Most website owners can tell you how many people visited last month. Very few can tell you where those people stopped. Was it the pricing section, the contact form, the checkout, or the moment the page took four seconds to load on a phone? Without that answer, every redesign is a guess.
The useful change in 2026 is that you no longer need to be an analyst to find out. You can put your own site data in front of an AI assistant, ask it where visitors drop off and why that might be, and get a short list of things to test. Here’s how to do it properly, what the free tools show, and where AI’s answers need a human check.
Key takeaways
- The question to ask is specific: at which step do people leave, on which device, and from which source. “Why is my site not converting?” is too vague to answer.
- Two free sources give you the raw material. Google Analytics 4 funnel explorations show where users drop between steps, and Microsoft Clarity, which describes itself as “free forever” with no traffic limits, shows what people did on the page.
- AI is best used as a first-pass analyst. It spots patterns and proposes explanations, but Clarity’s own documentation calls its AI insights a starting point, not a replacement for watching real sessions.
- Benchmarks help you judge the cause. Baymard’s research across 50 studies puts average cart abandonment at 70.22%, with extra costs the top reason at 40%.
- Fix one thing at a time and measure again. If you change five things at once, you won’t know which one worked.
Start With the Right Question
“Where do my customers stop?” has three parts, and each one changes the fix:
- Which step. The page, section or form field where people leave.
- Which segment. Mobile or desktop, new or returning, paid or organic.
- What they did first. Did they scroll, click something that didn’t respond, or leave within seconds?
A drop-off on the contact page for mobile visitors from search is a different problem from a drop-off at checkout for returning customers. AI can only help you separate them if you give it data split that way.
Step 1: Get the Data Out of GA4
Google Analytics 4 has a built-in report for exactly this: the funnel exploration. In Explore, you choose the Funnel exploration template, define your steps by selecting events, and the report shows abandonments, abandonment rate and completion rate between each step. You can add a breakdown such as device category, and export the result as a Sheets file, CSV or PDF. (Source: UK government GA4 guidance, funnel explorations)
Google’s own example for a lead-generation form uses four steps: a visit to the home page, a view of the form page, a form start (form_start) and a form submit (form_submit). Comparing user counts between consecutive steps shows the percentage lost at each stage. New events can take up to 24 hours to appear in explorations, so don’t panic if data is missing on day one. (Source: Google Analytics Help, reporting on a lead generation form)
Two settings matter. An open funnel lets users enter at any step, while a closed funnel only counts people who start at step one. Use closed when you want a clean “of everyone who began, how many finished” number. And the funnel is only as good as your events: if your form doesn’t fire a start and submit event, there’s nothing to measure. On a custom-built site, that’s a few lines of JavaScript pushing events, and it’s worth doing before any analysis.
Step 2: Add the “What Did They Do” Layer With Clarity
A funnel tells you where people leave. It can’t tell you what they were doing when they left. That’s the job of a behavior tool like Microsoft Clarity, which describes itself as “free forever” with “no limits on traffic.” It records sessions, shows click, tap and scroll heatmaps, and includes AI summaries and an AI chat that lets you ask questions about your data. (Source: Microsoft Clarity. This is marketing copy, so check the terms yourself.)
The behaviors worth looking for are the frustration signals: rage clicks (repeated rapid clicks in one spot), dead clicks (clicks that do nothing), excessive scrolling and quick backs. Scroll heatmaps also show the percentage of visitors who reach each point on a page, so you can see whether anyone ever gets to your call to action. (Source: Toptal on Microsoft Clarity. The author says they worked on the Clarity team, so treat the praise accordingly.)
One firm caution from the same source: it’s not advisable to use Clarity on sites with sensitive data such as medical records or financial information. Clarity has masking modes, with “Balanced,” which masks sensitive text, as the default, but if your site handles that kind of information, choose a different approach.
Step 3: Ask the AI the Right Questions
You have two routes. Use Clarity’s built-in AI chat and summaries on the behavior data, or export aggregated numbers from GA4 and paste them into a general AI assistant. For the second route, send only aggregated figures such as step names and user counts, never names, emails or anything personal.
Prompts that tend to work (the percentages below are made-up examples, so swap in your own):
- “Here are my funnel steps with user counts and abandonment rates, split by mobile and desktop. Which step has the biggest drop, and is it unusual compared with the other step?”
- “On my pricing page, 62% of mobile visitors never scroll past the first section. What are five possible reasons, ranked by how easy they are to test?”
- “Summarize the most common frustration patterns across these sessions, and tell me which page they happen on most.”
- “My form starts-to-submits rate is 31%. Based on typical form-abandonment causes, which fields or design choices should I check first?”
- “Write a one-week test plan for the biggest drop-off, with a single change, the metric to watch, and what result would count as a win.”
Notice what these do. They give the AI real numbers, ask for ranked hypotheses rather than a verdict, and finish by asking for a test. That keeps you in charge of the decision.
