Fatrank Podcast

James Dooley: Hi, today I’m joined with Ben from AISO, also known as Get AISO. Pleased to meet you Ben. Let’s jump straight in. I want to go deep into AI and LLMs because you have a lot of data through the tool you’ve built. What are people actually searching and asking in ChatGPT and other large language models?

Ben Akrout: This is a really good question. I think a lot of marketers are currently wondering whether they should really pay attention to AI search. If they do pay attention, they want to know whether it is worth investing serious money into it. A related question is what is actually new about search. Is it just different applications producing conversational answers while the same SEO and digital marketing rules still apply, or is something fundamentally changing?

One useful way to think about this is by looking at the user journey. The very first step in the search journey is understanding what people are actually asking AI systems. Are they asking the same things as traditional Google keywords but with a few extra words added around them? Are they asking the same things they used to ask on forums like Reddit, where conversations are naturally question and answer based?

That question is important because if people are asking the same things, then the strategies that work for SEO, digital PR and branding will probably still apply. But if the questions are very different, then marketers need to step back and ask whether their content is actually addressing these new types of questions.

At AISO we try to analyse what people are truly asking inside AI tools. A lot of companies today provide analytics showing charts of how you rank for prompts, which is useful. But we wanted to go a step earlier in the process and understand the prompts themselves.

There are some indirect methods people already use. For example you can analyse long tail queries in Google Search Console that might trigger AI Overviews. You can also analyse what people are asking on Reddit about your product category. But those are still approximations.

We wanted a more direct approach to understand real conversations happening in tools like ChatGPT and Gemini. Our previous startup actually worked on large scale data collection in real estate, so we tried to adapt some of those techniques to AI search.

One of the systems we built gives users certain benefits when they interact through our interface. For example they might get access to more advanced AI models or additional features compared to the free version of ChatGPT. Around ninety five percent of users actually use ChatGPT for free, which means they only access the limited models.

When users try stronger models for the first time they are often amazed by the difference. Because of this system we are able to observe anonymised conversations happening across models like ChatGPT and Gemini.

From that dataset we started seeing clear patterns. The first insight is that the prompts people ask AI are not simply Google keywords with extra words added. They are structurally different.

One thing that surprised us is that most AI conversations are actually one shot interactions. A user asks a question, receives an answer, and then leaves the AI interface. They might move to Google or another platform to click links and continue their research.

However the questions themselves are much more specific. People rarely type something short like “best pizza New York”. Instead they ask something more like “What is the best restaurant in New York for someone with a gluten allergy within a certain budget?”

The three most common categories in our dataset are travel, consumer electronics and beauty related topics like skincare or health adjustments.

In these categories people add a lot of personal context to their prompts. For example someone searching about restaurants may mention allergies, dietary restrictions, or a specific price range.

This level of specificity is very different from traditional search behaviour. People feel comfortable putting everything into one prompt because they expect the AI to process all the constraints for them.

Another interesting difference compared to social media is how personal the questions are. Even on anonymous platforms like Reddit people still limit what they reveal publicly. With AI tools people are far more open.

For example users may share health concerns, financial constraints or personal preferences that they would not post publicly. One of our users described it by saying that people sometimes share more with ChatGPT than with their romantic partners. This is particularly true in areas like beauty and health.

So AI queries are not only more detailed, they are also more personal and contextual.

Another interesting observation is something we call search density. Sometimes users do have multi step conversations where they gradually refine their request.

For example we analysed one conversation about planning a trip to Italy. The user started by asking about travel destinations. Then they asked about restaurants in a specific city. Then they asked about dietary requirements because they had celiac disease. Finally they asked for pricing and contact details.

In one conversation the user effectively performed fifteen different searches that would normally happen across multiple Google queries. The conversation moves from informational queries at the top of the funnel all the way down to transactional intent.

This compression of the funnel inside a single conversation is something completely new compared to traditional search.

James Dooley: So when you talk about search density, you are essentially saying that someone can move from the top of the funnel all the way down to the bottom of the funnel within one ChatGPT conversation. The queries themselves are longer and more conversational compared to traditional keyword searches.

From a content marketing perspective, what should marketers do differently? Should they create more semantic content, deeper content, or more pages to address these detailed prompts?

Ben Akrout: That is the million dollar question and there are several dimensions to the answer.

If we start with the customer journey again, the first step is understanding what people are asking. Once you understand those questions, you can start deciding which ones are strategically important.

There is always a trade off. The more specific a query is, the smaller the search volume. But those queries are often much closer to conversion.

Inside ChatGPT or Gemini, if your content answers a very specific question extremely well, it is more likely to be selected as a source or included directly in the AI generated answer.

That does not necessarily mean you should create ten thousand extremely narrow pages. Your content still needs to work for human readers and traditional SEO.

In many cases the first step is simply updating existing pages to cover slightly more specific questions. Many websites do not currently address these details because people were not searching for them on Google before.

Updating existing pages to include those additional specifics can already make a big difference.

The next step is thinking about distribution. AI systems do not only find information on your website. They also gather information from third party sources such as Reddit, LinkedIn, YouTube and media publications.

Classic digital PR channels with strong authority are still very important. AI models gather signals from across the web.

So after updating your website content, you can decide whether certain topics deserve broader distribution. If you have a strong competitive advantage on a very specific topic, you may want to push that information across multiple platforms.

For example if you are the only restaurant in New York offering a very specific type of pizza that meets a certain dietary requirement, it may be worth amplifying that message through different channels.

The key starting point though is making sure your website already answers the most important questions in a clear and relevant way.

James Dooley: That makes sense. Ben Akrout, it has been an absolute pleasure speaking with you. For anyone listening, this is the first part in a series where we explore how people are searching in ChatGPT, Gemini and other large language models.

In the next parts of the series we dive deeper into whether generative engine optimisation is really different from traditional SEO, we explore query fan out, and we also discuss personalisation within large language models.

Ben Akrout, thanks again for joining.

Ben Akrout: Thank you. It was a pleasure.