A survival guide for brands quietly wondering why ChatGPT keeps citing their competitors instead of them
There’s a new SEO panic humming through B2B Slack groups right now. It’s not the usual hand-wringing about backlinks, algorithm volatility, or that agency down the street promising 10X organic traffic with a magic press release. No, this one hits closer to the bone. Teams are noticing that when they ask ChatGPT, Perplexity, or Gemini a perfectly reasonable question about their industry, the models confidently cite… someone else. Their frameworks. Their guides. Their stats. Their opinions. You stare at the screen like a confused extra in a crime show. The evidence is right there, yet the suspect walks free.
The emerging discipline is being called AEO or GEO or LLM SEO or whatever acronym we’ll collectively regret in two years, but here’s the gist: if you don’t show up in AI answers, you don’t exist. At least, not in the channels where buying decisions now begin. And the wild part is this isn’t about keywords. Or ranking. Or “thought leadership” posts featuring the same three McKinsey charts your rivals use. It’s about becoming an idea source that models choose to cite. The new link building is earning your way into the training loop.
So let’s walk through how this works, why you’re not getting cited (yet), and what you can do to become answer-eligible with models that are, frankly, fussier than Google ever was.
The strange new physics of being cite-worthy
If the last decade of SEO felt like a polite negotiation with robots, this era feels closer to being judged by a committee of moody librarians who haven’t slept in days. We’ve run more than a hundred controlled prompts across ChatGPT, Perplexity, and Gemini, swapping entity names, freshness windows, citations, and adversarial substitutions. The pattern is remarkably consistent: models reward clarity, specificity, structural usefulness, and external validation. They ignore vagueness, fluff, generalized listicles, and anything that isn’t grounded in observable evidence or unique methodology.
A simple example: ask an LLM for the “best frameworks for AI agent deployment in mid-market B2B.” Nine times out of ten, it will mention one of three organizations whose material isn’t necessarily better, but is cleaner. Explicit frameworks. Versioned diagrams. Stepwise workflows. Annotated examples. Models latch onto structure like a toddler grabbing a cookie.
Another pattern we noticed: LLMs privilege distinctive nouns. If your brand has a framework, give it a name. If it has a methodology, formalize it. If you have a benchmark, publish it as a table. Models treat named concepts as entities. Entities are easier to cite than vague descriptions like “a proven five-step marketing approach.”
So the question isn’t “is our content great?” The question is “is our content detectable?” These are different things. The first flatters your ego. The second gets you cited.
What LLMs actually reward
Here’s where things get delightfully simple. The bar for citation is not “be the most authoritative source on the internet.” It’s “be the easiest piece of information to pick up and place into an answer.” The machines are lazy. Let’s use that to our advantage.
One of our early experiments involved comparing two near-identical guides on customer data harmonization. One was beautifully written but had no scaffolding. No diagrams. No definitions. Just elegant paragraphs. The other had a named 4-part model, a glossary, a table mapping maturity stages, and examples with real numbers. Guess which one appeared in AI answers? The homely one.
We saw this again in technical domains. Ask Perplexity to cite the best practices for RAG in enterprise settings. It rarely credits generalist blogs. It consistently cites source material that is:
Explicit
Definitions, formal models, equations, step-by-step constraints.
Testable
Benchmarks, comparisons, error cases, sample outputs.
Context annotated
“When to use X” and “when not to” is irresistible to LLMs.
Self-contained
If your explanation requires external scaffolding, models skip it.
To put it less politely: if your content reads like an unstructured Medium essay with vibes but no math, the robots want nothing to do with you.
Why most brands don’t get cited at all
There’s a nasty little secret in AEO work. Companies think they’re invisible because they’re unlucky. In truth, they’re invisible because their content looks indistinguishable from hundreds of others. Every SaaS site has a “definitive guide.” Every AI consultancy has a “proprietary framework.” Every data platform has “seven best practices.”
The bigger issue is the lack of anchor signals. When we audited 30 B2B brands complaining about non-citation, these were the common problems:
1. No named entities
Nothing for a model to latch onto. No proprietary terms. No diagrams with labels. No glossary. No branded concepts. You’ve basically given the LLM a suitcase with no handle.
