Search used to follow a simple pipeline. Crawled, rendered, indexed, ranked. That model doesn't hold anymore. AI search runs on a different pipeline entirely, and most brands are still optimizing for the wrong one.
From Crawled to Retrieved
Indig's framework replaces the old pipeline with three new stages: retrieved, cited, and trusted.
- Retrieved: An LLM pulls your content into consideration, either from training data or live web retrieval (RAG).
- Cited: A small slice of retrieved content actually gets linked to as a source.
- Trusted: The final and most important stage, because trust is what turns a citation into a customer.
Here's the misconception Indig runs into constantly: being crawled does not mean being cited. Getting pulled into an LLM's search just puts you in a pool of possible sources. Only a fraction of that pool makes it into the actual answer.
The Ghost Citation Problem
Getting cited doesn't guarantee you get mentioned by name either. Indig calls this the ghost citation problem. A brand can be the literal source behind an AI answer and still never appear in it.
And mentions are what matter. As Indig put it, mentions "are really what users see as a sort of recommendation at the end of the day, and that influences their purchase behavior the most."
This ratio between citation and mention isn't consistent across platforms. Gemini cites few sources but mentions a lot of brands. ChatGPT cites heavily but mentions fewer brands by comparison. Brands tracking a single aggregate score across all LLMs are missing this entirely.
Why Trust Beats Position
Indig ran a user behavior study with roughly 50 US adults shopping for high-consideration items like laptops and washing machines. When AI tools returned a shortlist, users picked the top result 75% of the time.
The exception was trust. If a user recognized a trusted brand anywhere on that shortlist, position stopped mattering. As Indig said, "If they trust it, they will always prefer it."
That single finding reframes the whole goal of AI search. It's not just about showing up. It's about showing up as a brand people already recognize and trust before they ever open the chat window.
Topic Clusters Drive 261% Growth in AI Search Results for Cruise Line
Prompts Are Longer, and Search Volume Isn't the Point Anymore
When an LLM doesn't have enough information in training data, it runs its own searches behind the scenes. These are called fan-out queries, and Indig defines them simply: "fan out queries is essentially the searches that LLMs conduct when they use web search."
The catch is that these queries often carry zero human search volume. Brands built entire content strategies around keyword volume data. That data doesn't capture what bots are actually searching for on a user's behalf.
Indig also points to a deeper shift happening in how people search. Prompts today run three to ten times longer than a typical search query, and each one carries a user's history and context with it. Rank tracking doesn't capture that anymore. As Indig put it, "prompt tracking is closer to polling than to rank tracking." Two people can type the same words and see different answers, so the old idea of a single, trackable position is gone.
His answer isn't more data. It's more focus on audience. Brands need to relearn "classic brand surveys, polling, focus groups, interviews," to understand not just who their audience is, but how that audience already perceives them.
Original Data Is the New Backlink
If there's one lever that consistently moves the needle, it's original research. Indig's own studies found that pages with 15 or more unique data points score far higher on information gain, the measure of what a piece of content teaches a reader that nothing else does.
For brands wondering where to start, Indig lays out a ladder:
- Top of the ladder: Product usage data. What can you learn about how your customers actually engage with your product?
- One rung down: Market data. Scraped, combined, or aggregated data that's still uniquely yours.
- Most accessible, but easiest to copy: Surveys and polls.
The mistake most brands make isn't a lack of data. It's the order of operations. As Indig put it, "the biggest mistake that I've seen brands make when it comes to data storytelling is that they just look at what they have and then they spin a story around it." The better approach starts with the conversation happening in your market right now, then asks what data you can bring to it.
Third-Party Trust and the Death of Duplicate Content Fear
LLMs don't just trust your own site. They form consensus using social networks, entertainment platforms, publishers, affiliates. That means third-party authority matters, but it has to be segmented by topic, not treated as one blanket score.
This also means the old SEO instinct to avoid duplicate content at all costs no longer applies. Indig is direct about it: "we absolutely should repurpose across different platforms." Posting the same article to LinkedIn, for example, doesn't cannibalize your own site anymore. It reinforces the same message across the sources LLMs already trust.
LinkedIn in particular has become what Indig calls a fast lane for AI visibility. Text is easy for LLMs to parse, professional identity adds credibility, and B2B topical relevance runs deep on the platform. Brands publishing consistently under a named author are seeing results within weeks.
Stop Confusing Tactics With Strategy
A lot of what gets called AEO or GEO strategy is really just a list of tactics. Indig draws a clear line between the two: "The strategy distinctly describes a very specific problem, a reason for why that problem is important, and then a unique approach to solve that problem."
A keyword list isn't a strategy. A real strategy names the specific problem, explains why it matters to the business, and identifies a unique approach competitors can't easily copy.
How to Actually Measure This
The old click funnel, search, click, convert, doesn't hold in AI search. Users rarely click citations or links. The exception is AI Overviews, where classic search results still appear.
That breaks traditional click attribution entirely. The only method that still works, according to Indig, is self-reported attribution: simply asking customers how they found you after a purchase or signup. It isn't perfect. People forget, and self-reported data always carries some noise. But it's the clearest signal available. As Indig said, "whenever I do this with my clients, we see values of like up to 10% and sometimes more of net new customers coming through AI."
For brands still waiting for a cleaner dashboard before they invest, that number is worth sitting with. The pipeline has already changed. The only question is whether the strategy keeps up.
Resources:
Sign up for the Growth Memo Newsletter: https://www.growth-memo.com/
Check out Kevin’s work: https://www.kevin-indig.com/
Connect with Kevin on LinkedIn: https://www.linkedin.com/in/kevinindig
Connect with Paxton on LinkedIn: https://www.linkedin.com/in/paxtongray/
Looking for an agency that'll be worth the investment? 97th Floor creates custom, audience-first campaigns that drive pipeline and conversions. Get started here: https://97thfloor.com/lets-talk/.
About Kevin Indig: Kevin has led organic growth at Atlassian, G2, and Shopify, and now advises companies like Meta, Ramp, and Upwork on building cost-efficient acquisition channels.
He is the author of Growth Memo, a weekly newsletter read by 26,000+ leaders, and speaks globally on growth strategy.
