The New Rules of AI Optimization How to Drive Traffic in the Age of ChatGPT

In this session, David Berkowitz and Elad Hefetz explore how AI platforms like ChatGPT and Google AI Overviews are reshaping digital marketing and SEO.

They discuss strategies, measurements, and technical insights for optimizing website content and structure to meet the new standards required by AI, while also addressing the implications for brand visibility and traffic.

0:05 – David:
“Welcome back everyone and welcome to another edition of AI Insiders by AI Marketers Guild. It’s great to have you joining us today. I’m excited to dive into one of the most pressing topics—AI optimization—and introduce our guest, Elad from Airfleet, who has some fascinating insights to share.”

1:10 – Elad:
“Thank you, David. It’s a pleasure to be here in this amazing forum. I want to share an interesting insight and even show my screen. I’m using Ahrefs—a tool for SEO keyword research, ranking, and reporting—to discuss a trend they call ‘the great decoupling,’ where clicks are down even though impressions are up, especially in Google AI overview views.”

2:11 – Elad:
“This specifically addresses Google AI overview views, not ChatGPT. For our tech B2B customers, impressions rise while clickthrough ratios drop. More impressions should naturally lead to more clicks; however, if the wrong audience is seeing the content or if the answer is provided directly by AI, there is little motivation to visit the website.”

3:14 – Elad:
“AI overviews simply use your content to deliver an answer, thus making a website visit redundant. Top-of-funnel queries will vanish. If users obtain a complete answer from AI or ChatGPT, there’s no reason for them to click through to your website.”

4:21 – Elad:
“ChatGPT works similarly—only on steroids. It not only summarizes content but also prompts for the next action. For example, if you ask, ‘what should I do about my problem?’ ChatGPT may suggest a solution, summarize relevant products, and even compare features, leading to an endless interactive conversation. This comprehensive answer makes visiting a website unnecessary.”

5:23 – Elad:
“That’s the situation we face. I have a full four-hour presentation on optimizing for AI, but I’ll walk you through a short version focusing on five key concepts. Our AGI optimization framework covers strategy, prompt formulation, monitoring, technical website adjustments, and content optimization.”

6:27 – Elad:
“Let’s begin with strategy. The first aspect is prompting—how do you craft the right prompts? Next, you monitor visibility and AI traffic, understand ChatGPT’s behaviors, and then focus on technical and content changes on your website to get noticed and recommended.”

7:30 – David:
“Before we dive deeper, I’d like to discuss keyword research. How many of you are actively doing SEO and producing content on a weekly basis? Knowing your ICP, personas, pain points, and relevant queries is essential. Without this basic research, optimization won’t succeed.”

8:37 – David:
“Now, let’s consider this: if we want to do keyword research for SEO, we have the tools. But for ChatGPT, we lack dedicated tools to understand what prompts people are using. How would you determine which prompt to optimize for? Please share your thoughts in the chat.”

9:42 – Elad:
“A good suggestion came through the chat: use your existing online community and tools. Just remember, ChatGPT might provide an answer that appeals to what you want to hear—it doesn’t reveal the full truth, as there is no prompt database built into it.”

10:48 – David:
“Great, another point raised: if you optimize correctly, what should you expect from ChatGPT? More engagement, higher conversion rates, and better organic ranking—even though its default response omits website links.”

11:54 – Elad:
“Exactly. ChatGPT’s initial answer doesn’t include external links because it relies on its internal dataset. Only if the dataset lacks an answer does it perform an external query, causing citations to appear.”

12:59 – Elad:
“Typically, most users won’t enable web search by default. External queries happen when you ask local questions (e.g., ‘recommend X near me’), seek new information, or use niche prompts that the internal dataset doesn’t cover.”

14:02 – David:
“What do you think users will do next when they receive a list without links—say, for the top five AI marketing communities? Chances are they will leave the session, copy the names, and then search on Google or LinkedIn.”

15:08 – Elad:
“Right. Once their initial query is answered without a clickable result, they’ll likely copy and paste the names into a search engine. Thus, while citation counts matter, what we really want is a high ranking that boosts brand and referral traffic.”

16:08 – Elad:
“Over time, you’ll notice an increase in organic brand traffic, referral traffic from AI tools, and even visits from AI bots. I recommend adding a self-attribution question on your website—ask visitors how they heard about you, listing options such as ChatGPT, Gemini, or Perplexity.”