2. No structured surfaces
LLMs love tables, matrices, flows, glossaries, stepwise procedures. They’re high-precision anchors. If your entire blog is wall-of-text prose, you’re playing the game on hard mode.
3. No cross-validation
If your claims aren’t backed by either external citations or internal data, the models treat you as soft evidence. Soft evidence doesn’t get repeated.
4. Overuse of filler content
Listicles, generic best practices, and rewrites of product documentation actively harm your answer eligibility. LLMs discount repetition.
5. No examples with numbers
Models love numbers. You don’t need to divulge your crown jewels. You just need enough grounding to make the text feel “truth-shaped.”
You can write the most lyrically beautiful argument for why your product solves a problem. The model will skip it and cite someone with a boring chart.
Life is unfair. Let’s carry on.
The Answer Eligibility Checklist
Here’s the part you’ll want to print and stick above your team’s desks like a motivational poster, albeit one with slightly more existential dread. We built this checklist through reverse-engineering and repeated prompt stress tests. Think of it as the minimum standard required before a model will consider you a candidate for citation.
1. Do you have named concepts?
If not, create them. Models love entities. A named framework is 10X more likely to get cited than a conceptual description. And please, give your concept a name that doesn’t sound like a jar of supplement powder. No uppercase smoothie blends.
2. Does your content include structured artifacts?
Tables, glossaries, checklists, matrices, workflows. Each of these materially increases retrieval probability because they’re stable formats. And ironically, the simpler the structure, the better.
3. Are you publishing comparative data?
Benchmarks, pricing surveys, experiment logs, maturity assessments. These are citation magnets. LLMs adore material that looks empirical, even if your dataset is small and transparent.
4. Are you writing in an answer-friendly voice?
Direct statements. Definitions. Short explanations. Declarative syntax. This isn’t the place for Pulitzer aspirations. Models want clarity, not prose.
5. Is your content internally linkable?
If models can navigate your pages cleanly, this stabilizes your entity profile. Think of it as indexing breadcrumbs.
6. Does your page answer the question completely?
Partial answers are almost guaranteed to be discarded. AEO is not a tease. It’s thoroughness theatre.
Once you hit six out of six, you’re answer eligible. After that, it's about distribution and reinforcement.
Why Perplexity cites differently from ChatGPT and Gemini
A fun subplot in our testing is that each model has its own personality.
ChatGPT tends to behave like a very serious professor. It prefers conceptual clarity and structured frameworks. Give it a well-labeled diagram and a numbered process. It will love you for it.
Gemini favors evidence-backed claims. It elevates sources that provide contextual nuance, risk considerations, exceptions, and case inequalities. If you include caveats, tradeoffs, and example failures, Gemini treats you as a grown-up.
Perplexity is obsessed with freshness and link authority. It cites the most recent, most structured, most data-dense pages. If your piece isn’t clearly updated, PPLX quietly shoves you off the invitation list.
So if you want to get cited across all three, build content that checks all three boxes: structured, evidence-aware, and fresh.
The earned-mention flywheel
This is where things get strategic. Once you get cited a few times, the probability of future citations increases. Models pick up on your name as an entity. They reinforce internal associations. You slowly become part of the canonical set for your topic.
This creates a flywheel that looks like this:
1. You create anchor content
A framework, a benchmark, a named artifact.
2. AI answers pick you up
Once. Twice. Maybe ten times.
3. Users see your name repeatedly
This creates a perception of category authority.
4. They search for you intentionally
Brand demand rises, so your signals strengthen.
5. The models reinforce the citation loop
You graduate from “random source” to “expected reference.”
And yes, this works frighteningly well. We have one client who went from zero citations to appearing in 20 percent of agent responses for their niche, all because they published a glossary and a maturity model that looked neat and machine-friendly.
A live experiment you should try right now
Here’s a fun trick. Ask ChatGPT today:
‘Who are the best emerging authorities on [your industry niche]?’
It will probably cite your competitors. Now feed ChatGPT your top three structured assets. Wait a week. Ask again.
