Tier-1 teams walk on stage already knowing which site bleeds, who force-buys on a loss, and where the AWPer stands at 0:16. They know because an analyst spent days in the opponent's demos. On FACEIT, you get the accept button and about four minutes — against opponents whose entire competitive history is public, sitting on FACEIT's servers, read by no one.
That gap — public data, zero time — is exactly the shape of problem AI is good at. So we wired Claude, Anthropic's family of AI models, into every stage of match prep on FaceitScout. This post is the full tour: what the AI actually does, where the numbers come from, and — because "AI-powered" is doing a lot of unsupervised work in marketing copy these days — what our AI is never allowed to do.
Every AI feature on FaceitScout follows one pipeline. Scanners and demo parsers produce verified numbers first — win rates from real match results, plant sites from real planter coordinates, buy habits from real dollars spent. Claude's only job is the last step: turning that sheet of numbers into a briefing you can absorb on the warmup screen, the way an analyst turns film notes into a game plan.
If the data isn't there, the AI says nothing. A lobby with no parsed demos returns no data — not confident fiction. That constraint sounds obvious; most of the engineering below exists to enforce it.
The scanner pulls up to 50 recent matches per enemy player — per-map win rates, K/D, ADR, entry success, clutch rates, streaks, Elo. That's ten players times a season of history: accurate, and unreadable at four minutes to knife round.
Match Intel hands the whole sheet to Claude and gets back the version an analyst would say out loud: a map-by-map veto read referencing actual numbers, a threat assessment per enemy player (who's genuinely dangerous vs. who's just loud), and a five-point "how to win" list. It factors in the live veto state, too — if three maps are already banned, the brief is about the maps that can still happen.
Stats tell you how good they are. Demos tell you what they do. Enemy Scout downloads your opponents' recent demos on the picked map and parses every round into a tendency digest: which site their CT side actually loses, where they plant on T (derived from real planter coordinates), pistol conversions on both sides, buy habits, opening-duel win rates.
Then Claude Sonnet 5 — the model we use for analyst-grade reads — turns the digest into a T-side plan, a CT-side plan, and an economy read. The output is opinionated on purpose: "hit A early, they lose it 62% of the time" is useful; a spreadsheet restated in prose is not.
One honesty detail we're oddly proud of: the digest's field names literally encode whose side each number describes — enemy_ct_site_lost_pct, not site_pct — because an early model version once read "they lose B" and coolly advised us to stack B on our own CT side. Perspective is now baked into the data itself, and the prompt polices which facts are allowed in which section. AI is a great analyst and a terrible mind reader; we stopped asking it to be the second thing.
The browser extension puts the same engine inside FACEIT itself. From the Enemy Scout tab in a live match room, Team Read checks whether we already hold parsed demos for the five people you're about to play — and if so, renders the full T/CT/economy briefing right in the side panel, no site visit, no wait, no quota spent on data we already have.
It comes with a defaults radar: a map overlay showing where each opponent actually stands 10–22 seconds into the round, clustered from their real position data across parsed demos. That one isn't AI at all — it's pure geometry from the demos — which is rather the point: we use AI where reading is the problem, and plain math where math is the answer.
For Coach-tier teams, the team prep hub runs the deepest version: a tendency profile of an upcoming opponent built across their parsed demos — T-side pacing from plant times, entry and post-plant hotspots, retake habits, and an inferred role for each player. Claude writes the anti-strat briefing off that profile, and a data-driven counter-setup lands directly on the tactics board as a draggable, editable step — the scout's report and the whiteboard in one motion.
Different jobs, different models. Anti-strats and Team Reads run on Claude Sonnet 5, where the tactical reasoning quality is worth the cost. Match Intel briefs run on Claude Haiku — fast and cheap enough to fire on every scan without making you watch a spinner. When the models improve, every surface above improves on the same day, with zero changes to the data pipeline underneath — the numbers were never the AI's job to begin with.
| Feature | What Claude does | Who gets it |
|---|---|---|
| Match Intel | Reads the full scan → veto read, player threats, how-to-win | Pro & Coach |
| AI Anti-Strat | Reads demo digests → T/CT/economy game plan | Coach unlimited · Pro & Free via ⚡ tokens (Pro also gets the data digest free) |
| Team Read | Same briefing, one click inside the FACEIT match room | Coach (AI) · Pro (digest) |
| Team prep anti-strat | Deep opponent profile → briefing + board-ready counter-setup | Coach |
Free players aren't locked out: Rewards pays ⚡ tokens for playing and winning with the scanner, and one token buys a full Enemy Scout run with the AI anti-strat included. Grinders who never spend a cent still get the analyst — just not on tap.
Your next lobby has five public histories nobody's reading.
Scan your next match free →The pattern behind all of it: the grind rewards preparation, preparation takes time, and time is the one thing a solo-queue player doesn't have between accept and knife round. The data was always public. Now it reads itself. ⚡