Convert.com

Robust A/B testing features at 1/5th enterprise tool prices | A/B Testing Tool of the Year at Experimentation Elite Awards 2024

[Convert.com @Convert]

Start Optimizing for Mobile Traffic: How Ecom Stores Can Sell More Without More Ad Spend. https://t.co/0T1I4Y9pOi

[Convert.com @Convert]

Prepare your website for AI & AI agents https://t.co/uUFSxh2wfy

[Convert.com @Convert]

He makes music, art, and furniture.

All the pursuits need discipline and rigour to shape creative urges.

That’s Matt Policastro’s strength.

He knows how to introduce desirable friction to enhance discipline, while never letting momentum erode.

In his session, he will let an LLM design & ship a test in real time.

That’s not the flex. The point comes later.

When a test costs almost nothing to run, half-baked ideas ship at scale - masquerading as flow, and "test velocity" turns quite meaningless.

The counter-intuitive fix: slow down, on purpose. Plug the process gaps.

Matt’s framework does just that. And it works. Period.

Thirty minutes, 16:05 BST on 9 September.

#WiE26

https://t.co/1MFyJHgStw

[Convert.com @Convert]

Everyone has AI.

The gap moved to data.

That's the short version of what came out of ROI Revolution client summit in Raleigh last week.

Four takeaways worth repeating:

-- AI on its own stopped being a differentiator.

What separates brands now is first-party data, clear positioning, quality inputs and disciplined execution.

-- Authenticity turned into a strategy.

In a world of generated content, the brands that sound specific and human are the ones getting noticed.

-- Measurement is changing shape.

Clicks and rankings are no longer the whole picture. Visibility, citations and influence are joining them.

-- The fundamentals got more valuable, not less.

Clean data and real customer understanding are worth more in an AI environment than they were before it.

None of that is new advice. Just harder to ignore now.

Thanks for the sponsor slot, and for sending the photos over!

[Convert.com @Convert]

Every tourism experimentation case study is about Booking, Airbnb or Expedia.

None of them look anything like your business.

Booking alone keeps over a thousand experiments running at any given moment, which makes testing look like a normal part of running a tourism company. It isn't. The industry is mostly smaller operators scattered across the globe, and most of them have never run an A/B test.

So we asked the people working at that end of the market what actually happens.

Three things make tourism different from the standard playbook.

1. Friction can help.

Ryan Thomas tested showing visitors the available homes on a home-exchange platform, expecting the properties to build desire. It backfired. What worked was a signup flow that explained how the platform worked first. Education created the desire, in that order.

Jono Matla saw a version of the same thing. A more intuitive booking engine produced inconclusive results, because intent ran high enough that people tolerated the old one and finished anyway.

2. Localisation decides the result, not just the language.

Olivia Bedford frames safari copy for UK audiences around health and safety: fenced lodges, malaria-free areas, child safety in open vehicles. The US market is aspirational and less safety-anxious, so the same copy underperforms.

Laura Duhommet found Spanish visitors at Disneyland Paris needed far longer to understand dining options than other markets, and got different results testing price-per-person against price-per-night across French, German and English users.

3. The purchase is logistical and aspirational at once.

Laura had to make people dream before checkout, which shaped what she tested: video against high-quality images, microcopy on product pages. Jono's addition is that the dreaming has to survive arrival, so the airport pickup and the check-in desk are testable steps too.

Then there's why so little of this happens. Olivia says CRO is still a brand-new concept in the hospitality market she works in. Jono found hotels selling up to 70% of their stock to travel agents at a discount for guaranteed bookings, so optimising direct conversion never became the priority. The money was already fine.

That's who this guide is written for. Small operators without a dedicated team, who have been reading case studies that don't apply to them.

One stat worth the click on its own: 60% of passengers said the environment mattered when choosing between two tour boats. 14% could say whether the one they picked was certified.

Priscillia Egbo and Trina Moitra put the full guide together.

https://t.co/SeyiF9LucB

[Convert.com @Convert]

The best ideas are useless - unless you can explain them in a way that makes sense to others.

