As Seen by AI webinar recording and transcript: how AI reads your brand and how to influence what it understands

This page shares an edited transcript from the webinar As Seen by AI, led by Marina Petrova, CEO and Co-Founder of Intentful. The transcript was lightly shortened and clarified, and the page also includes a recording of the session. The explanations and observations in this transcript reflect Intentful's best understanding of current AI system behavior as of October 2025. While no one truly knows the exact internal mechanisms of the AI systems, the summary is based on observed behavior and hands-on testing.

Marina Petrova opened the session by framing the topic around how AI systems interpret websites and, more broadly, how they understand brands. The webinar centered on what matters in October 2025, how AI reads actual websites submitted by volunteers, and how an agentic search demo can help a company or destination evaluate how AI understands its web presence. The session then moved into recommendations for discoverability that Marina and the Intentful team consider important right now.

The underlying shift is not only about search. AI is changing how customers, visitors, and users interact with information and content.

Intentful began in 2021 and has focused since its founding on building AI that understands brands, destinations, and companies. The company works globally with organizations in tourism and travel, performing arts, CPG/FMCG, telecommunications, agencies, and other expanding categories.

The guidance in the webinar comes from Intentful’s daily AI work. Its products are powered by AI search, so the recommendations are drawn from practice, ongoing testing, and direct product-building rather than from general online commentary or thought leadership summaries.

Grounded in practical expertise, not theory

Although search comes up throughout the webinar, Marina emphasized that the deeper transformation is about how information is used overall. The familiar topics — changes to search, traffic drops, and shifts in user behavior — are already widely understood. The bigger issue is how AI systems now gather, evaluate, and assemble knowledge.

AI responses draw from hundreds of sources

In the older model, optimization largely revolved around a company or brand’s main website. In the current AI environment, a single answer to a single query may be shaped by hundreds of sources, and sometimes more. Optimizing the website still matters, but it is no longer the whole picture. AI may consider many places where a brand is mentioned, and that becomes especially clear in live or recorded agentic search testing.

Marina noted that one of Google founders had referred to thousands of information sources in the analysis of a single query. Whether that was meant literally or as an illustration of scale, Intentful would not be surprised if AI systems do evaluate thousands of sources before producing an answer.

That shift creates a challenge beyond website optimization. If a brand does not actively shape the available information, AI will define the brand from whatever it can find, not from what the brand intended to communicate. This applies to content programs, PR campaigns, third-party mentions, and any other public source connected to the brand.

The simplified version is that brands now have two major goals: first, to be discoverable by AI systems, and second, to influence the story AI tells about them.

Goal One: Discoverability

During preparation for the webinar, Intentful communicated with organizations that had volunteered their websites for analysis. One destination and its agency later asked not to be used as an example because AI’s description of the destination did not match how they wanted to be represented. Marina used that example to underscore how much work can be involved in taking ownership of the story AI tells.

Content remains extremely important, but Marina argued that it is not the right starting point. The first priority is discoverability. Only after AI can access and understand the website does content begin to matter fully.

Most websites are not currently optimized for AI systems because they were designed for a different era and a different playbook. In many cases, their structure makes key content hard or impossible for AI to see. During the demo, examples showed that some content visible to humans was not visible to AI.

There are three core considerations.

First, the website must be open to AI bots and AI crawlers. Some organizations hesitate to allow AI bots because of hosting costs and because these crawlers behave differently from Google. Those concerns can be managed by setting rules around how often bots may access the site, but the first requirement is allowing them in at all. If AI crawlers cannot access the site, the content cannot be included in their knowledge bases.

Among the volunteer websites reviewed by Intentful, some blocked about half of the bots. For some companies based in the EU, websites were fully blocked from AI systems, meaning the systems reached the site but were not allowed to enter. In those cases, even excellent website content cannot be seen by AI.

When Marina referred to AI, she meant bots from OpenAI, Perplexity, Anthropic, and Google. Traditional SEO continues to matter for multiple reasons, but brands also need to optimize for discoverability by AI systems, not only traditional search.

Every website now has two audiences: the human audience and the AI system.

