How to Build an Authority-Led AI Search Strategy

Your brand sits at the top of Google. The content is solid. The backlinks are strong. The technical foundations are clean.

Then someone asks ChatGPT the exact same question your page answers, and your brand does not exist.

This is the disconnect that SEO practitioners, agency teams, and in-house marketers are running into across industries right now. Traditional search performance and AI search visibility are not the same thing. They are measured by different systems, driven by different signals, and require different strategies.

In Episode 137 of SEOTalk Spaces, host Malhar Barai and co-host Parth Suba sat down with SEO consultant Gagan and practitioner Isha to break down what authority actually means in the context of AI search, why most of what the industry is selling around “AI SEO” is not producing results, and what practitioners working with real brands are doing instead.

The Problem: Ranking Well But Getting Cited Nowhere

Consider a brand that ranks for “best project management tool.” The page is optimized. The backlinks are in place. It performs well in traditional search. But when a user asks ChatGPT, Claude, or Gemini the same question, that brand is rarely or never mentioned.

As Gagan explained during the discussion, this gap is not a ranking problem. It is a recognition problem. “Ranking measures page relevance,” he said, “while citation measures recognized authority.” These are two completely different tests, and passing one does not guarantee passing the other.

The keywords can be right. The semantic scores can be strong. The internal linking can be clean. None of that determines whether an AI system will name your brand in an answer. The signals that drive traditional rankings and the signals that drive AI citations operate on different logic entirely.

This is not a theoretical concern. Practitioners on the ground are feeling it across markets. Gagan noted that agencies in Australia, the UK, and the US are all experiencing significant client churn. The hype around AI SEO is loud online, but the reality of selling it and delivering measurable results is far more difficult. Budgets for AI SEO exist in conversation, but when it comes to actual spending, most companies are holding back because they are not seeing returns.

One example from the discussion stood out. A company spent roughly $80,000 USD with an AI search visibility platform. Their team spent months monitoring dashboards. When leadership asked what the investment had delivered, the answer was a handful of additional citations in ChatGPT that could not be connected to any meaningful increase in revenue. The VPs were not impressed.

The gap between what the industry is promising around AI search and what brands are actually experiencing is real. And for practitioners trying to advise clients or build internal strategies, the first step is understanding why traditional authority does not automatically carry over.

Why Traditional Authority Doesn’t Transfer to AI Search

For years, authority in SEO has been built through a familiar set of signals. High-quality backlinks from trusted domains. Strong domain ratings. Relevant internal linking. Well-optimized content with the right semantic coverage. These signals told Google that a page deserved to rank, and for the most part, the system worked. If you built enough authority through these channels, you earned visibility.

AI search does not use the same test.

When ChatGPT, Gemini, or Claude generates an answer, it is not evaluating your page the way Google’s ranking algorithm does. It is not checking your backlink profile or measuring your domain authority score. It is asking a different question entirely: does this model recognize this brand as a credible source worth naming out loud?

That recognition comes primarily from the model’s training data, not from what is happening on the web right now. The model formed its understanding of your brand, your competitors, and your industry during its training process. If your brand was well-represented in the data the model trained on, you have a foundation of trust. If it was not, no amount of real-time web optimization will fully close that gap until the model is retrained.

This is a fundamentally different dynamic from traditional SEO. In traditional search, you can publish a strong page today and start earning rankings within weeks. In AI search, the model’s confidence in your brand was largely set during training, and the window to influence that confidence only reopens when the next training run happens.

As Gagan put it during the conversation, the measurement has shifted. It is no longer about how well you rank a page. It is about whether a machine trusts you enough to name you in an answer. That distinction changes the entire approach to building authority.

How LLMs Decide Which Brands to Trust and Cite

Understanding why traditional authority falls short is only useful if you also understand what AI systems are actually looking for. The mechanics behind LLM citation decisions are not widely discussed outside of technical circles, but the panellists broke them down in practical terms during the conversation.

Three factors shape whether your brand gets named in an AI-generated answer: what the model learned during training, how it constructs its search queries, and where it looks for information.

Training Data Sets the Baseline

Every large language model goes through a training process where it ingests massive amounts of text from across the web. During that process, the model builds an internal understanding of brands, topics, and the relationships between them. This is where the baseline of trust is established.

Gagan outlined a simple but useful framework for thinking about this. If your brand had an active web presence before 2023, it is likely included in the training data of most major models. That means the model has some foundational confidence in who you are, what you do, and where you fit in your market. If your brand launched after 2023, the situation is more difficult. The model may simply not know enough about you to feel confident citing you, regardless of what you are doing on the web right now.