That’s the promise.

Of Ester Serra - founder at Dataverto.

And a WiE Summit speaker with the spunk to say: maybe a stat sig result isn’t cause to celebrate.

Ester’s session won’t be stats chatter and Goodhart’s Law repackaged. It is a structured reminder that experimentation is supposed to find solutions to business problems. Not improve metrics in isolation.

Her framework is about reading familiar numbers as what they often are → hiding places for the trade-offs that sabotage a CRO program.

What you’ll learn: How to balance primary metrics against guardrails, without cherry picking, to create genuine value through experimentation.

Thirty minutes on 9 September, 15:15 BST. You’ll be comfortable defending that test result in any room.

#WiE26

https://t.co/1MFyJHhqj4

[Convert.com @Convert]

There's a version of your website nobody has ever looked at.
Including you.

Slobodan Manić opened the agent-facing version of his own Shopify store to find out what was actually on it. Product title, price, sizing, stock status.

That was the entire page. Everything a designer put on it is gone.

"It's not a bug," he says. "That's the interface."

Shopify switched it on across the entire platform on 5 August.

And this isn't a Shopify story. Cloudflare is building in the same direction.

It's turning into a web standard.

Which leaves a question worth sitting with.

On that side of your website, what is your brand?

The design doesn't make it through. Neither does the copy, or the social proof you spent six months collecting.

What's left is the data.

Whether your prices are right.

Whether your stock is accurate.

Whether a machine can parse your returns policy.

Every website now has a non-visual layer, and almost nobody is designing it.

That's what Sani is covering on 27 August. Four pillars, live audits of well-known sites taken apart on the call, and a scorecard plus instant site checker you keep afterwards.

Eighteen months of running sites through his own audit sits behind it.

No hype, plain language, receipts.

Free, and the replay goes to everyone who registers.

Submit your own site when you sign up if you want it on the call.

https://t.co/s4EAYGXZ51

[Convert.com @Convert]

The algorithm can’t lick a stamp.

A robotic dog can’t lick you…. Can it?

You’ll have to ask our speaker Lauren Wong. She’s the one who pitched it, and won “Toy of the Year” to boot.

Lauren was at HelloFresh Canada.

She’s advised consumer brands like Huggies, Spin Master, and Sano Genetics.

Lauren knows how to find brilliant, underdog ideas.

And if she says GenZ is bringing back landlines -- the trend is worth paying attention to.

Her session focuses on building offline efforts into a performance channel.

Measurable & testable. Yes -- this *IS* the “leverage the unscalable” playbook that viral LinkedIn posts talk about but never break down.

Probably the last real moat left for growth.

Don’t miss it!

14:20 BST on 9th September.

#WiE26

https://t.co/1MFyJHgStw

[Convert.com @Convert]

Adding llms.txt lifts your LLM visibility by 97%.

Wrong number. There's a 97% chance nobody ever reads it.

A study published this summer tracked 137,000 domains with llms.txt on them. Three percent of those files were ever accessed.

That's Slobodan Manić's shorthand for the state of AI visibility right now.

Plenty of it sounds like a good idea.

Very little has proof. Some of it can actively hurt you.

llms.txt might work eventually.

Today there's no evidence it does anything at all.

What Sani covers on 27 August is the other category. The fundamentals that already worked for humans, then for search engines, and now for models.

Four pillars:

- Why machine-ready websites are a CRO concern in the first place.

- How machines actually parse a page. Raw responses with no JavaScript, the accessibility tree, and tokens as a budget you can overspend.

- Semantic structure, schema accuracy, content parity and entity clarity, including a live test where a page goes to a model and its answers get graded.

- Provenance, freshness and consistency. Why answer engines that can be held liable get fussy about sources, and which AI visibility metrics are worth tracking.

He's spent 18 months putting sites through his own audit to work this out. In his words: not magical, boring work, but work that needs doing.

Live audits of brands you'll recognise, and you can submit your own site when you register.

Free, 27 August, 10am EDT / 4pm CET. Normally a paid workshop.

