The New Rules of AI Optimization How to Drive Traffic in the Age of ChatGPT
Elad Hefetz · AI Marketers Guild · July 14, 2025
The game has changed. With the rise of AI Overviews and ChatGPT, organic website traffic is on the decline.
The game has changed. With the rise of AI Overviews and ChatGPT, organic website traffic is on the decline. In this AI Insiders session, David Berkowitz, founder of AI Marketers Guild, is joined by Elad Hefetz, CEO of Airfleet, to dissect the new reality of digital marketing.
Key moments
Elad introduces his GenAI Optimization (GEO) Framework, a five-part strategy to adapt to a world where AI-powered answers are replacing clicks.
Learn why traditional SEO is no longer enough and how to optimize your content, not just for search engines, but for the AI models that are increasingly becoming the gatekeepers of information.
[01:17] What is the "Great Decoupling" in SEO, and how does it affect website traffic?
Answer / Description:
The "Great Decoupling" is an organic search trend where website impressions in search engines increase while click-through rates (CTR) and organic traffic decrease. This occurs because search engines like Google use AI Overviews to answer user queries directly on the search engine results page (SERP), eliminating the need for users to click through to external websites.
Elad Hefetz explains that this trend is especially prevalent in B2B technology sectors. Because AI overviews use a website's content to immediately formulate and present an answer, top-of-funnel informational queries are seeing dramatic traffic drops. Users no longer search for links; they search for immediate answers, making website visits redundant for simple queries.
[03:31] How does ChatGPT influence user search behavior compared to traditional search engines?
Answer / Description:
ChatGPT shifts search behavior from simple link-finding to conversational, multi-step problem solving. It acts on "steroids" compared to traditional search engines by prompting the user for follow-up actions, comparing products, and summarizing tables directly within the chat interface, which keeps users entirely within the AI environment.
Elad Hefetz notes that when users query ChatGPT, the system proactively leads them to next steps. For example, if a user asks how to solve a problem, ChatGPT will recommend strategies, offer to list relevant products, and then offer to compare product features side-by-side in a table. This seamless, conversational workflow creates a highly engaging experience where users have no practical reason to visit individual company websites.
[05:19] What are the core pillars of a Generative Engine Optimization (GEO) framework?
Answer / Description:
A comprehensive Generative Engine Optimization (GEO) framework consists of five core pillars: Strategy (prompt research), Monitoring (measuring visibility), Off-Page Authority (third-party brand optimization), On-Page Technical Adjustments, and On-Page Content Optimization.
Elad Hefetz outlines these five concepts as the roadmap to optimize a brand for AI discovery. It starts with converting target keywords into natural prompts and setting up monitoring systems to track how the brand ranks inside AI tools. Next, marketers must manage off-page channels like PR, directories, and third-party media that train the AI. Finally, on-page optimization requires technical elements like configuring the robots.txt and adding structured content blocks designed specifically for AI models to digest.
[07:06] How can marketers perform prompt and keyword research for AI search engines?
Answer / Description:
Marketers can conduct prompt and keyword research for AI search engines by converting traditional SEO keywords into natural language prompts, analyzing popular questions on community forums like Reddit, and mining customer conversations from transcription tools like Gong.
Because platforms like ChatGPT do not publish public prompt databases or search volume tools, Elad Hefetz recommends synthesizing traditional search volume data from SEMrush or Google Keyword Planner and translating those terms into conversational questions. Marketers can also extract real customer language by analyzing subreddits or transcribing sales and customer support calls. Using AI, marketers can then consolidate hundreds of raw inputs into a core list of 30 to 50 high-intent, recurring prompts to optimize against. He cautions against asking ChatGPT directly for search volumes, as LLMs do not possess this data and will fabricate answers.
[10:38] What metrics and results should you expect when optimizing for ChatGPT and AI search?
Answer / Description:
Optimizing for ChatGPT and AI search primarily results in an increase in organic branded search traffic, a rise in AI crawler bot activity on your site, and modest referral traffic from citations.
Elad Hefetz clarifies that because ChatGPT does not display links by default in its database-driven answers, users typically read the AI's recommendations, copy the brand name, and manually paste it into Google or LinkedIn. Therefore, successful GEO manifests as an uplift in organic branded search traffic. Additionally, marketers should monitor direct referral traffic from AI platforms (though CTR on citations is exceptionally low) and track the frequency of AI bots crawling their websites. To capture these offline and conversational attribution channels, Hefetz recommends adding a self-attribution question like "How did you hear about us?" on sign-up forms with specific options for ChatGPT, Gemini, and Perplexity.
[11:33] What triggers ChatGPT to perform an external web search (RAG) and display source citations?
Answer / Description:
ChatGPT is triggered to perform an external web search using Retrieval-Augmented Generation (RAG) and display citations when a query involves local intent, asks for real-time news, targets a highly obscure or niche topic, or explicitly requests an external web search.
By default, ChatGPT attempts to answer questions from its static offline training data without showing links, as explained by Elad Hefetz. However, if a user types a local query (e.g., "near me"), asks about current events (e.g., today's weather or recent news), probes a specialized niche where the internal data is insufficient, or specifically instructs the tool to search the web, the system initiates an external search. Marketers should understand these triggers because citations and links only appear when these external queries are activated.
[22:18] How does the AI training data cutoff impact brand visibility in generative answers?
Answer / Description:
The training data cutoff means that AI models like ChatGPT cannot show a brand in their default offline answers if the brand's key information, repositioning, or launch occurred after the model's last training update.
Elad Hefetz points out that LLMs update their foundational training datasets infrequently—typically every six months to a year. For example, if an LLM's dataset was last updated in June 2024, any rebranding, product launches, or messaging changes made after that date will not exist in the AI's default offline memory. Consequently, a brand will only appear in conversational search results if the user triggers a real-time external web search. This makes it critical for marketers to optimize their online presence continuously so that their updated authority signal is ingested during the next model training cycle.