How AI reads a website differently from a person

The examples in the webinar came from Destination Marketing Organizations, DMOs, which promote destinations to visitors and support local communities. Marina described DMO websites as visually rich and often beautiful, with strong photography, color, style, and a clear sense of care and local connection.

For people, these websites can feel natural, inspiring, and well organized. But the same site may look very different to AI.

One detailed example came from the destination in Wisconsin called Oconomowoc. The website was typical of many destination sites, with information about things to do, places to stay, dining, and related visitor needs. For a person, the homepage appeared polished and engaging, with an embedded live video and strong visual inspiration.

For an AI system, the same homepage looked very different. What a person sees and what AI can read are not the same.

Another example focused on a Things to Do page with useful information for a DMO of that size, including beaches, fishing, water activities, boat rentals, theater, concerts, and sports. A person could recognize the effort involved in organizing and presenting all of that information. AI, however, could read only a portion of what was present.

Compared with many other DMO websites, that page was relatively strong; some sites expose only a paragraph or a few lines to AI while hiding most of the page’s useful information. Even so, the visible content was still only a fraction of what existed for human visitors. When an AI bot reaches a page and cannot find the relevant information within seconds or milliseconds, it moves on. If AI cannot see what it needs, it leaves, and the brand loses a chance to connect with a potential visitor.

Another example used a section called Water. In that case, AI could detect only the page title and then encountered a blank page. No meaningful page content appeared.

Across several volunteer destination websites, the same pattern appeared: apart from the header, AI could see little or nothing. That makes discoverability the essential first step. It ensures that the energy, budget, and effort spent on content can benefit both human visitors and organic AI discovery. Advertising can still drive traffic, but for organic visibility, pages cannot appear empty to AI.

Agentic Search demo in ChatGPT

Marina also demonstrated Agentic Search in ChatGPT using a pre-recorded video made the day before the webinar so the example would be current. The prompt was fictional: Marina was not actually speaking at the conference mentioned in the prompt, but because the search was run inside her ChatGPT Business account, the system already had some context about her. The prompt asked where to stay in a destination she had not visited before and requested recommendations for restaurants and related trip planning.

The key instruction was to activate agentic mode rather than using standard ChatGPT search. Agentic mode makes the model’s reasoning and decision process visible, which provides insight into how it chooses sources, evaluates information, and moves through the web.

The example again focused on Oconomowoc in Wisconsin. Marina recommended that teams try this for their own destination or business, ideally recording the session so they can pause, read the reasoning, and understand exactly how the system is interpreting their web presence.

What the Agentic Search demo revealed

• The model first chose to open the official DMO website and treated it as a credible local source for lodging information beyond TripAdvisor. This mattered because it showed the model distinguishing between official and unofficial websites, something earlier versions did not do as effectively.

• It reviewed the dining section and noticed that the main page included restaurant names but lacked additional information. It then followed a restaurant link to look for richer listings.

• When the site did not provide enough detail, the model returned to search results and began checking individual restaurant websites.

• It reviewed those restaurant sites and observed that, while the pages were simple, they included useful areas such as Our Story and Local Partners. The model opened those areas to gather more context.

• The whole search process was faster than previous versions, taking about 5 minutes rather than 10–15, while still collecting and combining information from multiple websites.

• The model then visited Travel Wisconsin, the state tourism website, but detected that automated access was blocked. It quickly stopped trying and shifted to open sources such as Twisted Fire.

• On the Twisted Fire website, the model observed that extensive image use reduced the amount of readable text. This was relevant to AI content accessibility: content matters, but technical accessibility, crawlability, and readable structure come first.

• The model attempted to use MapQuest and other sources and also returned intermittently to Travel Wisconsin despite the block, showing persistence in its reasoning process.

• It visited Yelp but could not extract content because of dynamic loading. The model recognized the limitation, left the page, and moved to the restaurant’s own website, Badger Burger.

• On the Badger Burger site, it encountered a long page, went to code line 888, and processed the content efficiently. This showed how quickly AI can scan and interpret large volumes of page content.

• It assessed phrases such as flavorful burger ingredients for traveler usefulness, then returned to the DMO website for additional context.