Attendees get the audit tools and installable skills afterwards.

Sani's picking a few of the submitted sites to audit live. Worth adding yours in the form. Register now:
https://t.co/s4EAYGXrft

[Convert.com @Convert]

Someone on your team is about to edit a live test.

Nothing will flag or block it. Until now.

The Request Approval System puts a reviewer in front of three actions: activating an experience, pausing one, and modifying one that's already running.

Each rule is separate, so you gate only what you actually worry about.

That solves:

"Nothing should go live on a client's site without me seeing it."
Turn on the Activate rule and add yourself as reviewer. Scheduling is covered by the same rule, because a scheduled test activates itself later and would otherwise slip straight past.

"My juniors should be able to build tests, not launch them."
Collaborators can request approval, never grant it. They build the whole experience, goals, audiences, variations, and then the launch waits for you.

"A running test should be harder to touch than a draft."
The Modify rule only fires on active experiences. Drafts stay exactly as fast as they were.

"When a client asks who approved that change, I need an answer."
Every request records who asked, what they asked for, who decided, when, and the reason if it was rejected.

"I need to say no without stalling the person who asked."
The rejection reason is required rather than optional. "The test is on day 9 of 14. Hold this change until it concludes." Then they revise and resubmit.

"One client needs tight control. The rest don't."
Scope is set per rule. Account-wide, specific projects, or a single experience inside one project.

Three things we decided on purpose:

- Approval doesn't apply the action.

- The requester returns and does it again.

- An approval is permission, not a remote trigger.

One approval covers one action, not a window of open access.

Nobody approves their own change. A reviewer's own request routes to the other reviewers.

Turning it on changes nothing that's already running.

The rule waits for the next attempt.

Beta on Convert Pro and Enterprise.
Optional, and nothing happens until you enable a rule.

[Convert.com @Convert]

What do fine wine and fine design have in common?

Ask Victoria Bakos and she’ll say “quite a lot”.

The opening speaker on Day 1 of the Women in Experimentation Summit prioritizes joy in all she does.

She had a fine wine business. And she picked up an important lesson there: sameness kills.

Victoria is bringing this insight to her talk.

Today, the Claude UX factory is running overtime.

The same hero, the same gradient, the same clean card grid - okay & totally forgettable. Time to shut down the cheerless assembly line. While retaining solid efficiency gains from AI.

Three things she will cover:

- The visual tells of AI-generated design, so you catch them in your own work before anyone else does.

- Where AI-built flows fail real users, which turns out to be a different question from where they look wrong.

- The exact point in your process where AI makes sense and the parts worth keeping manual.

Victoria works across UX, UI and CRO at Miso.

Her solution is born of hands-on experience.

9th September, 13:50 BST.

#WiE26

https://t.co/1MFyJHgStw

[Convert.com @Convert]

Your website looks great.

But the version AI agents read is almost entirely blank.

Load your top page with JavaScript disabled and see what survives.

Everything that vanished was decoration.

A thousand Google searches now produce about 300 website visits. The other 700 got their answer somewhere else, increasingly through ChatGPT, Perplexity, Claude and Gemini.

Check your referrals. If you're running tests, that traffic is probably already sitting in your samples, unsegmented.

Four questions decide whether any of it reaches you.

Identify. Can a machine tell who you are?

Read. Can it get your content at all?

Understand. Can it tell what your content means?

Act. Can it do something once it's convinced?

Teams tend to score well on Understand and fail quietly on Read. Fix Read first and every other score goes up.

The carousel takes the five mechanisms sitting behind those four questions: structured data, third-party proof, documentation, whether an agent can actually complete your conversion, and entity clarity.

Three of the sharpest fixes in there:

Accuracy beats coverage on schema. One correct Organization and Product markup outranks twelve types that contradict your visible copy.

Put a visible last-updated date on every doc page. Freshness is a signal a model can verify in one pass.

Write one canonical 30-word description of your company and ship it everywhere unchanged, schema included. Three different descriptions read as three different companies.

