How to Optimize Website Content for Generative AI
Ranking as highly as possible on the search engine results page has been the sole object of content optimization for over two decades. Now, as more and more people rely on tools like ChatGPT, Perplexity, and Google’s AI Overview to produce their answers, digital marketers have been faced with an entirely new question: how does one optimize content for AI?
That is a difficult question to answer, not least because “AI” has become a buzzword of maximal ambiguity. Yet deeper lies the problem that generative artificial intelligence is an immensely complex technology which cannot be reduced to the relatively straightforward principles that govern a search-engine algorithm. The unfortunate result has been a flood of well-intentioned theories and tip-sheets across the internet that is hardly proportional to the very limited body of genuine research on the matter.
Nonetheless, the research is out there—and it’s very promising indeed. A 2024 paper by Aggarwal, Murahari et al. determined that a handful of specific approaches under the umbrella of Generative Engine Optimization (GEO) could increase AI citations by up to 40%. The results were even more dramatic for low-ranking sites, which experienced visibility improvements of up to 115% with certain optimization methods.
The paper defines "generative engines" as systems that retrieve relevant sources and then use a large-language model (LLM) to synthesize them into responses. Its methods therefore apply to both AI-enhanced search engines like Google's AI Overview and search-augmented LLMs like ChatGPT, but do not apply to LLMs that lack real-time web access nor to the blue-link rankings of a traditional search-engine results page. What follows is a practical guide to implementing these research-based GEO tactics for improved AI-visibility in a new era of search.
1. Use as Many Citations as Possible
Generative engines are trained to fetch verifiable information. Given the sheer volume of data these engines can comb through in the course of a search, it should not be surprising that self-verifying content is quickly singled out from the rest. The more sources a webpage cites, the less likely it is to be self-promotional, subjective, or simply mistaken.
Effective GEO therefore requires that any non-original factual claims are accompanied by an inline citation or reference link. The more reputable the source, the more trustworthy the claim. Peer-reviewed research, reports form independent or government bodies, and authoritative industry voices signal to the engine that the accompanying message is well-grounded and fit to be given as a response to the user’s query.
2. Include Statistics for Quantifyable Information
The use of statistics to accompany factual claims carries two benefits. First, it once again reinforces the credibility of the webpage and increases the likelihood that a generative engine will trust it as a source. Second, it makes for good answers. Statistics are concrete, memorable, and digestible—precisely the kind of answer that generative engines are conditioned to produce.
Content creators should quantify anything that can be quantified and ensure that each non-original number is backed by a reputable source, per tactic one above. Writing “a 50% increase” will always be preferable to “a significant increase,” just as “over 600” is far more impactful than “numerous.” Then, when the engine collects hard numbers for its answer, the webpage is often cited right along with them.
3. Quote Relevant Sources Directly
As the final rule in the credibility triad, content should include direct quotes from relevant authorities to show that it is well-researched and that it represents its sources faithfully. This tactic is especially useful in cases where the subject of discussion is more qualitative than quantitative. Historical contexts and speculative insights are unlikely to fit into a statistic, regardless of their veracity.
In such cases, content creators should seek out authoritative voices and quote them directly—in quotation marks—followed by an inline citation. This allows the source to literally speak for itself. Generative engines can then interpret any new qualitative claims in the context of the existing, cited claims and determine that the content is not mere opinion but part of a broader conversation.
4. Employ Stylistic Emphasis
With credibility established, it is time to convey the intended message in a way that generative engines are likely to carry over into an answer. This essentially means improving readability. The same things that allow readers to digest the message of a text help engines to locate, synthesize, and extract information.
Practically, this means taking time to ensure that all content has a clear, logical structure where each section addresses a specific aspect of a coherent message. Sections must be indicated through descriptive headings and subheadings. The writing itself should be professional, direct, and—crucially—conversational in order to appeal to the components of generative engines responsible for processing natural language.
5. Avoid Keyword Stuffing
As referenced in tactic four above, generative engines are designed to read and write natural language. They evaluate information semantically—understanding meaning, context, and relationships between ideas—rather than just matching keywords and evaluating technical factors. As a result, traditional search-engine optimization (SEO) tactics like keyword-stuffing performed about 10% worse in the Aggarwal study than content which had not been optimized at all.
Instead, content creators should weave long-tail “keywords” naturally into flowing text that prioritizes reader understanding and comprehensively answers user queries. These keywords must center around concepts, rather than particular strings of letters, and make use of immediate context as they would in conversation. If the content appears optimized, it is doing something wrong.
Conclusion
Search is changing, and content optimization is changing with it. But the GEO tactics outlined above reveal a remarkable principle that is not so novel at all: optimized content is good content. “How do we manipulate AI to recommend our brand?” is perhaps the wrong question. One might instead ask, “How do we create content that genuinely deserves recommendation?” Those who answer that question by creating high-quality, well-researched, well-written content will find themselves optimizing for AI almost by default.
Be sure to check out our other blogs on useful and interesting public relations topics, like demonstrating thought leadership through regulatory insights.
