How to Find Out Why Visitors Leave Your Website

There's one page on your site you check more than the rest. You open the analytics, see how many people landed last week and how many left, and that's where the trail ends.

Counts are the easy half. The number leaves out the spot where it happened — the button that didn't respond, the line of copy that made someone reconsider, the shipping total that showed up at the last screen. That part gets recorded by a different set of tools, and most of them are free.

In this post I'll show you those tools, plus ChatGPT and Claude, that get you from "people are leaving" to one thing to fix this week.

This is part of the "Why They Leave" mini-series. What ties it together: find the exact spot before you change anything.

Turn a Wall of Numbers Into One Fix

You open a dashboard with fifty numbers on it, feel stuck, then stop opening it. Pick a tool that shows a handful.

PostHog's free plan gives you a million events and 5,000 session recordings a month with no credit card. Plausible is simpler still.

Then hand the reading to a model. Screenshot the page where people drop off, paste it into ChatGPT or Claude, and ask:

"Where do people seem to drop off most, and what's the one change I'd test this week to fix it?"

Make that one change. Look again in a few days.

Before you trust any of it, prove your tracking has ever fired. Pageviews are the easy part — a script tag collects those on its own. The events you care about, add to cart and form submitted and checkout completed, only exist if some line of your own code calls them, and that call is the part that gets skipped. Mine was, for months: the attribution tool loaded on every page and logged every pageview, and nothing anywhere ever called its conversion function.

So check it. Search your codebase for the conversion function's name; if the only hit is the file that defines it, your dashboard has been showing you traffic and you've been reading it as performance. On a hosted platform where you can't grep anything, the same check is a test purchase — buy something from yourself, then go look for that order in the report an hour later. If it isn't there, fix the tracking first.

Watch One Person Give Up

In a recording you watch the ten seconds someone spends jabbing a button that won't respond. Microsoft Clarity does this free with no traffic limit. Hotjar does the same thing.

1. Add Clarity to your site.
2. Pick your weakest page and watch five sessions back to back.
3. Write down the first thing that made you wince.

Fix it, then watch five more.

Put the watching on the calendar before you install anything. Recordings pile up whether or not anyone opens them, and a free-forever tool nobody has opened since March is telling you nothing in July. Twenty minutes on a Friday covers it. On a paid trial, book the sessions inside the trial window or the trial expires with the recordings unwatched.

One thing recordings won't tell you is that the page is slow. A stall looks like a person sitting still, and you can't see their connection from a replay. Run that page through Lighthouse in Chrome DevTools for the load numbers, and keep the recordings for what people did after it loaded.

Safety note: these tools can capture what people type. Blur or skip any clip with real personal information.

Read Your Site's Search Bar

Someone typing in your search box has told you what they came for, in their own words. Shopify keeps that report; most platforms do.

1. Pull last month's search terms.
2. Paste them into ChatGPT or Claude, ask for themes, and ask which searches came back empty.
3. Then decide per term: rename it, or admit you don't carry it.

An empty result page is a person who came in ready to buy — search visitors convert at about 4.63% against 2.77% for everyone else, per Econsultancy figures shared by Algolia, roughly 1.8 times higher.

So pull your own empty-search list before you change anything else on the page — it's sitting in that Shopify report already.

See Why a Competitor Outsells You

When a competitor converts better than you, the reason is usually nameable with both pages side by side.

1. Screenshot their product or landing page. ChatGPT and Claude both read uploaded images.
2. Ask what makes the page convincing and what yours is missing.
3. Check Perplexity for how real customers talk about them, with sources you can open.

Skip the things you couldn't copy anyway — their ad budget, their eleven thousand reviews. Look at what the page decides for the visitor: what the first screen promises, where the price sits, which objection gets answered before anyone has to ask it, how many clicks stand between wanting the thing and paying for it. Those are the parts you can match this week. None of it needs a paid research tool.

Then take what works and write it your way.

Safety note: learn from their page and write your own copy. Republishing theirs gets you in trouble.

Find the Leak at Checkout

The worst leak I've found on my own store — custom pet paint-by-numbers kits — was a missing "the." The kit configurator sent every finished cart to dailypawblog.store; the storefront is thedailypawblog.store, so every "Add kit to my cart" click landed on a parked domain with ERR_CONNECTION_TIMED_OUT. One curl -sI against the real URL surfaced it.

Yours is probably something that ordinary too. Baymard Institute's read of 50 studies puts cart abandonment around 70%, and the top reasons are all fixable: surprise shipping and tax (39%), a forced account (19%), not trusting the site with a card (19%), a long clunky checkout (18%).

Take those four in order on your own checkout. Is the shipping and tax total visible before the last screen. Can somebody buy without creating an account. Does the payment step still look like your site, or does it hand people off to a page they've never seen. And count the fields — every one between the cart and the pay button is another place to stop.

1. Screenshot your funnel — product page, cart, checkout, purchase, with counts.
2. Hand it to a reasoning model and ask where the biggest drop is and the one thing to test first.
3. Fix that one, then check again.

Then go click your own checkout on your phone tonight, all the way to the payment screen. No dashboard surfaces a missing "the" — a cart that never reaches the store never registers as a lost checkout either.

Quick Recap

• Screenshot the drop-off page and ask ChatGPT or Claude one question. That beats fifty numbers you never open.
• Confirm your tracking has ever fired before you read anything into it. Mine hadn't, for months.
• Watch five sessions on your weakest page. Fix the first thing that makes you wince.
• Your site search report lists people who told you what they wanted. Pull last month's.
• Click your own checkout on your phone tonight. My worst leak was a missing "the" in a domain.

Start Here

You'll want to settle which leak to chase first before you start rewriting pages.

At daisyguti.ai/work-with-me, there's an AI intake assessment that maps where your business stands before any call. It takes a few minutes and gives you a clear read on what to fix first. Daisy is a 20+ year engineer who builds these systems for small business owners, so the assessment reflects how real sites run.

Sources

1. PostHog pricing, free plan with 1M events and 5K session recordings - https://posthog.com/pricing
2. Microsoft Clarity, free session recordings and heatmaps ("free forever") - https://clarity.microsoft.com/
3. Algolia, ecommerce search statistics (Econsultancy: search visitors convert at 4.63% vs 2.77%) - https://www.algolia.com/blog/ecommerce/e-commerce-search-and-kpis-statistics
4. OpenAI vision docs, ChatGPT can read images - https://developers.openai.com/api/docs/guides/images-vision
5. Anthropic Claude vision docs - https://platform.claude.com/docs/en/build-with-claude/vision
6. Baymard Institute, cart abandonment rate statistics - https://baymard.com/lists/cart-abandonment-rate