There's a self-test at the end. Five checks, ten minutes. Two fails is normal. Four means agents are choosing your competitor by default.

Slobodan Manić built the scorecard behind all of it, from 18 months running sites through his own AI-visibility audit. He found patterns.

On 27 August, he hands them over.

You won't find this in the GEO cheatsheets currently flooding your feed.

Attendees get the scorecard, an instant site checker and the audit skills.

Live audits of brands you'll recognise, and you can submit your own site when you register.

https://t.co/s4EAYGXrft

[Convert.com @Convert]

The needle on female representation moved too slow.

So we tested the excuse behind it.

The excuse was that there are few seasoned women speakers around.

In 2024, Daphne Tideman, Lucia van den Brink, Trina Moitra and Marcella Sullivan flipped the script on the 'buts'.

A unicorn summit - that wasn't thought possible. Made for the community, by the community with excellence through (not despite) diversity at the core.

This year, 121 people applied to speak at the 3rd edition of that same event: Women in Experimentation Summit.

Human judges spent hours and hours scoring and curating frameworks and knowledge rare in the busy world of AI-isms.

What survived runs from queering CRO to learning experiment design with dolphins.

Speaking: Victoria Bakos, Lauren Wong, Ester Serra, Matt Policastro, Julie van der Hoop, PhD, Sabine Harnau, Himani Kothari, Nikki Elbaz, Iqbal Ali and Liliann Malcolm
.
Judging all 121: Carmen Apostu, Abi Hough and Talia Wolf.

Also on the agenda: a Women in Leadership panel with Ioana Iordache, Krys L. and Deborah O'Malley, an unconference on today's job hunt with Kevin Anderson and Daphne Tideman, and music from Josh Silverbauer👨‍🎤.

9 and 10 September, BST / AM EDT both days.

Save your seat.

https://t.co/1MFyJHhqj4

#WiE26

[Convert.com @Convert]

1000 searches on Google.

Only 300 end in website visits.

Every brand out there is competing for this shrinking pool of traditional traffic.

There's a silver lining though: AI agents & answer engines are a growing segment and they do convert well when they show up with intent.

If you'd like to meaningfully offset the losses from zero-click search behavior, machines must be able to: identify you, read you, understand you, and act on you.

Attend the first of our 2-part masterclass where Slobodan Manić will unhype and untangle the knots of AI discoverability and site navigation.

Slobodan (Sani) has spent the last 18 months running websites through his own technical audit that evaluates AI visibility and conversions. He has found patterns. Patterns he will now share with you.

We'll cover:

1. Why AI and machine-ready websites are a CRO concern.

2. How machines parse websites: raw responses without JavaScript, the accessibility tree, tokens as a budget.

3. Semantic structure, schema accuracy, content parity, entity clarity, plus a test where we feed a page to a model and grade its answers.

4. Provenance, freshness, and consistency. Why answer engines that can be held liable get picky about sources, and which AI visibility metrics are worth tracking.

You won't find this information in the AEO/GEO cheatsheets flooding LinkedIn.

We will turn the learning actionable with live audits of public, well-known brands and how their sites score against Slobodan's machine-ready criteria.

Attendees will also receive tools to audit and improve their own websites.

** Replay available.

PS: This is a paid signature workshop made free for you by https://t.co/WQWQf9trqB.

https://t.co/s4EAYGXrft

[Convert.com @Convert]

You can A/B test on AI crawlers.

Almost nobody is doing it yet.

How is it possible?

llms.txt is a proposed standard from Jeremy Howard: a plain text file at /llms.txt giving AI models a token-efficient summary of your site, so they can answer questions about you without parsing your HTML. Robots.txt for AI, roughly.

Adoption is still under 1%. But there is enough noise and research to warrant a closer look. Convert Experiences allows you to do that.

Through our Node SDK.

Here are the few things to keep in mind:

1. Bucket without cookies

Most AI crawlers don't accept cookies. Fall back to a hash of User-Agent plus IP so the same crawler lands in the same variation every visit.