• The demo showed agentic AI behaving in a way that resembles human research: exploring, checking, comparing, and synthesizing. The difference is that it operates at a scale and pace people cannot match.

• The final output was a coherent summary assembled from the DMO, restaurant, and travel-related websites.

• Running a recorded agentic search for a business or destination can show what AI reads, what it misses, which sources it favors, and how it interprets the online presence overall.

Marina did not ask follow-up questions in the demo because the goal was to show the search process itself. A real user would likely continue with more specific follow-up questions about their visit.

She encouraged brands and destinations to observe how this happens with their own websites and to identify which third-party sources AI uses alongside the official site.

Three core steps for making sure your website appears in that source set

Marina then moved into how organizations can help ensure their own website is included when AI searches and reasons across sources.

The framework has three core steps. Some attendees had heard Marina present these before, and the guidance remains relevant.

Step 1: Discoverability

The first step should become a checklist for conversations with the web team or developers. Teams should review each point and ask the technical team to show whether the website is open to Perplexity, ChatGPT, Google, Apple, AI, and others. If access is restricted, the team should understand how and whether bots are still allowed in at defined intervals, such as once a day or three times a day, so they can collect information.

If a site receives significant traffic and hosting costs are a concern, teams should review the current robot.txt rules and what is allowed. A development team can show the setup, and a non-developer can still understand enough to determine whether the rules make sense.

Sitemaps need special attention. Marina said that in 95% of cases, sitemaps are a mess.

The first requirement is having a sitemap at all. Among the websites Intentful has reviewed, probably 35% do not have one.

Once the sitemap is accessible online, teams should spend time reviewing which pages are included. A sitemap is not only a website map; it also helps AI understand what content is available. Many sitemaps include outdated or irrelevant pages. In the older search environment, keeping old pages sometimes made sense because someone might land there and continue to another page. In the current environment, that logic is less useful. Content should be current and updated.

Teams should also confirm that all desired pages are included in the sitemap. Intentful often sees newly published pages that are absent from the sitemap. Unless traffic is driven through ads, those pages may never be found. Paid traffic remains one route, but organic discovery should not be overlooked.

After reviewing one sitemap, the process becomes clearer. Teams should clean it up, keep it current, and understand whether updates are manual, automated, or handled through some other process.

Website accessibility also matters. In this context, Marina meant accessibility standards and optimization according to accepted best practices as much as possible. Accessibility affects what AI can see.

The ChatGPT agentic reasoning demo showed an example involving dynamic rendering. Marina was not suggesting that teams remove all dynamic loading elements, which is often unrealistic. Instead, web teams should explore whether dynamic elements can be represented in static ways so AI can recognize that content exists. An agent may later click through dynamic content when searching for tickets, hotels, or dates, but it first needs to understand that the content is present and not see a blank page.

Step 2: Structure and signals

The second step is structure and signals. Much of this overlaps with traditional search and standard SEO hygiene.

If a page loads too slowly, AI may skip it and choose another source. PageSpeed Insights can help teams evaluate speed without needing deep technical knowledge. Google provides tools for this. If the site falls outside speed guidelines, the web team should identify improvements. This connects to the demo example where ChatGPT decided not to load content because images were too heavy.

Basic SEO tags still matter. Open graphs still matter. Structured data is especially important and should be near the top of the list after page loading time.

Structured data helps AI understand and read content. This is not limited to ChatGPT or Perplexity; structured data has long mattered for Google and will continue to matter.

Marina placed structured data alongside the sitemap in importance. Based on the organizations registered for the webinar, many attendees had events or other content types that require structured data. For teams unfamiliar with structured data, Google provides documentation on available types, and developers or web teams can add it relatively easily.

Semantic markup is also important. Teams should avoid unnecessary pagination that hides content behind additional clicks.

Step 3: Content interpretation

Content remains crucial. Keywords are no longer interpreted exactly as they were in traditional search, but they still serve as signals and should not be ignored completely.

Text content is necessary. AI can read images, view images, and watch videos, but text is still the first point of entry. Pages should include descriptive content that clearly explains the subject matter. The content should not rely only on marketing language; it should provide real information.

Freshness is also important. AI systems increasingly verify whether content is recent, and outdated references, such as material from 2016, still appear on websites.

