Yes, a little. A chat widget adds weight and code to your page, but nearly all of them load after your main content, so the part your customer sees first is usually not held up. The cost shows up later, in how fast the page answers a tap.
What the measurements actually show
Chat scripts are not unusual. The Third Parties chapter of the 2025 Web Almanac from HTTP Archive, published January 15, 2026, found that around 90% to 92% of pages use at least one third party. It files chat and contact scripts under a customer success group and says those scripts are generally heavier in weight.
DebugBear tested 21 chat widgets and printed the results. Each widget went on an empty test page, and the page was then run through Google’s Lighthouse tool. That article was first published July 14, 2019 and last updated November 10, 2025.
Almost none of the widgets held up the first paint, since they load below the closing body tag. The one exception in that test was FreshChat, which asks for a script in the head of the page, and it was the only widget whose test page was slow to show anything at all.
The sizes ranged wide. Tawk.to pulled about 749KB. FreshChat came in at 564KB and Zendesk at 533KB. At the light end, Zoho Desk and MyLiveChat both measured 67KB, and LiveAgent 79KB.
That is an elevenfold gap between the lightest widget and the heaviest one. Same job. Same little bubble in the corner of the screen.
DebugBear’s own summary is fair about it: “Chat widgets impact page performance, requiring extra network bandwidth and increasing CPU usage. The good news is that the immediate user experience of your primary content isn’t normally affected.”
Where the real cost lands
Download size is the easy number to grab. Main thread time is the one that bites.
Google’s Chrome for Developers documentation has a page called “Reduce the impact of third-party code.” It says the audit fails when third-party code blocks the main thread for 250 ms or longer. It defines a third-party script as “any script hosted on a domain that’s different to the domain of the URL that you audited with Lighthouse.” That page was last updated May 2, 2019.
Your chat widget is exactly that. It is hosted on the vendor’s domain, not yours.
DebugBear measured how long each widget’s code took to run. Smartsupp came in at 1000ms. Zendesk at 991ms. The quickest of the group, Zoho Desk, still took 259ms.
Run time and blocking time are not the same measure. Lighthouse counts only the stretches where the main thread is stuck, so a widget can run for a second in small pieces and stay under the line. Read these run times as the size of the job being handed to the phone, not as a pass or a fail.
Main thread time is why a page can look finished and still feel stuck under your thumb. That is a separate measure with its own article: why a loaded page still feels slow.
The widget nobody sees for nine seconds
DebugBear also timed how long it took for each widget to actually appear on screen. LiveAgent showed up at 2.5 seconds. FreshChat at 2.8 seconds. Kayako took 9.4 seconds. Zendesk took 9.3.
Sit with that one for a second. A homeowner lands on your emergency page at 11pm with water coming through a ceiling. If the chat bubble does not exist for nine seconds, most people have already scrolled, or called someone, or left.
So on the slow end the widget is charging your page weight and CPU time the whole while, and it is still not on screen when the person decides what to do.
The limits of these numbers
Here is the honest part, and it cuts against using any of the figures above as gospel.
- The test page was empty. Yours is not. You have jobsite photos, a font, a booking form, maybe a map. A widget’s cost stacks on top of all that, so your real numbers will be worse than the lab numbers, not better.
- The Google page is old. “Reduce the impact of third-party code” carries a last updated date of May 2, 2019, and the page itself notes the audit has since moved into another insight in Lighthouse 13. The 250 ms figure is what that page still prints. The tooling around it has moved on.
- Vendor numbers and independent numbers do not agree. LiveChat published a post on July 20, 2021 saying its widget went from 412KB in June 2020 down to 256KB. DebugBear measured LiveChat at 385KB. Those two are not counting the same thing. That is the whole point. Do not take any vendor’s figure about its own script and skip the test on your own page.
- Widget versions change. A number measured in 2025 is not a promise about the script your site loads today.
How to check yours in ten minutes
You do not need to trust a list. Measure the page you actually own.
- Open your site in Chrome, press F12, and go to the Network tab. Reload the page.
- Sort by size. Find the requests coming from your chat vendor’s domain. Add them up.
- Run the same page through Google’s PageSpeed Insights and find the part that lists scripts loaded from other domains. Your chat vendor should be named there, with the time its code spent working.
- Now open your chat inbox. Count how many real, booked jobs came out of it in the last 90 days.
- Put the two next to each other, along with what an AI chat agent costs you per conversation.
That last step is the one people skip. A widget that costs 400KB and books eight jobs a quarter is earning its keep. Leave it alone.
What to do next, and when to do nothing
If chat brings you work, keep it, and make sure it knows what to do in a real emergency. Do not rip out a widget that books jobs in order to win points on a speed test. A quick page that gives a panicked customer no way to reach you is not a better page.
If chat brings you nothing and the script weighs half a megabyte, remove it and put a tappable phone number where the thumb already is. That is free and it takes an afternoon.
If you are changing what loads on the page, order matters more than effort. Weight and main thread work are only two of the levers, and the changes that move a page the most are not always the ones that feel biggest.
One more fence, and it points back at this site. An AI assistant on your site is a third-party script too. It has a download size and it runs code on the main thread, the same as any chat widget in that DebugBear list. Anyone who tells you their chat costs your page nothing is selling, and the same goes for anyone pitching the widget as a way to get found by AI. Ask for the number, run the test yourself, and hold their script to the same line Lighthouse draws at 250 ms.
The only page this site can vouch for is its own, which scored 100 on Google’s PageSpeed test, desktop and mobile, when it launched in August 2026. There are no client case studies here to point at, and none are claimed.