[24:50] Which third-party websites have the greatest influence on ChatGPT's recommendations?
Answer / Description:
ChatGPT's brand recommendations are heavily influenced by highly authoritative sources, including Wikipedia, Google Business Profile (formerly Google My Business), major industry publications, niche review platforms, and open social forums like Reddit and LinkedIn.
Elad Hefetz emphasizes that an individual brand's website makes up a microscopic fraction of the internet, meaning AI models do not rely solely on your own site to form opinions about your company. Instead, they scan established, trusted platforms. For general credibility, Wikipedia and Google Business Profile are essential. For B2B products and service agencies, platforms like G2, Capterra, Trustpilot, Clutch, and Design Rush dictate recommendations. Additionally, ChatGPT crawls open social spaces like Reddit and LinkedIn for user sentiment, whereas it cannot access gated ecosystems like Meta or X (formerly Twitter).
[35:33] How should you write website content to optimize for LLMs and AI search engines?
Answer / Description:
To optimize website content for LLMs, marketers should write using specific repetition of brand names, define clear conversational categories, and directly align page headlines with question-based queries.
While traditional SEO algorithms discourage repetitive brand phrasing to prioritize natural reading for humans, Elad Hefetz states that AI engines thrive on "specific repetition" to build vector associations between tokens. Instead of replacing your brand name or product name with pronouns like "it," "this," or "our tool," you must explicitly repeat the exact brand and feature names. For example, rather than writing "our wizard does X," you should write "the Airfleet GEO Wizard does X." This ensures the LLM's token-matching algorithms clearly associate the brand with the solution. Headlines should also be framed as natural questions (e.g., "What is the best B2B startup CRM?") to make it easy for LLMs to map answers to user queries.
[40:22] Does using LLMs to generate your website's content provide an advantage for AI search optimization?
Answer / Description:
Yes, using LLMs to generate or structure website content can provide an advantage for generative engine optimization (GEO) because machines naturally find it easier to read, parse, and analyze outputs formatted by other AI models.
While Elad Hefetz discourages using raw, unaltered AI content for human audiences because it often feels unnatural or dry, he notes that LLMs digest AI-structured copy exceptionally well. When combining clean, proprietary content inputs with an LLM-guided structuring process (such as a Retrieval-Augmented Generation or RAG system), the final text aligns better with the tokenization patterns and cognitive processing styles of modern search crawlers.
[41:42] How do on-page structure and layout elements improve AI content retrieval?
Answer / Description:
On-page structural elements like AI-dedicated summary boxes, structured lists, glossaries, and clean FAQ blocks improve AI content retrieval by lowering the processing energy (token computational cost) required for LLMs to parse and understand a webpage.
Elad Hefetz recommends adding a "TL;DR" (Too Long; Didn't Read) or an "AI Summary" box containing one to three condensed paragraphs near the top of long-form pages. Because LLMs assign the heaviest weight to the beginning of a document and can lose focus when analyzing overly wordy pages, these summaries prevent semantic dilution. Utilizing highly structured formats—such as bullet points, numbered lists, alphabetical glossaries, and explicit question-and-answer FAQ structures—helps search crawlers and AI systems convert page data into clean vector embeddings with minimal noise.
[44:35] What technical and file adjustments are required to make a website accessible to AI agents?
Answer / Description:
To make a website technically accessible to AI agents, webmasters must configure their robots.txt files to permit AI crawlers, verify that firewalls or CDNs are not blocking AI bots, and implement an llm.txt file as an AI-specific sitemap.
Technical SEO fundamentals map directly to technical GEO, according to Elad Hefetz. However, many IT, security, or legal teams accidentally or intentionally block AI crawlers via Web Application Firewalls (WAFs), Content Delivery Networks (CDNs), or restrictive rules in the robots.txt file. To facilitate better crawling, brands should implement the emerging llm.txt file standard. Functioning as a markdown-based sitemap built specifically for LLMs, the llm.txt file helps AI crawlers instantly identify which pages are highly authoritative, which ones are worth indexing, and how the site's overall content is structurally related.
[45:56] Is traditional SEO dead because of conversational AI engines and AI Overviews?
Answer / Description:
Traditional SEO is not dead because AI search engines still heavily rely on search indexes (like Bing) to source real-time data, and a vast portion of web searchers continue to use standard search interfaces.
Elad Hefetz argues that SEO remains highly relevant. When tools like ChatGPT perform a real-time web search (RAG), they pull data directly from Bing's search index. If your website is ranked in the top 20 organic results on Bing for the query generated by the AI, you have an exponentially higher chance of being integrated into the final conversational answer. Furthermore, traditional Google search volume still dwarfs AI search volume, meaning standard SEO practices are still critical for capturing current organic search demand while simultaneously laying the groundwork for GEO ingestion.
[50:14] How can brands balance content optimization for AI engines and human readers?
Answer / Description:
Brands can balance human-centric content and machine-centric optimization by using dedicated web pages for machine digestion—such as glossaries, FAQs, and AI summary modules—while serving human-friendly content on the main pages.
To prevent human-facing copy from reading as repetitive or awkward, Elad Hefetz advises creating specialized content repositories. Sections like alphabetical glossaries and deep technical FAQs are highly structured and keyword-dense, which machines love but humans rarely read in full. On mainstream landing and product pages, adding condensed "TL;DR" modules at the top or bottom of a page provides a clean vector for AI crawlers without disrupting the natural, engaging narrative structure written for human visitors. Hefetz also anticipates a future rise in middleware solutions that dynamically serve different, structured text variations to AI bots versus human web visitors.