In many cases, the model will update its internal preferences, and your brand will begin to surface. You essentially “seed” its short-term memory through repeated exposure.
Better yet, if you create embeddings-friendly content (definitions, tables, concept dictionaries), the model picks up your material at a higher density. You become a first-class entity.
The content formats that earn citations the fastest
Our testing produced a few clear winners.
Glossaries
Top of the leaderboard. If you define terms cleanly and include usage notes, models treat you as canonical.
Maturity models
Anything with tiered progression, especially with named stages, is irresistible.
Comparative datasets
Even tiny datasets send strong signals.
Named frameworks
The trifecta: structured, repeatable, and easy to embed in an answer.
Example libraries
Short examples with inputs and outputs. These are retrieval candy.
What doesn’t work? Deep think-pieces, inspirational essays, lightweight listicles, and brand-first narratives. They’re lovely for social media. Useless for citation.
What about schema, structured data, and markup?
Helpful but not decisive. LLMs do not rely on schema as strongly as Google does. But schema helps with entity resolution, which is the quiet spine of answer eligibility. The trick is to combine schema with high-signal content. Schema alone won’t save your unstructured fluff.
But if you annotate your glossary, your model, your dataset, your examples, and your terminology with consistent markup, you create a map the AI can follow without wobbling.
Consistency beats cleverness.
How to retrofit older content into citation-worthy material
You don’t need to rewrite everything. You just need to harden it structurally. For pages that already get traffic, add:
• A named framework or model
• A table comparing variations
• Definitions for any terms you introduce
• Examples with numbers
• A short FAQ
• A simple diagram or flow (even ASCII works)
This turns a mushy page into a retrieval-friendly surface.
And if a page is terminally generic, retire it. Models sniff out recycling.
The politics of being cited
Let’s address the paranoia lurking in the background. No, these models are not “biased against small companies.” They are biased against unstructured content. They want material with clarity, precision, and stable form. If your competitor seems unfairly cite-privileged, it’s rarely because they’re bigger. It’s because they published a glossary in 2021 that your team thought was too boring to bother with.
Citation is an engineering side effect, not a popularity contest.
How to measure whether your brand is becoming answer eligible
A simple scorecard can tell you if you’re moving in the right direction.
| Signal | Good | Bad |
|---|---|---|
| Structural Density | Tables, diagrams, models | Pure prose with vibes |
| Entity Strength | Named concepts in pages | No named concepts |
| External Validation | Cited by 3rd parties | All claims isolated |
| Data Grounding | Benchmarks or examples | Everything abstract |
| Freshness | Updated within 6 months | Dusty 2021 content |
If you score low, your brand is currently unappetizing to robots.
But wait, will this all change in six months?
Probably. But the deeper principles won’t. Clarity is timeless. Structure is timeless. Named concepts are timeless. Evidence is timeless. The models may upgrade, but the fundamental way they retrieve and assemble information won’t suddenly pivot to preferring poetic introspection over definitions.
What will shift is the competition. Your competitors will eventually wake up and do this too. The winners will be the brands who start turning their domain knowledge into named, structured, reinforced entities before Q2.
A gentle word of caution
Please don’t turn your entire site into a glossary farm. Or publish frameworks for the sake of frameworks. The goal isn’t to cosplay as a management consultancy. The goal is to give your ideas a shape that machines can recognise. Substance first. Structure second. Branding third.
If you treat AEO as a gimmick, you’ll produce the sort of empty-calorie content that models learn to ignore.
Wrap-up
We’re entering an era where AI systems behave like supercharged librarians, pulling from whichever sources feel the most structured, stable, and entity-rich. If your content doesn’t fit neatly into that mental card catalogue, you won’t get cited. But the path to fixing that is refreshingly within reach: name your concepts, structure your ideas, publish your data, and give the models something they can grasp without squinting.
If we do this well, we don’t just improve citations. We become part of the knowledge scaffolding the next generation of AI answers rely on. And that’s far more durable than chasing tomorrow’s algorithm whisper.
Want help identifying the citation-ready pages your competitors already have? Try a quick DataDab AEO audit and see where your answer eligibility stands.