2. Take consistent decisions across routes

Make the A/B decision once. Then every page the crawlers hit - /pricing, /llms.txt, all of it - should respect it.

3. Cache in a way that doesn't kill the test

* Private caching → each visitor (and each Google crawler) sticks with the variant they were served.

* Public caching → your CDN hands everyone the same saved copy. The test looks live. It isn't.

* Running a CDN? Add Vary: Cookie - it tells the CDN "not everyone gets the same page."

4. Measurement, since you can't track a text file

There are no pageviews, no events, and no pixel to fire. Put UTM parameters on every link inside the file, tagged by variation, then read the traffic in GA4.

Monitor the quality of AI-referred traffic that lands. Bounce rate, pages per session, conversions.

llms.txt is the narrow version of a much bigger question.

A thousand Google searches now produce roughly 300 website visits, and a growing share of what replaces the rest arrives through answer engines and agents.

Any advantage you can get is worth keeping.

Slobodan Manić is covering the entire AI visibility & conversions playbook in a signature workshop. 27th August 2026. Made free for you.

Implementation guide: https://t.co/xjk8Ikq6S7

Workshop:
https://t.co/s4EAYGXZ51

Submit your own site in the registration form. Sani will pick a few to audit live.

[Convert.com @Convert]

Day 2 is where it gets interesting.

Five talks, each poking at something you stopped questioning.

Thursday 10 September, kick-off BST / AM EST

· Sabine Harnau

Whose Conversions Are You Optimising For? A Feminist Lens on Experimentation

Every KPI you've ever set carries a quiet assumption about who the average user is. Sabine's argument is that AI is now hardening those assumptions into fact at scale, which means a program optimising for one narrow default gets very good at excluding everybody else. She uses queer and feminist theory to pull those assumptions back out of your test metrics, and it's a more practical exercise than it sounds.

· Himani Kothari

The Most Dangerous Moment in Any Experiment Is When It Starts Working

Everyone has a plan for a losing test. Almost nobody has one for a winner. Himani's point is that a stat sig uplift is a promise the rest of your business then has to keep across demand, capacity, support and fulfilment. She has watched wins scale faster than the systems built to hold them, which is why mature programs ask what breaks when this works before launch rather than after.

· Nikki Elbaz

You Shipped Everything. Why Isn't NRR Moving?

Nikki hosts the Churn Flags podcast, and her case is that expansion revenue isn't hiding in new features or new markets. It's sitting in the product you already have and the customers you already paid a record CAC to acquire. Churn and upgrade decisions get made in-product rather than on AE calls, and she maps nine specific moments where you can tip one your way.

· Music with Josh Silverbauer
The break, with an actual soundtrack.

· Iqbal Ali

From Fragile Frameworks to Antifragile Experimenters

Iqbal's opening move is to note that "don't outsource your thinking to AI" is rich advice from an industry that already outsourced it to acronyms. He has spent serious time training language models and studying how they work, which is what makes him worth listening to on why that isn't thinking. The session is practical techniques for rebuilding deep understanding, on the grounds that it's the one capability staying expensive. Founder and Builder at Ressada.

· Community showcase
Women in leadership, and lessons from overcoming imposter syndrome and feeling invisible.

· Liliann Malcolm

Silent Customer Signals: Rethinking Customer Value

Ask a team who their best customer is and you'll usually get the biggest spender. Liliann thinks that model is dying, and with CAC where it currently sits the blind spot is expensive. Her session covers the behavioural signals that give away real intent (recency, frequency, site visits, email engagement, purchase patterns beyond gross AOV) and how to re-segment the base you already have around them.

Four hours, five talks, and 72-hour replays.

Save your seat.

https://t.co/1MFyJHgStw

[Convert.com @Convert]

"Let's make the pricing page simpler"

is not a hypothesis.

Here's what it takes to become one.

𝟭 - 𝗕𝘂𝗶𝗹𝗱

The Hypothesis Generator won't let you off the hook.

