What is a source graph?
When ChatGPT, Perplexity or Google AI Mode answers “what is the best X for Y”, the retrieved passages rarely come only from the winner’s website. They come from review platforms, comparison articles, Reddit and forum threads, Wikipedia and Wikidata, industry directories, YouTube transcripts and news coverage. That constellation of external sources is your source graph. Engines treat it as corroboration: grounding checks your site’s claims against it, and citation often points at the third-party source rather than at you.
Why does the source graph outweigh your own site?
Every claim on your own domain is self-interested; engines know this. A product described as “industry-leading” on your homepage is marketing; the same description in three independent reviews is evidence. This is why Perplexity’s sentence-level citations so often point to third parties, and why ChatGPT shopping recommendations lean on review coverage. Your website defines the facts; the source graph validates them. Both are required, but most brands invest 95% in the first and 5% in the second.
How do you map your current source graph?
- Run your fixed question set against the major engines and record every cited domain — this is the empirical source graph for your category.
- Classify sources: review platforms, editorial, community, reference (Wikipedia/Wikidata), directories, video.
- Mark presence: are you mentioned, accurately described, and linked — on each?
- Score gaps by engine: the sources Perplexity cites differ from the Bing-heavy set ChatGPT surfaces, so map per engine where resources allow.
How do you build the graph?
- Reviews first. Systematic review generation on the platforms engines cite in your category. See review management.
- Reference data. A Wikipedia article where notable, and complete Wikidata entries either way. See Wikipedia and Wikidata.
- Editorial coverage. Digital PR that earns genuine coverage — data studies, expert commentary — not link placements. See digital PR for GEO.
- Community presence. Helpful participation where buyers actually discuss your category. See community seeding.
- Consistency enforcement. Every new mention repeats the same facts as your registry — the source graph multiplies the damage of inconsistency. See grounding fact consistency.
Common pitfalls
The dominant failure is buying placements: engines increasingly discount sources that exist only to sell mentions, and one pay-to-play “review” can poison grounding trust. Second, brands build the graph once and abandon it; sources decay, listings go stale, and engines notice. Third, teams ignore community sources because they cannot control the message — but an unanswered negative thread is part of your graph whether you like it or not. Finally, do not neglect measurement: track which sources engines cite for your category quarterly, because the trusted-source set itself shifts.
Budget your effort honestly across the graph. A pragmatic split that works for most brands: the largest share to reviews, since they feed both grounding and sentiment; a meaningful share to editorial coverage via data-driven PR; a small but non-negotiable share to reference data like Wikidata; and continuous light investment in community presence. Rebalance quarterly based on which source types your question-set answers actually cite.