Methodology
How every number on this site is produced
Including the thirteen things we published wrong, what they said, and what is true now. If a figure here cannot be checked by somebody who does not trust us, it should not be here.
Last checked 1 September 2026
The short answer
Can you check our numbers yourself?
Yes, and that is the point of this page. Our own widget size is measured by gzipping the files we serve, and the command is printed below. Competitor weights come from DebugBear rather than from us. Every price is read off the vendor's own pricing page and carries the date it was read.
Where a figure cannot be verified it is left out. 5 of the 12 vendors we cover have no published weight, and those cells say so.
Measuring our own widget
The widget loads in 3 requests. A tiny loader that decides whether to run at all, a core bundle, and exactly one theme chunk, because only the theme you selected is downloaded. That last part is why the total is a range rather than a single number, and why any single number you see quoted somewhere else is either the best case or the worst.
| Request | Bytes | Note |
|---|---|---|
| w.js (loader) | 1,339 | Decides whether to load anything at all |
| index.mjs (core) | 32,193 | The chat itself, sent on every page that shows the widget |
| Theme chunk, smallest | 5,298 | The "ui" theme |
| Theme chunk, median | 6,606 | The "glass" theme, which is what most people pick |
| Theme chunk, largest | 12,400 | The "singularity" theme |
| Typical total | ~39KB | Range across all 33 themes: 38KB to 45KB |
Measured 1 September 2026 against the artifacts served from production, not against a development build.
The command, if you want to run it against the files we are serving right now:
curl -s https://tglivechat.com/index.mjs | gzip -9 -c | wc -cTwo things we do not do. We do not quote the uncompressed size, because nothing is served uncompressed and the bigger number would flatter nobody. And we do not exclude the theme chunk from the total, which would let us print a smaller figure than a visitor actually downloads.
Where competitor figures come from
Every competitor weight on this site is DebugBear's measurement, attributed to them, linked to the article, with the publication and update dates shown. We do not measure other companies' widgets and we would not expect you to believe us if we did. A vendor benchmarking a competitor gets to choose the configuration, and that choice is the whole result.
7 of the 12 vendors we write about appear in that test. For the rest there is no number here at all. That leaves gaps in the tables, and the gaps are correct.
How prices are collected
Read off the vendor's own pricing page, in a normal browser, not from a review site, a comparison directory or an old screenshot. Each figure is stored with the date it was read and the page it came from, and both are shown wherever the figure appears.
Where a vendor bills more than one way, we say so rather than picking the flattering version. Intercom is a seat price plus a per resolution AI charge, so both are quoted. Tawk.to charges branding removal per website rather than per account, which is the detail that changes the answer for an agency, so that is quoted too.
Corrections
Thirteen things were published on this site that were not right. They are listed here rather than quietly edited, because a corrections log that never has anything in it is not a corrections log. Nine of the thirteen were found by internal audits on 1 September 2026 and are marked as such. Nobody reported any of them, which is rather the point of running an audit: the errors you hear about are never the whole set.