This creates a meaningful split in strategy. For brands with existing model trust, the optimization window is open. Changes you make today, such as publishing on high-authority domains or strengthening your presence on key platforms, can produce results relatively quickly. Gagan shared an example where a brand with established trust published a single page on a high-quality domain, and within three hours, AI Overviews was citing the brand name. The model already trusted the brand. It just needed a strong signal in real-time search results to surface it.

For brands without that baseline, the timeline is different. Gagan described companies that had spent $20,000 or more on PR campaigns, placing articles across respected publications, but still could not get mentioned in AI answers. The content was on the web. The backlinks were in place. But the model itself had not developed enough confidence in the brand to include it. That confidence will only update when the next training run happens, which could be seven to twelve months away.

This does not mean newer brands should do nothing. Publishing content, building presence, and earning coverage now means that when the next training cycle arrives, the model will have more material to learn from. But it does mean that expectations around ROI need to be realistic. If the model does not know you yet, the returns are not immediate. They are an investment in the next version of the model’s understanding.

Fan-Out Queries Decide Where the Model Looks

When a user types a prompt into ChatGPT or a similar tool, the model does not simply search the web with that exact query. Instead, it generates a set of sub-queries, sometimes called fan-out queries, to gather information from multiple angles before assembling its answer.

Here is the part that most practitioners miss: those fan-out queries are shaped by the model’s internal knowledge, not by what is currently available on the web. The model decides what to search for based on what it already knows about the topic, the industry, and the brands associated with it.

Gagan gave a clear example. When a user searches for financial information on ChatGPT, the fan-out queries sometimes include specific brand names like Bankrate or Forbes. ChatGPT is not just searching the open web for financial content. It is actively seeking information from domains it has learned to associate with financial authority. Those brand names were not in the user’s prompt. The model added them based on its own training.

The same pattern applies to other query types. The panel discussed research showing that when ChatGPT handles queries with a user-generated content angle, it frequently appends “reddit” to its fan-out queries. The model has learned during training that Reddit is a strong source for certain types of information, so it actively steers its search toward that platform even when the user never mentioned Reddit.

For brands, this has a direct implication. If the model has not learned to associate your brand with a particular topic or query type, your brand will not appear in the fan-out queries. And if you are not in the fan-outs, your web content may never even be evaluated, no matter how well-optimized it is.

This is why Gagan described it as a structural advantage for established brands. They are already embedded in the model’s internal map of “who to ask about what.” Newer or less visible brands need to build that association over time through consistent, high-quality presence across the sources models are trained on.

Each AI Surface Pulls From Different Platforms

One of the most practical points from the discussion was that “AI search” is not a single channel. Different AI systems draw their information from different platforms, and a strategy that works for one surface may be completely irrelevant for another.

Gagan walked through the landscape platform by platform. Grok, built by xAI, pulls heavily from X (formerly Twitter). If your brand is not being discussed on X, you are close to invisible in Grok’s answers. For tech companies targeting Silicon Valley audiences or developer communities, this makes an active X presence a direct input to AI search visibility, not just a social media play.

Meta AI, which surfaces in Facebook and Instagram search, draws from posts, Reels, and content shared on those platforms. If your audience lives on Instagram and your brand’s presence there is thin, Meta AI has little to work with when users ask questions in your category.

Google’s AI Overviews lean heavily on YouTube. Brands investing in video content and building a YouTube presence have an advantage in Google’s AI-generated answers that text-only strategies cannot replicate.

And as the fan-out query discussion showed, ChatGPT appears to favour certain high-authority domains and Reddit for specific query types. Presence on those platforms feeds directly into ChatGPT’s ability to find and cite you.

Gagan made a broader point that stuck with the panel. Authority now operates on two layers. The first layer is your own assets: your domain, your content, your profiles. The second layer is what other people are saying about your brand on the platforms these models monitor. Earlier, collecting reviews on Google or on your own product pages was enough to signal credibility. Now, you need third-party voices across social platforms, forums, and independent publications.

This does not mean you need to be everywhere at once. It means you need to understand which AI surfaces matter for your audience, identify where those surfaces pull their data, and build your presence accordingly.

What Most Brands Are Getting Wrong

Understanding how AI search works is one thing. Avoiding the most common mistakes is another. The panel spent a significant part of the conversation on two patterns they are seeing repeatedly across the brands they work with: scaling AI content without checking whether it converts, and over-engineering technical SEO in pursuit of AI visibility.

