How Convertlyft works: from the tag to what to do

After sign-up a three-step setup opens: Your site, Add the tag, First visit. Every step can be skipped. It asks for your site address first, looks at the homepage to see what the site is built with (Shopify, WordPress, Wix, Squarespace, Webflow, Framer, Next.js, React, or a tag manager already on the page), and leads with the instructions for that platform. There are two ways to add the line: paste it into your site’s head yourself, or paste a paragraph into Lovable, Bolt, v0 or whatever built the site, which says exactly what to add and to change nothing else. On WordPress you can install the Convertlyft plugin from WordPress.org instead: it adds the tag, and its one setting is your site key. There is nothing to press at the end: First visit ticks by itself when the first page view from your site reaches us.

Install the Convertlyft app from the Shopify App Store, open it from your Shopify admin and switch Convertlyft on under App embeds in your theme. The app connects your checkout itself, so there is nothing to paste. Shopify bills your plan. See Plugins for what it records and what it asks for.

Or paste it yourself: on a Shopify store the line goes in the theme, and a second script goes in Shopify admin under Settings → “Customer events” as a custom pixel. Shopify’s checkout and thank-you pages do not run your theme’s code, so the first line cannot see a sale; the second paste is what counts purchases. Settings › Install shows both, with your key already filled in, when it recognises a Shopify store. If someone buys without having browsed the store in the same visit, we count the sale but cannot tie it to a visit.

Live, opened from Home’s “on site now”, is the human check: the people on your site now, one row per visit (up to 50), refreshed every 10 seconds; open a visit to see it in place. Home opens on Needs you, the top three to-dos, each one click from What to do, then one card of visits, visits from paid ads, and sales or enquiries, then your pages and where people are. A visit counts as a sale only when the purchase you mapped actually reaches us.

Traffic adds how many visits bought or got in touch by entry channel, campaign tags, and “Search (Google)” from Search Console (finalised about two days late), plus scroll depth and a list of what people did. Recordings lists every visit newest first, up to 200 shown, and each one has a recording to watch, up to the daily safety limit, unless you turned recording down in Settings; Traffic › Pages lists your 200 busiest pages by views, each with how many of the visits that started there bought or got in touch and how many views were the last page of the visit, and above them the 5 pages losing the most visits against pages like them; open a page for the 3 pages most often before and after it. Nothing is sampled; the lists are capped, and the pages say so on screen.

On Funnels a funnel is up to 6 steps, each a page or an action with an optional “Page contains” filter; people are counted once per step and only if they did the earlier steps first. The one step that lost the most people prints in amber, in words: “Most people were lost between “A” and “B”: X of Y did not go on.” Each step links to the recorded visits that reached it; for a page step that means visits to that exact address, so the list can be narrower than the step. You can compare with the period before, or export the steps as CSV. Suggested funnels carry their own caveat on screen: we picked the steps, nobody has confirmed them. More on funnel analysis.

Replays and heatmaps are evidence: you open them from a to-do, a funnel step, an error, a stuck-spots row or a page that already names what to look at. Heatmaps also have a tab of their own, Recordings › Heatmaps. Recordings show the page as the visitor saw it: its words and prices readable, every typed value hidden, and personal details such as email addresses and phone numbers masked. More on visit recordings. Any visit that hits a JavaScript error, rage-clicks or dead-clicks, gives up on a form, or reaches a key page such as checkout, cart, contact or sign-up is kept. Recordings are kept 90 days. The list opens on the ones you have not watched, ordered by where people got stuck most, and What to watch picks the day’s three recordings where people got stuck most, each with a line on why.

Heatmaps rebuild the page from a kept recording and draw each click on the element it landed on. There is a map for all clicks, rage, dead, error and first clicks, and a scroll map; movement is a desktop view. Elements are ranked, with how many of the people who clicked each one went on to buy or get in touch once there are enough clicks to say, and a link to watch those visits. You can compare two maps side by side. A page with no kept recording yet falls back to a photo our crawler takes; “Photograph this page” queues one it missed.

Site health › Errors groups JavaScript errors into issues, each with how often it happened, the visits and the people, split into your site’s code versus external issues. You can mark an error not a problem (and bring it back), separate reports we joined, copy it for your developer, or fix it with your AI: the AI app you connect to Convertlyft reads the write-up and proposes the change. Site health rates how fast it loads for most visitors (LCP), how quickly it reacts to a tap (INP) and how much the page jumps (CLS), with FCP and TTFB alongside, for the typical visitor (p75) against Google’s thresholds, and lists the slowest pages. Site health › Stuck spots lists clicks where nothing happened and clicks made again and again (rage clicks) on each page, against that page’s views, and each row links to the recorded visits behind it.

The daily job runs every day at 06:00 UTC over the previous complete UTC day. The Monday report goes out on Monday at 07:30 in your site’s time zone; the daily email is off until you turn it on. Settings › Email updates has one switch for each email, just for you, and keeps the record: one card per day, one row per to-do with one link, and a “Watched” line of visits, pageviews, pages seen, funnel steps checked and new issues. A quiet day reads “Nothing leaked above noise level.” Each to-do compares a day to its own trailing 7-day normal — it carries no change marker, so it never attributes a move to a deploy. Alerts you set under Settings › Alerts are looked at once a day and arrive by email the morning after one goes off.

What to do is fed by a run every Monday over the last 7 complete days. The header names the week, when it was computed, and how many to-dos were held back as too small to trust. It is one to-do list with four tabs: To do, Good news, Fixed and Hidden. Each row carries a tag (Fix, Grow or Good news), the to-do with its counts, and one button, Show me. An opened to-do says what is wrong, where, and the change to make, with a button to the evidence itself (the issue, the error, the visits) and your marks: Later, Not a problem and I fixed it, which reads Fixed — checking until the next check. When an AI app is connected to Convertlyft, it also offers a prompt to copy for that assistant. Confirmed fixed and Came back are set by the next check, never by a button. When the to-do is about a number, Worked follows: 28 days after the fix, the weekly run reads that number again and compares it with the same days before the fix, and says the rate rose or fell, from what to what, with the counts: Worked, Got worse or No clear change yet, or “Too few visits to tell yet” when either side had too few visits or results to judge. Home shows the top 3. Add a to-do puts one of your own on the same list. There is no approve button and no ship step on this path. Worked compares before and after; it is not a controlled test, so other changes in those weeks move the number too. It does every day the reading a consultant does once; an alternative to hiring a consultant for more sales and enquiries says where that fits and where it does not, and CRO tools, by the job they do says which job to start with.