| Claim | What we said | What is true |
|---|---|---|
| Widget size | Stated as roughly 10KB, roughly 15KB and 14.8KB on different pages | About 39KB with the median theme, 38KB to 45KB across all of them. Every page now reads the figure from one file. |
| Theme count | Stated as 20, 25 and 32 in different places | 33. Four more theme files exist in the repository but are not registered, so nobody can select them, so they do not count. |
| WordPress plugin | Written in a way that implied a listing on WordPress.org | The plugin is real and works. It is distributed by us and is not in the WordPress.org directory, and the pages now say that. |
| Founding year in structured data | foundingDate 2024 | Corrected to the real year. Nobody reads structured data with their eyes, which is exactly why it is worth getting right. |
| Theme count on the themes page (audit, 1 Sep 2026) | The page title said "20 Free Chat Widget Themes" and the description said "and 18 more" | 33. The title now reads the count from the same file every other page reads it from, so it cannot drift again. |
| Theme count on two comparison pages (audit, 1 Sep 2026) | Both said "12 widget themes" in the body while their own feature tags said 32, and both named a theme nobody can actually select | 33, read from the shared file, with only registered themes named. |
| Our own prices in machine readable files (audit, 1 Sep 2026) | llms.txt told AI assistants that Pro was $12 a month, and llms-full.txt said the lifetime licence was $149 | Pro is $19 and the Founder licence is $299. These are the files assistants read, so a stale price there is a wrong answer given to somebody who trusted it. |
| Our own prices inside two articles (audit, 1 Sep 2026) | The lifetime deal article said "we sell a lifetime tier at 149 dollars" and computed break even against a $12 Pro plan | $299 and $19, with the break even arithmetic recomputed. The article now says the price changed and points here. |
| Industry statistics with no source (audit, 1 Sep 2026) | Six articles carried more than a hundred percentage claims with no citation, including "studies show that over 40%" and figures attributed to named reports we could not find | Removed or rewritten. Where a real source exists it is cited and linked. Where it does not, the argument stayed and the number went. |
| Pricing table on the Telegram chat widget page (audit, 1 Sep 2026) | Showed Pro at $12, Agency at $29 and a lifetime tier at $149, on one of the highest intent pages on the site | $19, $59 and $299, read from the shared pricing file. This one is the worst of the set: it was a visible price table, not a number in a footnote. |
| Agency price on the agencies page (audit, 1 Sep 2026) | The headline card said $29 a month while the FAQ on the same page said $59, and the per client figure of $2.90 was derived from the old price | $59 a month across ten client workspaces, which is $5.90 per client site. Both figures now come from the pricing file. |
| A theme our own themes page never showed (audit, 1 Sep 2026) | The widget registry has 33 selectable themes and every other page said 33, but the themes page listed 31 and its counter agreed with itself | All 33 are listed. The missing one was Insights, a light analytics style theme that customers could select and could not preview. |
| Widget core size in the speed benchmark article (audit, 1 Sep 2026) | Said our widget core sits around 16KB gzipped | The core is 32,193 bytes gzipped, and a typical install totals about 39KB. Being wrong about your own size in an article about widget sizes is a special kind of careless. |
| Our own retention figures (audit, 1 Sep 2026) | An article presented a 14 / 38 / 68 percent retention comparison as an experiment across three products, with no sample size and no controls | Labelled as an uncontrolled internal observation, with the self selection problem stated, and explicitly marked as not citable as research. |
The rules the dataset enforces
These are not aspirations, they are comments at the top of the file that holds every competitor figure, and they are the reason the tables have empty cells in them.
Every number carries a source and a date, and a figure with neither does not go in. Weight figures are DebugBear's and stay attributed to them. When a price changes, the checked date changes with it. If something cannot be verified it is left null, and every renderer handles null by omitting the row rather than printing a guess.
Our conflict of interest
We sell a live chat widget and we publish comparisons of live chat widgets. That is a conflict and it does not go away by being disclosed. What we do about it is structural rather than promised: every comparison page carries a section naming when you should stay with the other product, written before the section arguing you should not.
Six of the twelve vendors we cover have a reason to stay that we cannot answer, because they do something this product does not attempt. Those sections are on the pages. There are no affiliate links on this site and no vendor pays us anything.
The bottom line
Check us. The widget measurement is one curl command, the competitor weights link to DebugBear, and every price links to the page it was read from with the date attached.
If a figure here is wrong, mail [email protected] and it gets fixed and added to the corrections table above. That table is meant to grow.
Frequently asked questions
How do you measure the TGLiveChat widget size?
Where do the competitor widget sizes come from?
Why not measure competitors yourselves?
How often do you check prices?
Have you published anything wrong?
Do you take money from any vendor you write about?
Who writes this and who is responsible for it?
What do you do when you cannot verify something?
Sources and references
- DebugBear: How Do Different Chat Widgets Impact Site Performance? Published 14 July 2019, updated 10 November 2025, checked 2026-09-01. Every competitor weight figure on this site.
- web.dev: Largest Contentful Paint The metric widget weight actually moves, and the thresholds we quote against it.
- Chrome for Developers: Total Blocking Time Why a second of third party JavaScript execution matters on a mid range phone.
Published by Outline Technologies LLC, 30 N Gould St, Ste R, Sheridan, WY 82801, US.