17:17 – David:
“Just to be clear, if a user asks for the top five AI communities and doesn’t see AMG, what should they do? They might simply copy the names and search on LinkedIn or Google.”

18:23 – Elad:
“Exactly. Now, let’s discuss how to understand your prompts. Start by converting your targeted keywords into potential prompts. I use a keywords-to-prompts converter, a Reddit scroller for top-subreddit questions, and even analyze FAQs, social conversations, and transcribed calls to discover what people ask in their own words.”

19:29 – Elad:
“Once you compile a huge list of potential prompts, use AI tools to consolidate them into the top 30–50 recurring, high-intent queries relevant to your personas. Knowing exactly what prompt to optimize for is crucial to success.”

20:29 – Elad:
“Next, establish monitoring before you optimize. Determine where your brand ranks per prompt—does it appear as number one, three, or five? Monitor website citations, referral traffic from AI, and organic brand growth. This benchmarking is essential.”

21:33 – Elad:
“Keep in mind: ChatGPT’s responses vary. To test objectively, never ask while logged in—use an incognito window. Some tools simulate queries with different contexts and locations, allowing you to analyze ChatGPT’s statistical responses.”

22:39 – Elad:
“Note that ChatGPT’s first response always comes from its internal dataset, with no external citations, because its dataset is updated only every six to twelve months. For instance, if your branding changes after the latest dataset update (June 2024), ChatGPT won’t reflect your new identity unless an external search is triggered.”

23:45 – Elad:
“This means that before the next dataset update, you must optimize thoroughly for AI. Otherwise, your brand may only appear when external searches occur.”

24:55 – Elad:
“Another factor to consider is the third-party websites that influence ChatGPT responses. These vary by industry—Wikipedia and Google My Business are key examples. For your brand to be recognized, ensure you’re listed on these authoritative platforms.”

25:56 – Elad:
“Depending on your industry, other influential sites might include established media or review websites like G2, Capterra, or industry-specific indices. Presence on these sites boosts the likelihood of ChatGPT referencing your brand.”

27:04 – Elad:
“Even social media plays a role; ChatGPT often pulls from Reddit and LinkedIn, rather than Meta or X. Understanding these influence channels is crucial when optimizing for AI.”

28:09 – Elad:
“This third-party influence applies to both the training dataset and real-time queries. When someone asks, for example, for the top five CRM platforms for startups, ChatGPT may first list popular choices and then offer to compare features—using content from your website and competitors if available.”

29:17 – Elad:
“If your website content is missing or irrelevant, ChatGPT might pull information from competitor pages or, worse, fabricate details. Ensuring your content is accurate and prominent in the AI’s dataset is essential.”

31:22 – David:
“Let me ask a question: In your experience, what aspect of on-page optimization matters most for ChatGPT and similar generative AI platforms? Is it website content structure, schema markup, or something else? Please share.”

32:34 – Ken:
“From my perspective, it initially appears as if the solution requires a complete CMS overhaul, a rip-and-replace approach. How do you enter relationships with clients who feel burdened by the idea of migrating platforms?”

33:32 – Elad:
“Our approach isn’t about replacing your existing CMS. We work as web agency consultants, integrating optimization techniques that work with nearly every CMS—WordPress, Webflow, HubSpot CMS, etc. The key is that the heavy lifting is external to your website’s underlying platform.”

34:40 – Elad:
“Core to our strategy is ensuring that external mentions, PR placements, guest posts, and citations across authoritative sites help ChatGPT notice your brand. Your website alone represents only a tiny piece of the internet from the AI’s perspective.”

35:44 – Elad:
“Let me show you a slide from my presentation… [Slide shown]. This slide outlines the concept of writing for machines rather than for humans when it comes to AI optimization.”

36:50 – Elad:
“For SEO, we write for humans and use natural language, but ChatGPT simply matches the next word based on data. Therefore, using specific repetition is crucial. For example, consistently using your brand’s exact name or product label helps the AI reliably recognize your brand.”

37:57 – Elad:
“If you refer to your product by its proper name—say, ‘Go Wizard’ instead of a nickname—the AI can make the connection more accurately. In other words, be specific and repeat key terms consistently across your site and external mentions.”

