Chapter 2: The core principles of Impact GEO

Chapter 1 laid out the problem in full: AI search has magnified what a piece of content can do, and the current incentive structure makes poisoning cheaper than building. The more basic question is what a different incentive structure should look like.

This chapter proposes four principles. They are not moral commandments. They are a systematic answer to a narrower question: which GEO practices win over a long enough horizon. Each one tracks a structural change already under way in AI search.

Principle 1: Source transparency — content must be traceable and verifiable

The deepest difference between AI search and traditional search is the relationship between the user and the information. In traditional search the user sees a list of links and judges each source personally. In AI search the user sees one synthesized answer, and does not know which sources went into it, which were given more weight, or which had an interest in the outcome.

The core requirement of source transparency is this: when AI cites content, that content's origin can be traced accurately; and when AI tells a user something is a fact, the user can check what it rests on.

This is not a moral ask. It is an algorithmic rule already being enforced. The mainstream large language models run explicit source tiering. Government websites, mainstream media, industry regulators and authoritative academic databases are classified as top-tier trusted sources, crawled first and cited with a standing advantage. Independent accounts, advertorials and content from unlicensed sites are demoted and read with caution. The higher a source's tier, the more likely AI is to cite it.

So source transparency is not only the right thing to do; it is the thing the algorithms recognize. Where a brand's content has clear origins, a complete evidence chain and visible interests behind it, AI systems return a higher citation weight. Where origins are vague and untraceable, the content may get crawled today and will still be flagged as low quality in later cross-checks, then phased out.

T/CAPT 026—2026, the association standard that took effect in August 2026, sets out an A/B/C/D four-tier credibility rating and requires brand knowledge bases to keep three zones separate: a fact zone, an opinion zone and a marketing zone. Its full title is Specification for Trusted Information Dissemination and Information Ecosystem Governance in Generative Engine Optimization (GEO). The logic behind that design is source transparency itself. Verifiable objective fact is held apart from judgment and from promotion, so that AI can tell "this is a fact" from "this is an opinion" from "this is an ad" when it cites.

For a GEO practitioner the principle translates into three working points. In production, every core factual claim carries an explicit source. In organization, factual content and marketing content are managed in separate zones. In monitoring, every piece of content AI cites can be traced back to its primary source.

Principle 2: Truth first — accurate content is the best optimization

In the traditional SEO era the logic was to make search engines believe your content was good. In AI search the logic changes: the model judges the content itself and cross-verifies it against other sources. Large language models both cross-check and remember. Exaggeration and false claims may be crawled for a while, then get marked as low quality and dropped.

The principle states it plainly: under AI's citation mechanism, accurate content is not the better option. It is the only option that lasts.

The data supports that reading. Weight for authoritative sources in AI search is rising across the board. May 2026 figures show Wikipedia's AI citation volume up 55.2% month over month and the U.S. National Institutes of Health (NIH) up 41.3%, moving it into the top three cited sources overall. The research conclusion is blunt: AI models are actively looking for authoritative sources.

At the same time the short-term payoff of poisoning is being cut back deliberately. After the 2026 Spring Festival, DeepSeek compressed the sources it reads closely from 10-15 down to 4-5, and citation rates for mass-distributed press-release content fell sharply. The odds of a single-source claim being blocked reach 95%, and content that contradicts itself across platforms is 82% less likely to be featured. The platforms are subtracting on purpose: dropping low-value content and raising the weight of content that is accurate and authoritative.

Truth first has three operational layers. Content: every thousand words should carry at least five verifiable statistics and at least two authoritative citations, with the core conclusion stated in the opening paragraph. Structure: the argument should have a clear chain of facts so AI can identify and extract the evidential relationships. Verification: a third party should be able to confirm it independently, rather than taking the brand's own word for it.

Truth first does not mean giving up marketing copy. It means keeping the two distinguishable: marketing should be recognizable as marketing, and factual claims must survive verification. The standard's rule that factual claims in marketing copy must match the fact-zone version is exactly this principle written into regulation.

Principle 3: Ecosystem health is the moat

The first two principles say what to do. The third says why keep doing it.

The argument is short. In a market where trust in AI search is falling, source quality stops being a moral bonus and becomes a competitive factor that directly determines retention.

The trend is visible in the numbers. In 2025, 82% of consumers found AI search more helpful than traditional search; by 2026 that was 54%, a fall of 28 percentage points in one year. More than half of consumers now doubt the reliability of AI answers, and 86% go check the primary sources themselves when the information matters.

What matters more is the change in behavior. 72.4% of users switch engines after AI gets a citation wrong twice in a row. For an AI search platform, the cost of source quality is therefore charged straight to retention. A product polluted by low-quality content loses users who vote with their feet; a product that keeps producing trustworthy answers earns loyalty. The same logic holds for brands and service providers: when your content becomes a source behind answers people trust, you gain more than exposure. You gain standing.

The direction of industry standards confirms it. The stated goal of T/CAPT 026—2026 is to move GEO away from the technical pursuit of exposure, mentions and rankings, and toward building verifiable factual assets, controlling dissemination for compliance, and governing the information ecosystem — in its words, shifting the ecosystem from traffic-driven to trust-driven. The China Advertising Association (CAA) and the China AI Industry Development Alliance (AIIA) are pushing trusted GEO ecosystems in parallel, and the first companies to pass the GEO service trust assessment have been certified.

Taken together, these moves point one way: ecosystem health is turning from an aspiration into an entry requirement. Once the regulatory framework matures and platform algorithms keep tilting toward credible sources, the brands and providers that built their content systems early will hold a structural advantage.

Principle 4: Positive externality — every optimization should add to the ecosystem

The first three principles are about getting your own content cited correctly. The fourth widens the frame from the self to the whole ecosystem: every optimization should leave the AI information ecosystem better off while it works for you.

Positive externality is an economics term for an act that benefits others without them paying for it. In the AI content ecosystem it looks like this: a careful, verifiable piece of content does not only help AI answer questions about your brand accurately. It adds trustworthy knowledge about the subject to the AI corpus. Later users asking related questions, and later models retrieving related topics, get better answers because that page exists.

Poisoning produces the opposite. A fabricated page misleads the reader in front of it and pollutes the corpus, so countless later users and conversations inherit the error. The Apollo-9 smart band case shows the real radius of a fake advertorial: it is not the people who read it, it is every later conversation in which AI repeats it as fact.

So the principle's requirement is about measurement. When you evaluate GEO performance, do not only count how often you were cited. Ask whether your content made the AI information ecosystem more trustworthy.

That has concrete implications at three points. Production: make content with public knowledge value, not just brand praise. A page that answers questions about the brand and hands the industry something reusable creates brand value and ecosystem value at once. Measurement: add an ecosystem contribution dimension, asking whether AI's answers on the topic became more credible after they cited you. Strategy: put ecosystem health into the long-term competitive model. In a market where everyone poisons, no brand is insulated. In a market where everyone builds credible content, every participant benefits.

The fourth principle closes the loop with the other three. Source transparency makes content traceable. Truth first makes it survive inspection. Ecosystem health as a moat makes good practice pay. Positive externality makes sure each optimization adds value to the ecosystem rather than withdrawing it. All four lead to the same conclusion: Impact GEO is not the ethical version of GEO. It is the long-termist version, in step with how algorithms evolve, regulation tightens and trust has to be rebuilt.