Six fields, all required:

We have observed that… / By… (what you saw, and the evidence)

We wish to… / For the segment… (the change, and who gets it)

This will lead to… / Measured by… (the outcome, and the metric)

Two optional fields most people skip and shouldn't: how long the test runs, and how you'll keep it ethical.

If you can't fill those in, you don't have a hypothesis yet.

Finding that out here is cheaper than finding it out in week three.

𝟮 - 𝗤𝘂𝗲𝘀𝘁𝗶𝗼𝗻

Convert Socrates takes what you built and runs the five whys at it, through questions written specifically for experimentation.

It surfaces the assumptions you didn't know you were making, and the spots where your logic doesn't actually join up. Those questions were developed with Jon Crowder of Another Web Is Possible, who has spent decades doing this to other people's hypotheses.

𝟯 - 𝗜𝗺𝗽𝗿𝗼𝘃𝗲

Convert Chorus puts your restated hypothesis in front of six expert perspectives:

- Alternative approaches and scope
- Risks
- Targeting and placement
- Evidence and rigour
- Boldness and detectability
- What-ifs

It then synthesises all six into a single read and drafts a variant design wireframe off the back of it.

Feed it your page URL, goal metric, current conversion rate, and site type (ecommerce, SaaS, subscription, publishing, lead gen) and the feedback stops being generic.

For context on why any of this matters: across more than 28,000 tests run on Convert Experiences, one in five reached statistical significance.

A tighter hypothesis won't fix all of that.

It will stop you spending three weeks of traffic on something that never held up on paper.

Free, no login.

[Convert.com @Convert]

121 people pitched.

Five made Day 1.

Here's the afternoon they built.

Wednesday 9 September, kick-off 13:30 BST. Free, online.

14:00 · Victoria Bakos

The Claude UX Factory Is Real: Break the Cycle of Sameness in AI Designs

If AI-built pages have started giving you a vague sense of déjà vu, Victoria has named the thing. She works across UX, UI and CRO at Miso, and her session is a practical split: what in your process genuinely benefits from automation, and what has to stay in human hands. Includes how to spot AI-generated design at a glance, and where those flows quietly fail real users.

14:40 · Lauren Wong

The Algorithm Can't Lick a Stamp

Lauren has taken consumer brands from zero to hundreds of millions, including the founding team at HelloFresh Canada plus work with Huggies, Spin Master, Sano Genetics and Tonies. That makes her the right person to argue that everyone's favourite advice, do things that don't scale, is also the advice nobody actually operationalises. Her session treats offline as a performance channel with measurement and tests attached. Strange thing to hear in 2026, which is probably why it's worth hearing.

15:00 · Ester Serra

The Experimentation That Destroys Value: When Optimizing One Metric Breaks Another

Ester's is the talk most likely to change how you write your next test doc. Her framework is for reading a stat sig result and asking what it cost somewhere else: trust, retention, brand credibility. It's the guardrails conversation with an actual structure attached, and a way to run it without leaving room for cherry-picking. Founder at Dataverto.

15:50 · Matt Policastro

Confidently Wrong: What Happens When AI Makes Experimentation (Almost) Free?

There's a live demo of an LLM designing and shipping a test, and Matt is upfront that the demo is the setup rather than the point. His argument is about deliberate friction. When launching a test costs almost nothing, volume stops meaning anything, so the useful question becomes what to slow down and which criteria make a result worth acting on.

16:20 · Community showcase

The art of job applications in 2026, plus Ikigai and monetising purpose.

17:20 · Julie van der Hoop, PhD

Experimental Design Lessons from Marine Biology

Julie runs privacy-first testing at Wikipedia, at a scale most of us will never go near. Before that she was a marine biologist running experiments on wild dolphins, where nobody gets to say "let's just rerun it" and discipline is the only thing between you and no signal at all. Her session is about building real control into conditions you can't control, which is a problem more experimenters have than admit to. Senior PM at the Wikimedia Foundation.

Four hours, five talks, and 72-hour replays if Wednesday afternoon doesn't cooperate.

Day 2 lands tomorrow.

https://t.co/1MFyJHgStw