39:01 – Elad:
“This concept also applies when generating content via LMS. Machine-generated content tends to be token-friendly and more easily parsed by AI. However, remember that overly “machine-sounding” content might not engage humans as effectively.”

40:10 – Ken:
“If you use LMS to generate content, does that provide a real advantage? I’ve seen some results indicating an increase in traffic when companies adopt this approach.”

41:13 – Elad:
“Indeed, machine-generated content is more digestible for AI because it’s crafted in the language the AI expects. I advise against using it solely for human audiences—balance is key. For AI-specific pages, structured and concise summaries work very well.”

42:22 – Elad:
“Here are some quick tips: When you have a full blog post or page, create a short summary specifically for AI. A TL;DR box with one to three paragraphs can help the AI quickly grasp your content. Structure your content with bullet points, lists, glossaries, and FAQs to keep it token-friendly.”

43:24 – Elad:
“Consistency is critical. Use natural language while ensuring that AI sees specific repetition of key terms. This not only helps with context but also increases the likelihood that your website will be recommended.”

44:35 – Elad:
“Let’s talk technical for a moment. Standard technical SEO practices apply for AI optimization as well. Ensure your site allows AI agents to crawl it—don’t block them with a firewall or misconfigured CDN settings. Additionally, implement an LLMTXT file, a new standard acting as a sitemap for AI.”

45:35 – Elad:
“When the next dataset update occurs, the AI must quickly recognize what your website is about. An LLMTXT helps guide the AI to the most relevant pages and signals.”

46:37 – Elad:
“Some argue SEO is dead because ChatGPT relies on Bing for external searches. But even if you’re ranked in the top 20 Bing results, your content can mix with internal data to determine ChatGPT’s answer. The opportunity to rank in ChatGPT results is exponential.”

47:45 – David:
“I realize I’ve talked extensively already. Before we move to questions, I want to underline that while niche content might see less traffic from AI’s zeroclick searches, overall SEO still holds importance—especially as Google remains the primary search tool for most users.”

48:53 – Ken:
“Thanks for the great presentation, Elad. I also have a question regarding the recent Cloudflare ‘Crawl Framework’ paper. It seems to suggest changes in how AI might access content. What’s your take on this?”

49:59 – Elad:
“From my longer presentation, I noted that Cloudflare’s approach appears to have a specific agenda. I don’t believe it will stick. While Cloudflare is strong, OpenAI and Google—backed by massive ecosystems—will ultimately set the standard. On another note, some say that longtail, niche content sites will suffer from zeroclick AI searches, and those sites must adjust their strategies accordingly.”

51:04 – Audience:
“There’s another question regarding the balance between optimizing for human users and for AI. How do you recommend managing that?”

52:12 – Elad:
“In the near future, middleware solutions will likely serve different content for AI and humans. For now, I recommend having dedicated pages for AI—FAQs, glossaries, and tailored summaries—so that human users can bypass heavily technical content while AI benefits from the structured information.”

53:11 – Audience:
“Additionally, I’ve compared Google’s organic results with ChatGPT and Gemini outcomes. There seems to be a significant difference in rankings and recommendations. Any comments on that?”

54:12 – Elad:
“ChatGPT blends data from its internal dataset, Bing queries, and contextual factors. When it queries Bing, it doesn’t use the same parameters as a regular user, so even ranking high on Bing may not guarantee inclusion in ChatGPT’s answer. The engine mixes multiple sources based on context, which explains the differences.”

55:17 – Elad:
“If your website isn’t included in the internal dataset or the top tier of Bing results, you essentially have zero chance of being mentioned in ChatGPT’s answers. It’s all about ensuring your brand is recognized across authoritative platforms.”


56:21 – Audience:
“What about adopting an LLMT.XT file or similar technology? Should companies implement it now?”

57:28 – Elad:
“Yes. While not all LLM providers have publicly adopted the LLMTXT standard yet, there’s no reason not to implement it now. It’s straightforward and may give you an edge when these standards are widely embraced. Also, integrating this with your overall optimization process is key.”

57:28 – David (Closing):
“Thank you all for joining us today. I learned a lot, and I’m sure many of you did as well. Feel free to connect on LinkedIn for more insights and follow us for future sessions. Have a great evening, and I look forward to continuing this conversation soon.”