Scaling AI Content Without Measuring Conversions

The temptation to scale content using AI tools is real. The economics are appealing. What used to take a writer five to fifteen working days can now be produced in minutes. For brands under pressure to grow their organic footprint, it feels like a shortcut that is too good to ignore.

Isha described this mindset directly. Finding a repeatable AI content process feels like hitting the jackpot. And once you have the jackpot in your hand, the instinct is to go all in, publishing dozens or hundreds of pieces without stopping to evaluate what they are actually producing.

The problem, as Gagan explained with a specific case study, is that scaled AI content often drives traffic without driving business results. He described a company that adopted a popular AI content scaling tool and began publishing aggressively around October of the previous year. By March, Google had penalized the domain. The company lost roughly 95% of its organic traffic, and the penalty did not just affect the AI-generated pages. It pulled down the rankings of their human-written content as well.

But the more revealing detail was in the sales data. Even before the penalty, when the AI-generated pages were still ranking and bringing in visitors, only about 5% of actual form submissions came from those pages. The remaining 95% of the pipeline was being generated by the human-written content. The scaled pages were producing traffic numbers that looked good in a report but were not converting visitors into customers.

Gagan’s question to the audience was pointed: the real issue is not whether you can scale AI content, but whether the people landing on those pages will ever become your customers. For most B2B and SaaS companies, the answer so far has been no.

Isha offered a more measured alternative. AI-assisted content production can work, but only when it is approached with discipline. That means knowing your ideal customer profile before you start. It means scraping real customer language from Reddit, support tickets, and sales conversations to inform what you create. It means producing in batches, evaluating each batch against performance data, and folding proprietary research or first-party data into every piece. And it means accepting that this process is not scalable for 100% of your content needs. Every batch should have a clear objective and a defined place in your overall marketing strategy.

The difference between the company that lost 95% of its traffic and a company that scales AI content successfully is not the tool. It is whether someone is asking “does this convert?” before asking “can we make more of it?”

Over-Engineering Technical SEO for AI Search

The second pattern the panel pushed back on was the growing pressure to adopt complex technical SEO strategies specifically for AI search. The industry conversation around ontologies, knowledge graph alignment, and machine-readable information architecture has intensified over the past year, and much of it is coming from well-funded companies with an incentive to make the problem sound more complicated than it is.

Gagan took a deliberately simple position. His approach to technical SEO for AI search is the same as his approach to technical SEO in general: get the basics right at the page level. If you have a product page, add the relevant product schema. If you have a service page, add service schema. If you have a local services page, add local schema. Make sure every page has proper headings and clean structure. Do this consistently across your entire domain, and the technical foundations add up.

He does not build strategies around ontologies. He does not design site architectures to plug into Google’s knowledge graph. He does not recommend that clients restructure their websites to be more “machine-readable” in some abstract sense. His reasoning is straightforward: if every page meets solid technical SEO standards, the domain-level picture takes care of itself without needing a top-down information architecture plan.

Isha reinforced the point with an analogy that captured it well. Schema is like a list of nutrients on a food packet. It labels the content, but it does not make the food healthy. The actual content on the page is what matters. Schema just helps machines read it.

The frustration both panellists shared was not with technical SEO itself but with the gap between what the industry is promoting and what most companies can actually execute. Gagan described a brand that hired someone with the title “AI SEO Architect” for two months. The result was nothing. The brand had been sold on a term that sounded impressive but had no defined scope, no clear deliverables, and no measurable outcome.

He gave another example that cut closer to the daily reality most practitioners face. Companies come to him wanting LLMs.txt files, advanced robots.txt configurations, and AI-specific technical audits. But when he tells them their product pages need seven or eight images from different angles so that customers can actually see the product, they say they do not have the budget. They want to optimize for AI search before they have finished optimizing for the humans already visiting their site.

Gagan’s view, which Isha shared, is that well-funded SEO companies are incentivized to over-engineer the conversation. They need to justify their funding rounds, so they introduce increasingly complex terminology and frameworks. But for day-to-day practitioners working with real brands, the number one problem is not AI traffic loss or a lack of ontology design. It is operations. It is getting basic changes implemented on the website. It is aligning with clients on strategy and actually pushing things live.

The advice from the panel was clear: do not chase AI SEO complexity before your fundamentals are solid. The brands that struggle most are not the ones without an AI search strategy. They are the ones that cannot get basic SEO implemented consistently.

