Janak Sunil, CEO, Bear (YC F25). Helping companies market to AI agents.
How humans retrieve information has changed forever. Unlike with Google, AI is trained to generate an answer tailored to the user’s intent. Results pages are making place for short, rich and definitive answers.
That shift has enormous business implications. According to the CEO of CustomGPT, leads originating from ChatGPT are almost seven times more likely to convert to paying customers than those from Google. At my company, we’ve seen a similar pattern: AI-search traffic accounts for just 0.6% of clicks, yet generates 12% of inbound revenue.
And yet, I see very few brands actively optimize, or even know their visibility on AI search, citing their black-box nature as the reason. But with a data-driven approach, using over 5.5 million data points collected by our company, we’ve been able to derive some valuable insights I’d like to share with you.
Why Trust The Insights?
This article draws from two sources of evidence:
First, our internal dataset comprises over 5 million query-response pairs and 15 million distinct sources, collected across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews. This set consists of high-fidelity data collected from the front end of these AI search models and sampled over many distinct domains, representing true-to-user prompts and answers.
We also back our research with recent peer-reviewed research. For example, this University of Toronto study analyzed the correlation between SEO factors and AI-engine citations, and a 2025 cross-lingual study looked into role-based intent optimization for conversational models. Yet another source, a flagship GEO paper, was used to complement our own vast dataset.
Many of the findings throughout these papers support our own conclusions. By analyzing a wide variety of parameters and factors across these millions of data points, including URLs, content structures, keyword inclusion, length and more, we’ve developed comprehensive insights into ChatGPT’s search algorithm. This is a framework we now use internally to guide our generative engine optimization (GEO) strategy. For this article, ChatGPT will be the primary focus.
Core Strategies For AI Search Optimization
1. Focus on the slug.
Among all technical variables, slug structure (the descriptive, end part of your URL) had, surprisingly, one of the highest correlations with AI citation frequency. Clean, specific slugs and perfect matches performed significantly better than overly general or keyword-stuffed slugs. For the search “best generative engine optimization tools,” the slugs of the top three sources were:
1. “best-generative-engine-optimization-tools”
2. “best-generative-engine-optimization-tools-ai-visibility”
3. “generative-engine-optimization-tools”
This nearly perfect match signifies an extremely strong correlation between query and slug. In contrast, traditional SEO metrics such as backlinks and keyword density showed weak predictive power. The earlier-mentioned University of Toronto study found that SEO correlation ranged from 40% down to 15% for localized or specialized topics.
2. Publish more, sooner.
AI search engines update and “crawl” content faster than Google. For our company, new articles began appearing in ChatGPT and Perplexity citations within two days of publishing. Consistent, high-volume output was shown to consistently increase chances of being found by AI search engines. This contradicts SEO’s more calculated, scheduled approach to content. In GEO, speed is everything.
3. Optimize for intent, not keywords.
As already alluded to, keyword padding is rapidly losing relevance. ChatGPT prioritizes intent modeling— understanding the user’s “why” behind the query. AI models are trained specifically to understand a user’s intent and role, and generate results that satisfy exactly those.
4. Prioritize readability.
AI models heavily reward clarity and structure. Coherent, focused paragraphs with clear argument flow consistently outperform bloated, keyword-stuffed posts. From our data, top-performing sources share common traits:
• An average Flesch-Kincaid reading score between 60 and 75 (conversational but informed)
• Clearly defined sections and transitions (especially prominent introductions and summaries)
• Logical, narrative progression from problem to solution
5. Forget about clickbait and backlinks.
Engagement signals like click-through rate and backlinks are nearly negligible in AI search. Models don’t “click”; they consume.
This is most clearly highlighted in Reddit’s recent stock slide, leading to a core realization: User-generated content is no longer a source of truth. AI search values structured, authoritative content over noise. If your entire content strategy is centered around sites like Reddit, I think your approach is fundamentally dated.
Best Practices For Creating AI-Friendly Content
The following components significantly improve visibility in AI answers, according to both research and our data:
1. Include a table of contents.
A hyperlinked TOC gives structure and saves tokens, helping both readers and language models parse context. It also increases section-level citations; many AI responses pull text from the first two or final two sections of an article.
2. Add author and citation metadata.
Models associate credibility with transparent bylines, bios and citations. Adding an author schema and linked citations (e.g., to studies, datasets or official sites) helps establish authority and domain trust.
3. Incorporate bullet points and numbered lists.
Chunked, “liftable” formats—especially lists of five to seven items—make it easier for AI to reassemble your content in an answer. Our internal tests found that bullet-formatted takeaways doubled citation frequency compared to dense prose.
4. Frontload and conclude with insight.
The opening and closing paragraphs are disproportionately cited. Begin with a clear thesis and end with a concrete insight or FAQs. In multiple experiments, restating your conclusion in the final 100 words increased the likelihood of AI lift by around 30%.
Summary
AI search is redefining the world’s medium of information. The search scene is shifting rapidly and content success now depends on semantic relevance, intent clarity and genuine value.
I view AI optimization as SEO’s successor. I aim to continue my studies regarding the expansion of this unexplored frontier.
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