Step 4: Check the AI’s Answer Against Reality
An AI assistant will give a confident, plausible explanation for almost any drop-off, whether or not it’s the true one. That’s why Clarity’s own documentation, as summarized by Toptal, describes AI insights as “most useful as a starting point for investigation rather than a replacement for direct observation.” (Source: Toptal)
So treat every explanation as a hypothesis, and confirm it before changing anything:
- Watch five to ten recordings of sessions that dropped at that step.
- Test the page yourself on a real phone, on mobile data, not just in a desktop browser.
- Check whether the drop is concentrated in one device, browser or traffic source. A problem that only affects one segment points to a bug or a mismatch, not a weak offer.
- Make sure the drop isn’t just a tracking gap, such as a missing event.
Also be honest about sample size. If a page gets 40 visits a month, a 10-point swing is noise. Wait for more data before acting on small numbers.
What Drop-Offs Usually Mean
Once you know the step, benchmarks help you guess the cause. For online stores, Baymard Institute’s research across 50 studies puts the average documented cart abandonment rate at 70.22%, and lists the top reasons shoppers give (they could pick more than one): (Source: Baymard Institute)
| Reason for abandoning | Share of shoppers | What to check |
|---|---|---|
| Extra costs too high | 40% | Show shipping, tax and fees early |
| Delivery too slow | 20% | Show delivery estimates before checkout |
| Didn’t trust the site with card details | 19% | Trust signals, recognizable payment options |
| Forced to create an account | 18% | Offer guest checkout |
| Checkout too long or complicated | 17% | Cut fields and steps |
| Site had errors or crashed | 17% | Test on real devices, check logs |
| Couldn’t see total cost up front | 12% | Show a running total |
Those are store figures. For a service business, the equivalent leaks are usually less dramatic and more basic: a contact form with too many fields, no phone number visible on mobile, a page that loads slowly, or a call to action buried below the fold. I covered the practical basics in what a small business website really needs to make the phone ring, and the structure side in a 7-section, SEO-ready service site built in one month.
Step 5: Change One Thing, Then Measure
The biggest mistake after finding a drop-off is fixing everything at once. Pick the single highest-impact, lowest-effort change. Write down the current number, make the change, wait long enough to collect meaningful data, then compare the same funnel step before and after. Google’s own lead-form example does exactly this: it compares the period after a form change with an earlier one to confirm fewer users dropped at each step. (Source: Google Analytics Help)
This also keeps you away from the trap I wrote about in stop adding tools, and what to prioritise instead: more tools and more changes don’t help if you can’t tell what moved the number.
A One-Week Plan
| Day | Task |
|---|---|
| 1 | Make sure key events (form start, form submit, checkout steps) are tracked. Install Clarity. |
| 2 to 5 | Let data collect. Build a GA4 funnel exploration for your main goal. |
| 6 | Export the funnel by device, ask AI for ranked hypotheses, and watch 5 to 10 matching recordings. |
| 7 | Pick one fix, write down the current numbers, and ship the change. |
A Note on the Numbers
The Baymard figures are survey-based averages across many stores, so your own rate will differ. The Clarity feature descriptions come from the company’s marketing page and a review by someone connected to its team, so confirm features and privacy terms on Clarity’s own documentation. And AI-generated explanations are guesses until you test them. The method here is sound, but treat each output as a lead to check, not a finding.
Frequently Asked Questions
Can AI really tell me where customers leave my website?
It can help you find and interpret the pattern, but the data comes from your analytics and behavior tools. AI is good at spotting the biggest drop and suggesting reasons, as long as you give it real numbers and then verify its ideas.
What free tools do I need?
Google Analytics 4 for funnel explorations, and a behavior tool such as Microsoft Clarity for recordings and heatmaps. Both can be used at no cost, though you should check current terms before relying on them.
Is it safe to paste my analytics into an AI assistant?
Paste only aggregated numbers such as step names and counts, and leave out anything personal. If your site handles medical or financial information, avoid session-recording tools altogether, as Clarity’s own guidance cautions.
What is a funnel exploration in GA4?
It’s a report that shows the steps users take toward a goal and where they drop off between them, with abandonment and completion rates. You can break it down by device or source and export it.
How much traffic do I need before this is useful?
Enough that percentages aren’t swinging on a handful of visits. With very low traffic, rely more on watching individual recordings and testing the page yourself, and be cautious about reading too much into small changes in the numbers.
The Short Version
To find where customers stop, track the key steps in GA4, add a behavior tool like Clarity, and put the aggregated numbers in front of an AI assistant with specific questions. Ask for ranked hypotheses, verify them by watching real sessions and testing on a real phone, then change one thing and measure again. The AI speeds up the analysis, but the decision about what to fix still needs a person who has looked at the page.
Want a second pair of eyes on where your site loses people? I’ll walk through your funnel with you and point out the first fix worth making. Book a call.
About the author: Rolijon Clemenia is a web designer and WordPress developer based in Iligan City, Philippines, with 5+ years building conversion-focused websites, landing pages and funnels for startups, businesses and marketing agencies across the US, UK and Australia. Read more about Rolijon.