The white paper adds a dedicated Industry Source Research chapter. This article distils the core findings: AI platforms differ sharply in source preferences; Appendix B uses 10-industry x 6-platform measured data to answer 'where should content be placed', and provides a standard order to reverse-engineer source strategy from target platforms.
The single biggest expansion in the 2027 edition turns "sources" from a methodology slogan into an executable engineering chapter - industry source research. This article distils the chapter's core conclusions and directly answers the question SMEs care about most: on which sources should my content actually be placed?
I. Why source research is the 2027 addition
For the past year, the industry has debated "how many articles to publish, how many platforms to cover" but researched too little "whom AI actually cites in real answers". The white paper fills this gap with systematically collected real citation data: explain "whom AI trusts" first, then talk about where the budget goes - the order cannot be reversed.
A counter-intuitive conclusion: placing content on the wrong ecosystem, no matter how well done, will not enter the target engine's answer. This is the most easily wasted budget for SMEs.
II. Two camps: ecosystem-native vs web-wide collectors
The six major AI platforms fall into two types with completely different playbooks:
- Ecosystem-native (Doubao, Yuanbao, Baidu AI): backed by their own content matrices, they prioritise citing their own ecosystem content. Playbook - complete the content layout inside the corresponding ecosystem first; the entry is inside someone else's house.
- Web-wide collectors (DeepSeek, Kimi, Qwen): no self-built content base, they rely on web-wide material collection. Playbook - produce professionally solid content; the official website and vertical sites are the main battleground.
The conclusion is direct: hitting the former hinges on "in-ecosystem entry", hitting the latter on "off-site content quality" - two very different budget and team configurations, and conflating them is the most common strategic mistake.
III. Hard data from Appendix B: ecosystem lock-in
Appendix B lines up the #1 cited source across 10 industries, and the pattern is written on its face: platforms with their own ecosystem lock the entry tightly; web-wide platforms scatter answers widely.
- Doubao: the #1 cited source in 10/10 industries is Douyin;
- Yuanbao: 10/10 industries is WeChat Official Account;
- Baidu AI: 9/10 industries is Baijiahao;
- while DeepSeek, Kimi and Qwen swap their #1 across industries frequently.
This means: if your audience is on Doubao, Douyin content layout is the highest-priority task; if on DeepSeek, Zhihu and in-depth industry articles are the main battleground.
IV. Source composition of AI citations: where the money should go
The white paper tallies the category share of sources AI actually cites, with illuminating implications for budget allocation:
- Vertical media 24.9% + general media 24.7% - nearly half combined - the main GEO battleground;
- Self-media 21.9% - the key to influencing Yuanbao and Baidu AI;
- Official media 17.8% - what Qwen eats up;
- Official website 8.4% - small base, but the only 100%-controllable piece, recognised by all platforms.
Two things to remember: ① "spread thin" beats "pick the right vertical" less; ② although the official website is small in share, it is recognised by all platforms and is the "cited object" of source building.
V. Two types of high-value "one-pitch-many-harvest" sources
The white paper distinguishes two high-value targets:
- Ecosystem anchor: a source that wins across one platform - Douyin (Doubao 10/10), WeChat Official Account (Yuanbao 10/10), Baijiahao (Baidu AI 9/10), Baidu Baike (DeepSeek 5/10), Sohu (Kimi 7/10).
- Cross-platform common base: a source that one placement covers multiple platforms - Autohome / Pacific Auto (auto 6/6), Sina system (home & more 6/6), Sohu (maternity / education / medical 6 platforms), SMZDM (several consumer tracks 5/6).
The latter is the key to budget efficiency: the same money covers more platforms, higher marginal return.
VI. Standard operating order: pick the target first, then diagnose
The white paper gives a directly repeatable order to avoid the waste of "covering all platforms from the start":
- Decide the 1-2 AI platforms you most want to influence (best match your audience);
- Expand the question library from brand terms to category, scenario, region, comparison, price, word-of-mouth and pitfall terms;
- Ask each question in the target engine and record which pages it cites - that is your source list;
- See which platforms in the list you can enter, whether there is an alternative path for those you cannot, and which positions competitors occupy;
- Generate channel-adapted content (differentiated expression under the same semantic coordinates);
- Monitor results, attribute changes, then do targeted re-investment - this loop must run continuously.
VII. Three recommendations for SMEs
- Pick the platform first, then spread content: do not cover all platforms at once; replicate after nailing 1-2;
- The official website is both foundation and source: make qualifications, certifications, cases, parameters, sources and dates all verifiable, so third parties verify against your site when citing you;
- Layered investment when budget is tight: layer 1 self-build (asset stays in hand), layer 2 purchase (media relations and time cost are high), layer 3 nurture (community accumulates over time).
Closing
The essence of source research is turning "what GEO should do" from metaphysics into cartography. When you can say "my audience is on Doubao, Doubao trusts Douyin, so my Douyin content comes first", the budget is truly spent where it counts.