Recommendation method
PC GameDex is built for players who know the feeling they want but do not want another popularity list. The recommendation system combines catalog facts, concrete gameplay signals, guide-specific editorial rules, and session feedback to rank PC games by fit.
Core approach
PC GameDex treats the text someone enters as the center of the search. A request is not reduced to one broad genre tag. The matcher looks for play style, progression, pacing, combat shape, management depth, co-op needs, platform filters, and explicit negatives such as genres or mechanics the player wants to avoid.
Genre names are useful, but they are often too loose. A game can carry a tag because it has one feature from that category. The stronger signal is the loop: planning, building, exploring, fighting, trading, managing resources, repeating runs, unlocking options, or improving a character over time.
Trending PC games are useful for the home and trending pages, but they do not get a free boost inside Discovery or guides. That separation matters because popular games often dominate generic lists even when they are poor fits for the player's actual request.
Right-direction, wrong-direction, and played-it actions reshape the current feed. They do not create a long-term taste profile and raw feedback notes are not sent to OpenAI recommendation curation. When feedback names a title family, the app can derive a simple exclusion for the current session without turning that note into a stored profile.
Catalog sources
The public beta combines a local baseline catalog with indexed provider records from RAWG and Steam-backed ingestion. Provider rows are normalized into titles, genres, themes, descriptions, cover art, store links, release years, ratings, and source URLs. The site does not scrape storefront pages for monetized v1; new providers should come through official, documented, or partner-friendly sources.
Some provider metadata is noisy, so PC GameDex adds local interpretation for known mechanics. Construction, automation, management, co-op, progression, survival, and story-heavy games can all need different matching rules. The goal is to compare games by the way they play, not by a single broad tag.
Quality checks
Guide logic
Guide pages are organized around player intent such as building, automating, managing, upgrading, surviving, exploring, or playing with friends. That keeps them broader than one franchise or one store chart.
Interactive discoverySearches can be tuned in the feedThe same catalog logic powers the interactive feed, where store filters, genre filters, and session feedback can narrow a broad guide into a more personal shortlist.
Separate trend surfacePopularity is visible but separatedTrending games are useful context, but Discovery and guide ranking are designed around fit rather than current visibility.
Privacy postureRecommendation data is minimizedThe current beta does not require an account for Discovery, keeps saved games local-first, and avoids turning private feedback notes into a stored taste profile.
Privacy posture
PC GameDex does not require an account for Discovery. Saved games are local-first in the browser. Raw private notes are not stored as a taste profile, and OpenAI curation only receives a bounded candidate list plus structured intent when AI features are configured.
Current limitations
The recommendation system is only as good as the indexed catalog and normalized metadata. Some provider rows still need better cleanup, and rare games can be missing until background ingestion or query coverage imports add them. The goal is to keep expanding coverage while staying clear about why each visible pick was chosen.