The context window will keep expanding. Previously, AI analyzed smaller amounts of text to understand a topic. Now it can work with a volume of text closer to a book, though Marina noted that this is an oversimplification.

This creates two content streams. Brands should continue creating inspirational content for humans: imagery, video, color, and emotional connection still matter. At the same time, both humans and machines need informational content.

Informational content must be specific. If a team knows something but has not placed it on the website, AI will not know it. AI also makes it possible to move beyond assumptions about what users may want and instead identify actual intent.

Marina described this as an extraordinary moment because companies can now connect with every customer, user, and visitor in ways that were not possible before.

In a webinar held a couple of months earlier, Intentful reviewed a sample of 15,000 questions asked through the Intentful AI Assistant installed on destination websites. Marina did not repeat that full analysis but referenced it as a powerful way to see what people are actually interested in.

There is no PII attached, we do not collect personal information, and the information is anonymized. Even without knowing who the individual users are, the questions provide deep insight into what visitors are seeking when they come to a website.

For DMOs, the question is whether the website contains enough information for AI to answer visitor questions. This applies whether the answer comes through an AI assistant on the site, a ChatGPT bot visiting the site, or Google crawling the site.

In DMO contexts, people ask AI as though they are asking a trusted local. They ask about hours, parking, average costs, restaurants, and other practical details. They want information, not only marketing inspiration. For content strategy, that means being as detailed as possible. Marina acknowledged that this requires significant work, but if the goal is to help visitors engage and find answers, content needs to be updated.

The broader change goes beyond search. Companies are moving from speaking at customers through websites and ads to speaking with customers. This is the beginning of two-way communication.

Summary: knowledge and content

Once discoverability is solved and AI no longer sees a website as a blank page, AI works with the knowledge that has been published. Today, that often means breaking information into chunks. This will evolve as the context window grows, but currently AI splits text into smaller manageable units, interprets context, and then assembles relevant pieces into responses when users ask questions.

To be included in those responses, a brand needs to be discoverable and needs to make the right information available on its website. Professional knowledge should guide what gets published, but brands should also listen to customers.

A content strategy for humans and machines should balance inspiration with information. People ask in-the-moment questions and often expect AI to know what is happening right now.

For DMOs in particular, Marina emphasized the importance of updating member pages. A restaurant name alone is not enough. An event name alone is not enough. Pages should include as much useful information as possible, not just marketing copy. The brand should behave like a trusted guide or front desk resource, not only like a marketer. Current information matters increasingly because people do not need outdated content from several years ago.

The Four Buckets: a practical checklist

Marina organized the recommendations into four buckets:

1. Discoverability. This is the first priority. Teams can use examples like those in the webinar or run ChatGPT in agentic mode to see what AI knows about them.

2. Structure and signals. Websites should have structured data, a sitemap, and related technical signals in place. Web teams understand much of what needs to be done, though AI search may still be relatively new for many of them. The issue is not that they did something wrong; it is that the environment has changed and they need to engage with these requirements.

3. Content. Brands should provide both inspiration and practical information. They should examine real user questions, make content informative and substantial, and keep it fresh and accurate. The broader communication shift is from speaking at users to speaking with them.

4. Ongoing AI perception. Brands should keep watching how AI describes them across multiple sources, not only on their own website.

Intentful products for discoverability and engagement

Intentful offers two product categories for customers.

The first focuses on discoverability and on understanding how AI interprets a brand or destination. The program is called As Seen by AI and was recently launched. It is a 12-month program, and Intentful is already working with several DMOs. Marina emphasized that this is not a one-time effort because AI keeps changing, and because the discovery work and website changes required for AI visibility can be significant.

The second category focuses on using AI to engage users. Intentful’s AI Suite includes an AI assistant, generative response ads that let someone have a conversation with an ad and receive real-time replies, and an on-brand content tool powered by the same AI that understands the brand.

Marina closed by emphasizing that the environment will continue to change, but the foundations remain steady. A brand already optimized for SEO has a strong starting point for AI search optimization. The additional work begins with discoverability: making sure AI can see the website and helping the brand take back control.