Hi Sumo-lings,

We've shipped a big update to Otter's hosted MCP server, which connects AI assistants to your A/B testing workflow.

Your assistant can now inspect a page, build an experiment draft, check how the treatment renders on desktop and mobile, and give you preview links to review before launch.

Here's what you'll notice:

- Better page inspection. Your assistant gets clearer information about headings, buttons, sections and responsive visibility, helping it choose the right elements to change.

- Rendered variant checks. Otter compares the control and treatment using the same visual-change code your visitors receive. Missing elements, refused edits and JavaScript errors are surfaced. Incomplete checks are reported as incomplete.

- Safer launches. Extra checks catch confirmed rendering failures and incomplete goals before API/MCP experiments go live. You can inspect the findings and decide what to fix.

- Clearer errors and instructions. Your assistant gets more specific feedback when a request fails, including which field needs attention. A guided authoring workflow helps it build a focused draft with a hypothesis and measurable goal.

- Ongoing change health. Sampled diagnostics can flag visual changes that stop applying after a site update. These signals don't change your experiment statistics or automatically stop your tests.

There are now 34 tools on the hosted server, covering projects, experiment setup, inspection, verification, previews, launch checks, results and health. Your account permissions and project restrictions still apply.

Start with the hosted MCP connection guide: https://www.otterab.com/mcp-docs

Try asking: "Inspect my landing page and draft a headline test aimed at increasing clicks on the main CTA. Verify it on desktop and mobile, and share previews before launching."

I'd love to hear what you'd like your assistant to help you test next.

Hey Sumo-lings,

Here's how theme changes usually go on WordPress.

You want to try a new layout. So you edit your live theme, push it, and every visitor gets it. If the numbers go the wrong way, you're unpicking your own changes by hand a fortnight later, hoping you remember all of them.

Otter now gives you the other option.

Duplicate. Change. Test. Delete the loser.

1. Make a copy of your theme — a child theme, or a straight duplicate

2. Make your changes in the copy. Templates, CSS, layout, whatever you want

3. Run the two against each other on real traffic, scored on your goals

4. Keep the winner. Delete the loser.

Your live theme is never edited. It runs as control, exactly as it does today. Everything you're trying lives in a separate theme — and if it loses, you delete a folder. Nothing to revert, nothing left behind.

That's the actual point. It isn't "which theme is better." It's being able to make real changes without betting the site on them.

Why this beats the usual approach

Most A/B tools patch the page with JavaScript after it loads. Fine for a headline. But a layout change means the page visibly rebuilds itself in front of the visitor — and a test that wins leaves you folding a pile of CSS overrides back into your theme by hand.

Otter picks the theme server-side, before WordPress loads a single template file. The first byte is already the right theme, so nothing flashes. And because the winner is a real theme, shipping it means activating it.

No staging site. No second URL. No redirect.

The safety rails

- Only themes you explicitly approve can be served

- Assignment cookies are HMAC-signed — hand-edit one and you get control

- Admin, REST, cron, WP-CLI, feeds and the Customizer always show your real theme

- Returning visitors stay on the theme they were given

- Theme testing stays switched off until you confirm the assignment cookie is excluded from your page cache or CDN. A cache that ignores it would serve one visitor's theme to everyone, so we block the feature rather than risk it.

Getting started

Install the Otter plugin, approve your copy under Settings → Otter A/B Testing, then choose Create a WordPress theme split test. Otter prefills a site-wide test with your current theme as control.

Works with or without WooCommerce.

Go break something safely.

Hey Sumo-lings,

If you run Google Tag Manager, this update removes a step you never should have had to do twice.

What shipped

Otter now has a DataLayer Event goal type. If your site already pushes an event into the GTM dataLayer — generate_lead, purchase, add_to_cart, form_submit, whatever you named it — you can score an A/B test on that event directly.

Setup is one field. Choose "DataLayer Event", type the event name exactly as your site pushes it, done.

Why it's good

No new tag. No code change. No GTM access needed.

That last one matters more than it sounds. If you run an agency, you know the drill: the client's GTM container is behind a login belonging to someone who's on leave, and your test sits in draft for a fortnight waiting on it. Otter reads window.dataLayer straight from the snippet you already installed. Whoever owns GTM never has to be involved.

It also means your test and your client's analytics are counting the same thing. You're scoring on the exact event their dashboard already reports, instead of a second click-selector that quietly drifts away from it.

Confirm it before you launch

Open the Preview Tool on your real site, go to the Goals panel, and push the event. The goal flips to ✓ Fired the moment Otter sees it. Preview bypasses conversion tracking entirely, so fire it as often as you like — nothing is recorded against your results.

Live now for everyone, on every tier, no upgrade and no action needed. Your existing tests are untouched.

Hey Sumo-lings 👋

You've been asking for this one, and as of today it's live for everyone: multi-armed bandit is now a traffic strategy you can pick on any new test.

Quick version of why you'd want it:

A classic A/B test holds your traffic split fixed until you call a winner - still the right tool when you want a clean, rigorous read. But for a promo, a seasonal landing page, or a short campaign, every visitor you send to a losing version is money left on the table. A bandit fixes that by shifting traffic toward the winner while the test is still running.

Here's how ours works:

- Starts fair. Every version gets an even share of traffic while it collects a warm-up sample (50 visitors per version).

- Learns from your primary goal. Conversion rate or revenue per visitor - whichever you set — decides which version is leading.

- Adapts automatically. Most new visitors go to the current leader, and 20% of traffic keeps exploring all versions, so a slow starter can still take the lead.

- Stays consistent. Returning visitors always keep the version they first saw - no flip-flopping mid-session.

- Reports in plain probabilities. Bandit tests use Bayesian reporting automatically, so you see each version's chance to be best instead of squinting at p-values.

You'll find it in the test wizard right next to the usual manual split, and the results page shows exactly how traffic is being allocated and why.

It's included on every AppSumo tier at no extra cost - lifetime deal, lifetime features.

Docs: https://www.otterab.com/docs/building-tests/creating-a-test

Go squeeze more wins out of your traffic 🦦