, unreviewed copy |
Best for | One carefully crafted experiment | Teams that want to test often and keep a backlog moving |
The verdict: use the agent for setup and analysis, keep the hypothesis and the final design call for yourself. The manual route still wins when a variant is a genuinely new layout that needs real craft, because the agent can produce a variant faster than you can judge whether it is good.
How to run your first agent-built test: the Ask, Approve, Audit loop
A simple loop keeps the speed without handing over judgment.
Ask. Write one hypothesis per test, and make it specific: "Replacing the hero headline with a pricing-first message will increase clicks on Book a call." Tell the agent which page, which element changes, and which link or form is the conversion step. Ask it to change only that element. A variant that rewrites the headline, swaps the image and moves the button teaches you nothing, because you will not know which change did the work.
Approve. Before the test starts, open every variant yourself. Check the copy for tone, the layout on mobile, and that the tracking ID really sits on the same call to action in each version. Framer's help article warns against changing variants while a test runs, so this is your one clean chance to fix things.
Audit. When you ask the agent for results, ask for the numbers Framer reports (visits, conversions, lift, Probability of Being Best), not just its summary. If the result is inconclusive, accept it. Framer itself says low traffic can mean a test never reaches a conclusion. Then stop the test, choose the winner, publish, and write down what you learned before the next experiment.
Country-based triggers deserve their own care. They are useful for regional events, shipping notices or language hints. They are also an easy way to annoy people with one more overlay. Treat every trigger as a design decision with a reason, and remove the ones that have outlived it.
Hype vs reality
The hype says the agent now "does CRO for you". The reality is narrower and still useful. The agent removes setup work and makes analysis conversational. It cannot generate traffic, it cannot invent a good hypothesis from nothing, and it cannot fix the same-day measurement window Framer describes. On a low-traffic site, faster test setup mostly means faster inconclusive results.
What holds up: for teams with steady traffic and a clear goal, the cost of trying an idea drops close to the cost of describing it. That changes behaviour. Testing stops being a quarterly project and becomes part of everyday design work. An earlier article on the future of landing pages argued that pages will be judged more by how they convert the visitors who still arrive. Agent-run testing is one of the tools that makes that judgment cheaper.
Where this takes design work next
Over the next one to three years, expect the loop between design and evidence to tighten. Designers who used to hand a page to a growth team will increasingly own the experiment backlog themselves, with an agent doing the plumbing. The valuable skill becomes framing: picking the question worth testing, designing variants that isolate one idea, and knowing when a result is noise.
It also pushes design systems further. If an agent builds variants, it should build them from your components and styles, not from improvised layouts. Teams that already treat their system as AI context (see the guide to Framer skills for designers) will get cleaner variants. And if you run your site from Claude, Codex or Cursor, the Framer MCP guide for designers covers how external agents connect.
Frequently asked questions
Will the Framer Agent replace designers in conversion work?
No. It replaces setup chores: duplicating pages, adding tracking IDs, configuring goals and reading dashboards. The parts that decide whether a test is worth running (the hypothesis, the variant design, the judgment on results) still need a designer. If anything, the role grows, because more experiments mean more design decisions to make well.
Do I need a paid plan to use Framer agent A/B testing?
Yes. Framer's help article says A/B testing and Funnels are part of Convert, an add-on for sites on the Pro or Enterprise plan, billed on monthly analytics events. Check Framer's pricing page for current add-on rates before you start, since they are not listed on the update itself.
How long should a Framer A/B test run?
There is no fixed duration. Framer declares a winner when Probability of Being Best passes 90 percent and the test has enough visits. Low-traffic pages may never reach that point. Because visitor identifiers reset daily, conversions that happen days after a visit are not counted yet.
Is agent-run A/B testing overhyped?
Partly. It does not make a site convert better by itself, and it is only as good as the hypothesis you give it. It is not hype that setup becomes much faster, which makes regular testing realistic for small teams that skipped it before.
Can I target visitors by country with Framer triggers?
Yes. The 3 November update added country-based triggers, so you can show overlays to visitors from a chosen country or region, and the agent can add or edit those triggers for you.
Getting expert help
If you want an experiment programme designed around your product, not just a few headline swaps, a team that pairs website redesign with AI product development can set up the hypotheses, the variants and the design system the agent works from. DIGCY is a Dribbble Selected Agency and a UI/UX agency in Casablanca working with clients globally. Start small: one page, one hypothesis, one conversion goal, and let the agent handle the rest.
Let’s keep in touch.
Discover more about high-performance web design. Follow us on Twitter and Instagram.




