Session replay vs AI visitors: Cold Visitor, Hotjar and Microsoft Clarity

The short answer

• Choose session replay (Hotjar, Microsoft Clarity and similar tools) when you have steady traffic and want to see how real visitors behave at scale.
• Choose AI visitors (Cold Visitor) when you do not have traffic yet, when you want to test a page before it goes live, or when you want each problem explained and timestamped on video without watching hours of recordings.
• Use both once you are live: AI visitors to find and re-test obvious friction, session replay to confirm what real people do.

A note on fairness: Hotjar and Clarity are established products with more features than we describe here. We stick to the basics that define the category and that we are sure of. Check their own sites for current features and pricing.

What session replay tools do

You add a tracking script to your site. From then on, the tool records how visitors interact with your pages - clicks, scrolling, mouse movement, page changes - and lets you replay individual sessions or look at heatmaps that aggregate many visits. Microsoft Clarity is free to use; Hotjar offers a range of plans.

The strength is that these are your real visitors, with real intent and real devices. Patterns across hundreds of sessions - everyone clicking a non-clickable image, nobody scrolling past the second section - are strong evidence.

The limits follow from the same design:

• It needs traffic. With a handful of visits a day, a heatmap says little, and before launch there is nothing to record.
• It shows what, not why. You see a visitor hesitate and leave; you have to guess the reason.
• It takes time to review. Finding the telling session among many is work.
• It runs in your visitors' browsers. In the EU in particular, that usually means thinking about consent and your privacy notice.

What Cold Visitor's AI visitors do

Nothing is installed on your site. You give us a URL and, optionally, a goal ("reach the order confirmation"). AI visitors with different personas - an impatient buyer, a skeptical evaluator, a careful senior on slow 4G, a keyboard-only visitor and others - open the site in a real browser, read it, click, scroll and fill forms with clearly fake test data. Each has a patience budget; friction uses it up, and when it runs out the visitor leaves.

Every step is recorded: screenshots, load times, console errors, a full video, and the visitor's own reasoning ("I clicked 'See prices' twice and still haven't seen how much it costs"). The report scores the friction, shows the funnel per persona, and links each finding to the exact second on video. Visitors stop before any irreversible step such as a payment. You can open a full sample report from our fictional Northpeak store to see the output.

The limits are just as real. These are synthetic visitors: they have no real intent or money at stake, they can be more or less patient than your customers, and their findings are directional. They find mechanical and communication problems well; they cannot tell you whether the market wants your product, and they do not predict conversion rates.

Side by side

Using both: a simple workflow

1. Before launch: run AI visitors against the landing page and main journey. Fix the dead buttons, hidden prices and slow pages they find. Our guide to why visitors leave lists what to look for.
2. At launch: install session replay so real data starts to accumulate.
3. Each time you change a key page: re-test with AI visitors before or right after shipping, so you see whether the change helped without waiting for weeks of traffic.
4. Every few weeks: watch real sessions for patterns that synthetic visitors would never produce - real confusion about your offer, unexpected use cases, specific devices.

If you are comparing all the ways to test without recruiting people, the guide on usability testing without participants also covers hallway tests, five-second tests and paid panels.

Frequently asked questions