What Actually Works: A Practitioner’s Approach

The panellists were critical of what the industry is overselling, but they were equally specific about what is actually producing results for the brands they work with. The common thread across their recommendations was practicality. None of what follows requires a six-figure tool budget or an AI SEO Architect. It requires clear thinking about where your brand stands, where your audience spends time, and what content you already have access to that competitors do not.

Assess Your Starting Position

Before spending anything on AI search optimization, figure out whether the models already trust your brand. This single assessment determines whether you are working with a short feedback loop or a long one.

The simplest way to do this is to run test prompts. Open ChatGPT, Gemini, Claude, and Grok. Ask the kinds of questions your customers would ask. Use variations: branded queries, category queries, comparison queries, problem-solving queries. Note whether your brand appears, how it is described, and whether it is recommended or just mentioned in passing.

If your brand was active on the web before 2023, you are likely in the training data of most major models. That means optimization efforts can produce results quickly. Gagan shared the example of a brand in this position that published a single article on a high-authority domain and saw AI Overviews citing the brand within three hours. The model already had confidence. It just needed fresh search results to pull from.

If your brand launched after 2023 or has a limited web footprint, the timeline looks different. The model may not have enough exposure to your brand to trust it yet. That does not mean the situation is hopeless. It means the work you do now, content, PR, platform presence, is an investment in the next training cycle rather than something that will pay off this month. Gagan estimated that brands in this position should plan for a seven to twelve month horizon before models start reflecting their authority.

Knowing which camp you fall into prevents the most expensive mistake in AI SEO right now: spending money on real-time optimization when the model has not built enough baseline trust to act on it.

Build Presence Where Each AI Surface Looks

Once you understand your starting position, the next step is to match your content and platform strategy to the AI surfaces your audience actually uses.

This is where the platform-specific nature of AI search becomes actionable. If your target audience is technical and active on X, building a strong presence there feeds directly into Grok visibility. If your audience turns to Google for research, YouTube content strengthens your chances of appearing in AI Overviews. If your audience uses ChatGPT for product research or recommendations, presence on Reddit and coverage on high-authority publisher sites increases the likelihood of being included in the model’s fan-out queries.

The key principle Gagan kept returning to was this: whatever channel you invest in, bring it back to your domain. If you are recording YouTube videos, generate transcripts and publish them on your website. If you are producing Instagram Reels, turn the best ones into blog posts or resource pages. If you are active on X, compile your threads into longer articles on your own site.

This is not an SEO play in the traditional sense. You are not creating these pages to rank in Google. You are creating them so that when the next training run happens and AI companies crawl the web again, your domain has a rich, consolidated record of everything your brand has published and discussed across platforms. Gagan was clear that brands should think of this as building an asset, not chasing rankings.

He pointed to Profound as an interesting example of channel-specific thinking. Profound’s primary marketing strategy is not SEO or AI SEO. It is events. They sponsor conferences, including their competitors’ events. When Ahrefs held a conference in New York, Profound had trucks driving around the surrounding streets with messaging that said “don’t use Ahrefs, use Profound instead.” That is a brand investing in the channel where its audience actually pays attention, rather than following a generic playbook.

The lesson is not that every brand should run guerrilla truck campaigns. It is that effective AI search strategy starts with knowing where your audience is, understanding which AI surfaces pull from those places, and building your presence there with intention.

Mine Your Internal Data for Content That Converts

The single most original insight from the discussion was about content sourcing. While much of the industry is focused on using AI to generate content from external research and keyword data, the panellists described a different approach: using internal company data as the raw material for content that is genuinely proprietary and far more likely to convert.

Gagan described three approaches that brands he works with are using right now.

The first is mining sales calls. One company processes roughly 400 sales calls per month, each running between 30 minutes and an hour. They extract the questions prospects ask, the objections they raise, and the pain points they describe. They strip out company-specific details and private information, then turn those conversations into bottom-of-funnel blog posts on their website. The content is not generic. It reflects the actual language real prospects use when evaluating the product. And it is working. Gagan noted that these pages are showing up in Google Discover, which suggests that Google is recognizing the content as fresh and genuinely useful.

The second approach comes from a B2B company that is studying how its own engineering team uses AI coding tools. The company has enterprise licenses for tools like Cursor, and they log the prompts their engineers use and the outputs they receive. They then use lower-cost models like Gemini Flash to analyse those logs, extract insights about common technical patterns and integration challenges, and create bottom-of-funnel pages about their product’s technical capabilities. They apply high-level masking to protect proprietary information before publishing. This process produces around 100 pages a month of detailed, engineering-grade content that no competitor can replicate because no competitor has access to the same internal data.

