The Winners and Losers of AI Search: What Kevin Indig’s Data Shows About Visibility

The factors that determine whether a brand gets noticed, trusted, cited, and chosen are broader than traditional SEO.

In an AirOps webinar, Kevin Indig walked through more than 20 months of data, over 1 million analyzed answers, and more than 100,000 citations to examine what separates the winners from the losers of AI search.

His conclusion was not that SEO is dead or that every brand needs to abandon its existing strategy. It was that SEO and AEO overlap, but they are not the same game. Search rank remains the foundation, while trust, audience ownership, first-party insight, and measurement now determine what happens beyond the ranking.

Top 5 Takeaways

1. Search Rank Still Predicts AI Visibility

The first shift is from thinking only about prompts to thinking about rank.

In research Kevin conducted with AirOps, ranking number one for a fan-out query earned 58% of citations. That was four times more than the result ranking number ten.

The implication is straightforward: teams trying to improve AI visibility should still start by looking at classic search performance. Are the pages that matter ranking in Google or Bing? Are they gaining or losing visibility? Are they competitive for the queries that models are likely to fan out into related searches?

AI search is not simply good SEO, but classic search is still the foundation. Kevin estimates that SEO and AEO overlap by roughly 70% to 80%. The remaining 20% to 30% is where the difference between showing up and being ignored often lives.

2. Trust Can Beat Position

Users carry some of their classic search behavior into AI search. When AI Mode presents a shortlist, users choose the first result roughly 74% of the time.

But trust is the stronger signal. When users recognize and trust a brand on the shortlist, its position matters much less. A trusted brand can appear lower in the list and still win the decision.

That changes how marketers should think about visibility. The goal is not only to rank highly in an answer. It is to become a brand users already know when the answer appears.

Kevin’s research found that users often filter results through two questions:

Do I know and trust the brand behind this result?
Could this answer my question?

The first question often comes before the second. That is why trust is becoming a shortcut through the search experience, and why brand building increasingly belongs inside the AI search strategy.

3. Visibility Matters More Than Referral Traffic

AI search is not primarily a traffic channel. It is a visibility and discovery channel.

When Google shows an AI Overview, clicks to classic search results decline by roughly 50%. At the same time, only about 1% of users click links inside the AI summary. Links are often used to validate or verify an answer, not necessarily to begin a traditional website session.

The same pattern appears in ChatGPT. Some links can generate hundreds of thousands of impressions while producing only a small number of clicks. Referral traffic is growing, but Kevin said AI referrals still typically represent only 1% to 5% of organic traffic.

That does not mean AI search lacks commercial value. One of Kevin’s clients lost almost 50% of organic traffic year over year while gaining nearly 20% in conversions. With self-reported attribution, AI search appeared to influence roughly 10% of new business.

The lesson is to measure what AI search actually changes. A brand can lose clicks while gaining consideration, direct visits, branded demand, and conversions from users who first encountered it in an AI answer.

4. The Citation Economy Is Concentrated

AI citations are not distributed evenly across every brand in a category.

For many topics, ChatGPT appears to draw from a relatively small pool of roughly 30 brands. That group captures about two-thirds of all citations, while the remaining citations are fragmented across a much larger set of sources.

This creates a sharp visibility gap. Brands do not necessarily need to dominate an entire category immediately, but they do need to become part of the consideration set. Once a brand enters that pool, moving higher becomes the next objective.

The same principle applies to trust signals outside a company’s own domain. Kevin pointed to YouTube as one of the strongest correlates of brand mentions in AI answers. AirOps research has also found that the majority of brand mentions in AI search come from off-site sources.

The practical strategy is a surround strategy. Vendor pages matter. So do review sites, publishers, competitors, communities, YouTube, and other third-party sources that models already use to form an opinion.

5. Commodity Content Is Losing to Discovery

AI is making content faster and cheaper to produce, which also means it is making generic content easier to replace.

How-to and what-is content has lost roughly 35% to 60% of clicks over the last 12 months, according to the research Kevin shared. Answer platforms such as Quora and Stack Overflow are also well below their traffic peaks.

Kevin described informational content as “the Yellow Pages of 2026.” If a page competes directly with the first paragraph a chatbot can generate, it is vulnerable.

The content that holds up is harder to synthesize:

Proprietary data
First-hand experience
Original research
Strong opinions
Real case studies
Product knowledge that helps someone make a decision

The bar for content quality is not simply higher because models write better. It is higher because the supply of adequate content is now effectively unlimited.

Best Practices and Key Learnings

Start With Search Rank and Citation Patterns

Teams should not throw out their existing search data. Classic rankings remain one of the strongest predictors of AI citations, especially for fan-out queries.

Start by identifying the commercial queries that matter most, then compare:

Where the brand ranks in classic search
Whether its pages are retrieved by AI systems
Whether those pages are actually cited
Which competitors consistently appear higher
Which sources are shaping the answer around the category

Prompt tracking is useful, but it should not be treated as a single source of truth. Kevin’s research found that when the same prompt is run three times consecutively, only 2.2% of citations remain consistent.

