HLTV Launches a Closed Beta of Its Own AI Stats Assistant

HLTV opened a closed beta of its first in-house AI assistant on October 13, inviting a limited group of readers to try a tool the Counter-Strike news and statistics site calls HLTV Brain. The chatbot works only from HLTV's own database of players, teams, matches, news, rankings and historical ratings, and the site says the trial is planned to run for about two weeks before more accounts are let in gradually.

The rollout follows a smaller change HLTV made three days earlier. On October 10, the site began generating a short automatic brief on every team's profile page, summarizing a club's latest results, upcoming schedule and standing in the current Major qualification race. HLTV Brain is a bigger step: instead of a fixed summary, a reader types an open question, and the assistant decides which of HLTV's own lookup tools to run before writing an answer in plain English, often with a chart or table attached.

From HLTV's October 13 announcement and its replies to readers in the comments.

Where this sits in HLTV's own history

HLTV was founded in 2002 by Martin Rosenbæk and Per Lambæk as a site for hosting recordings of early Counter-Strike matches, and it has spent the years since building the record HLTV Brain now draws on: match results, player and team statistics, news and transfer history. Its in-house rating formula has gone through several versions over that time, from the original Rating 1.0 introduced in 2010, through Rating 2.0 in 2017 and the Counter-Strike 2-era Rating 2.1 in 2024, to Rating 3.0, the version currently in force. Because the formulas differ, HLTV Brain treats them as separate measures rather than one continuous scale, flagging in several of its example answers that figures from different rating versions are "not calibrated against each other."

How the assistant works

HLTV built HLTV Brain to read its own back catalogue of results, ratings and roster records the way a human analyst would, without asking the reader to filter several separate stats pages by hand. The assistant works out what a question is really asking, pulls the matching data through HLTV's internal tools, and writes the reply itself, which is why its answers can come with a generated chart or table rather than only text.

It is "not allowed to 'make up' any facts or numbers on its own," the HLTV account that runs the site's official login said in the comments under the announcement, replying to a reader who asked how the tool avoids the kind of invented statistics that have dogged other AI products. The same reply said the assistant can also decline to answer outright when the underlying data cannot support a question, rather than guessing at one.

The assistant's launch follows a smaller, related change three days earlier. On October 10, HLTV began generating a short automatic brief on every team's profile page, summarizing a club's latest results, its upcoming schedule, and its standing in the current Major qualification race each time a reader opens that page. Where the team briefs run on a fixed template, HLTV Brain is built to handle an open-ended question and choose its own path through the data to answer it.

What early testers have asked it

HLTV published several example exchanges to show the kinds of questions the assistant is meant to handle.

In the duo example, the assistant was asked to reconstruct which pairs of players on the same team had defended outnumbered rounds most successfully at Big Events between 2025 and late September 2026, among duos sharing at least 50 qualifying rounds. It returned a Swedish-Danish Astralis pairing, Love "phzy" Smidebrant and Victor "Staehr" Staehr, with a 38.33 percent win rate across 60 qualifying situations, ahead of a Turkish Aurora Gaming duo and a Russian PARIVISION pairing further down the list.

The roster-history query asked how many teammates Ninjas in Pyjamas player Artem "r1nkle" Moroz, who has since moved on to G2 Esports, had played alongside during his time with the Swedish organization. HLTV Brain counted 12 players whose primary roster spells overlapped his between April 2024 and April 2026, then noted on its own that the figure would rise to 14 if a later benched spell at the club were counted too.

The event-rating example asked how unusual Finn "karrigan" Andersen's individual numbers were on the Falcons lineup that beat FURIA to win the IEM Cologne Major in June. The assistant found his Rating 3.0 for that event at 0.71, among the weakest of any player on a title-winning roster across its sample of 44 Big Events held since January 2025, and said a player rated that low had gone on to win the event in only about one case out of every 155.

All three examples share a pattern: each question needed the assistant to cross a boundary that HLTV's existing filter pages keep separate. Comparing two players' performance only in rounds where both were outnumbered, tracing one player's full list of ex-teammates across a two-year window, or ranking a single player's rating against every other title-winning roster since early 2025 would otherwise mean exporting several filtered views and combining them by hand. HLTV Brain is built to run that combination step itself and show only the final figure, with the underlying match list available if a reader wants to check it.

Where it still gets things wrong

HLTV's co-founder, posting under the site's official account, acknowledged in the comments that the tool is still being refined and pointed to that scoring mix-up as a case the team is working to catch. A separate reader reported that asking the same statistical question twice, in two different conversations, produced two different answers, and that one query returned only a partial explanation of why it had failed rather than a usable result.

Both problems point to the same underlying limit: the assistant can only be as reliable as the lookup tools and match records it chooses to run, and a wrong choice of tool, or a record with an ambiguous team order, can produce a confident-looking answer that is still wrong. HLTV has not said how many of its early testers' queries have run into this kind of error, or whether the scoreline mix-up was an isolated case or a pattern the team is still working through as the closed beta continues.

Why a stats site built its own chatbot

HLTV has spent close to two decades building its record of Counter-Strike results, player ratings and roster histories, and has long offered a search and filter interface across much of it. Several readers said in the comments that they had asked for a tool like this before, and welcomed it as a way to answer specific, cross-referenced questions, such as comparing a player's form across different ratings systems, that the existing search filters cannot easily combine in one pass. Other commenters argued that HLTV's filters already expose the same underlying data, and that a chatbot adds cost and the risk of a wrong answer without adding real capability on top.

The debate echoes a wider one playing out across competitive data sites generally, as more of them experiment with natural-language tools layered on top of existing statistics databases rather than building new ones from scratch. HLTV's approach keeps the assistant restricted to data it already owns and structures internally, which is also why the closed beta is being run gradually: a small invited group lets the team watch for cases, like the reversed scoreline, where a correctly retrieved number is still presented the wrong way.

What's next

HLTV said in the comments that it began inviting the first wave of waitlisted readers on October 14, a day after the announcement, and that further accounts will be added as the two-week beta continues. Readers can join the waitlist from the page HLTV has set up for the assistant. HLTV did not say when, or whether, HLTV Brain would move out of closed beta and open to every visitor to the site.

The .esports registry publishes Esports Observer, so the whole scene can read its news for free.