The third approach is simpler but equally valuable. Companies that track why customers cancel their subscriptions, through exit surveys or feedback forms, can analyse that data to identify recurring concerns. Those concerns become FAQs on product pages, addressing objections before the next potential customer encounters them. This is a direct line from churn data to conversion optimization, and it requires no external research or AI content generation tools.

Isha added that the same principle applies to other internal sources. Payment gateway data can reveal subscription trends. Support tickets surface recurring product questions. Performance marketing campaign data shows which messages resonate with which audiences. The more internal data you can feed into your content process, the more dots you can connect, and the more ideas you can generate that are rooted in what your customers actually care about rather than what a keyword tool suggests they might search for.

What makes these approaches powerful is not just the quality of the content. It is the fact that it is structurally difficult to copy. A competitor can replicate your blog post about “best practices for project management.” They cannot replicate content built from your sales conversations, your engineering team’s workflow, or your customer churn data.

Get the Basics Right Before Chasing AI SEO

The panel closed the tactical section with the point they agreed on most strongly: none of the above matters if your fundamentals are broken.

Every page on your site should have the correct schema for its content type. Product pages need product schema. Service pages need service schema. Local pages need local schema. Headings should follow a logical hierarchy. Content should be structured clearly. Images should be present and useful, not placeholder stock photos or missing entirely.

This is not new advice, and that is exactly the point. Gagan noted that the number one problem he hears from practitioners is not about AI search traffic, citation counts, or AI Overviews. It is about operations. It is about getting a client to approve and implement basic changes on their website. It is about aligning on a strategy and actually pushing updates live.

When companies skip these fundamentals and jump to AI SEO terminology, they create a gap between what they are talking about and what they are actually capable of executing. The brand that wants advanced AI search optimization but cannot budget for proper product photography is not ready for AI SEO. It is not even finished with regular SEO.

Gagan’s advice was direct. If every page on your domain meets solid technical SEO standards, the domain-level picture takes care of itself. You do not need to start from a top-down ontology or design your site to plug into a knowledge graph. You need to make sure the basics are done well, consistently, across every page. That foundation supports everything else, including AI search visibility.

The Uncomfortable Truth About AI SEO Right Now

The panellists did not shy away from the broader state of the industry, and the picture they painted was sobering.

There is a visible gap between what is happening in online conversations about AI SEO and what is happening when practitioners try to sell, deliver, and measure it. Gagan described it as feeling like two different countries. Online, the hype around AI SEO, GEO, and visibility tracking is enormous. Offline, agencies are losing clients, budgets are tight, and the companies that are spending on AI search tools are struggling to connect that investment to business outcomes.

This is not limited to one market. Gagan mentioned conversations with practitioners across Australia, the UK, and the US, all reporting the same pattern. Churn is high. Clients are asking hard questions about ROI. And the answers are not coming fast enough to keep engagements alive.

Part of the problem, as the panel saw it, is that parts of the industry are incentivized to make AI search sound more complicated and more urgent than it is. Well-funded companies need to justify their funding rounds, so they introduce new terminology, new frameworks, and new categories of work. Some of this is useful. Much of it creates confusion and sets expectations that practitioners cannot meet. The “AI SEO Architect” title that produced nothing in two months is not an isolated case. It is a symptom of an industry that is selling solutions before the problems are fully understood.

None of this means AI search is irrelevant. The shift in how people find information is real, and brands that ignore it entirely will fall behind. But the panellists were united on one point: the path forward is not through panic, hype-driven spending, or complex strategies layered on top of broken fundamentals.

It is through the work that has always mattered. Clear content. Strong technical foundations. Genuine authority earned through consistent presence and real expertise. The platforms and surfaces may have changed. The principles have not.

Frequently Asked Questions

Q1. Why does my brand rank on Google but never get mentioned in ChatGPT?

A1. Ranking measures how relevant your page is to a specific query. AI citation measures whether the model trusts your brand enough to name it in an answer. These are different systems applying different tests. A brand can perform well in one and be completely absent from the other because the signals each system relies on are fundamentally different.

Q2. Does my brand need to be in LLM training data to show up in AI search?

A2. It helps significantly. Brands that were active on the web before 2023 are likely included in the training data of most major models and have a baseline of trust. Brands that launched more recently may need to wait for the next training cycle, which could take seven to twelve months, before models develop enough confidence to cite them consistently.