The better approach is to run important prompts repeatedly and measure what persists across multiple answers. The stable patterns are more meaningful than a single result.

Build Trust Beyond the Website

Trust is difficult to create through one page or one optimization. It accumulates across the places users and models encounter a brand.

That includes:

Review and comparison sites
Industry publications
YouTube videos and transcripts
Communities and discussion threads
Vendor and competitor pages
Original research and data
People who can explain the category with credibility

YouTube deserves particular attention. Kevin’s data showed that YouTube mentions had one of the strongest relationships with brand mentions in AI answers. Users also turn to YouTube and Reddit to validate AI-generated recommendations, with clicks to those platforms increasing when AI Overviews appear.

AI answers and human validation are working together. Brands need to be present in both parts of that journey.

Make On-Site Content Easier to Extract

Long-form content can still be valuable when it supports search rank and answers a meaningful commercial question. The issue is not length by itself. The issue is whether the page gives models and users something clear to use.

Kevin highlighted several on-site elements that showed measurable lift in his testing:

FAQs that answer specific questions directly
A TLDR or key takeaways section near the top
Clear headings with a descriptive hierarchy
Short, declarative sentences
Lists and comparison tables
Visible “last updated” dates

One client tripled its citation rate after adding last updated dates to a set of pages.

The writing style matters too. Kevin recommended simple, direct language with fewer nested sentences. The goal is not to flatten the content. It is to make the important claims easy to identify, understand, and quote.

Avoid Scaling Content Away From the Core Business

Kevin identified three common warning signs on sites that lose traffic after aggressively scaling AI content.

First, the topic selection drifts away from the company’s core expertise. When production is no longer the bottleneck, teams often expand into areas that are further removed from their product and what they are known for.

Second, the content develops the recognizable style of AI output: verbose, hyperbolic, flat, and interchangeable.

Third, the content becomes commodity content. It summarizes what is already known instead of adding experience, evidence, or a useful point of view.

AI-generated content is not automatically a problem. Kevin’s point was that well-made AI-assisted content can perform as well as human-generated content. The danger is using AI to create more of the same content without adding expertise or experience.

Use AI for Execution, Not Direction

Kevin described AI as a great intern but a terrible founder. It can analyze data, automate repetitive work, and increase the output of knowledgeable teams. It is much less reliable at choosing the right direction or developing a strategy from scratch.

His preferred approach is to use AI to build deterministic tools and workflows. Instead of asking AI to complete the entire job, use it to create a system that performs repeatable tasks with clearer rules and fewer opportunities for hallucination.

The distinction matters because AI capabilities are uneven. Models may be excellent at coding or data synthesis and surprisingly weak at a task that requires nuanced judgment. Subject matter expertise helps teams recognize where the model is useful and where it needs close review.

Measure AEO With Incrementality and Self-Reported Attribution

Traditional attribution is poorly suited to AI-assisted journeys. More than 70% of AI traffic can appear as direct traffic in analytics, and click-based reporting misses users who see a brand in an AI answer and later navigate directly to the company.

Self-reported attribution is one useful counter-lens. Ask new customers where they heard about the company, then connect those answers to CRM and pipeline data where possible.

Incrementality testing is another. Kevin shared an example where adding last updated dates to a group of pages tripled the citation rate. The test worked because the team could add the change, measure the lift, remove it, and see whether performance returned to baseline.

This is the standard marketers need more of: a measurable change, a defined cohort, and a result that can be compared against a baseline.

How to Put This Into Practice

Start with one commercial category where the brand already has some visibility.

Identify the priority prompts and run each one multiple times.
Compare classic search rank, retrieval, citations, and mentions.
Find pages that are being retrieved but not cited.
Improve extractability with direct answers, FAQs, tables, short sentences, and current dates.
Add proprietary information or first-hand experience that competitors cannot easily reproduce.
Map the third-party sources shaping the category and build a credible presence there.
Track citation persistence, branded demand, direct traffic, self-reported attribution, and pipeline together.

The work should not be limited to the website. On-site structure helps a model understand and use the content. Off-site trust helps determine whether the brand is considered in the first place.

The New AI Search Playbook

Kevin’s closing thesis was that AI changes humans more than it changes search. The important shifts are happening in how people allocate attention, decide what to trust, and use assistants that know their context.

That is why the next phase of AI search will be about more than rankings and prompts. It will be about becoming a trusted source, building an audience that can be reached directly, and creating information that models cannot easily synthesize from somewhere else.

The companies best positioned for that future will not simply publish more content. They will build stronger points of view, collect better first-party data, show up across the sources AI already trusts, and measure whether visibility is changing business outcomes.

As Kevin put it, the next 12 to 24 months may bring Media Wars 2.0, with companies investing in journalists, original research, video, and owned audiences to build a direct relationship with buyers.

AI search is still evolving. But the direction is becoming clearer: rank gets a brand into the conversation, trust helps it win, and owned audiences make the relationship durable.