Author: Kita Yohei Published: June 2, 2026
In GEO discussions, you'll often hear "primary source information is important." Most companies respond with "we don't have survey data" or "we're not a big enough company for that." I used to think the same thing. But after consistently observing AI responses, I started to realize it's not that simple.
What I Got Wrong at First
I'll be honest: I used to think primary source information meant large-scale surveys.
"We surveyed 1,000 people." "We published a report covering the entire industry." — I thought only things like this counted as primary sources, and that companies who couldn't reach that level had no business with primary source strategy.
But when I actually observed what gets cited in AI responses, it wasn't necessarily that at all.
What Is Primary Source Information?
Let me start with the definition. Primary source information is "information directly observed or collected by yourself." It's distinct from secondary information — interpreting, summarizing, or citing information someone else has compiled.
In other words, scale doesn't matter. A "10,000-person survey" is primary source information, but so is a "5-customer interview." What matters is whether you directly observed it.
→ What Is a Primary Source?
Primary Sources Are Everywhere
Looking around with this lens, you start to realize that most companies hold large amounts of primary source information they haven't noticed.
- Records from customer interviews and consultations
- Compiled inquiry content and frequently asked questions
- Usage data from your own tools or internal systems
- Performance figures from case studies
- Records of failures and improvements
- What your frontline staff has observed in the field
All of these are primary source information. And most are either "too obvious to think of as information worth sharing" or "buried somewhere inside the company."
An Observation I Ran Myself
Let me share something personal. For a period, I checked every week whether my X account appeared in Grok's guest mode search results.
Here's what I found: search for "AI search optimization" and it appears. Search for "GEO" and it doesn't.
That was an interesting observation for me. The same person, the same account — but whether it gets cited changes depending on the search query. In other words, AI isn't evaluating "what this person is an expert in" at the keyword level — how concepts are linked to a person determines whether they appear in results.
This observation isn't a large-scale survey. But it's something only I could have observed — which makes it legitimate primary source information.
AI Looks for Uniqueness, Not Scale
What I've come to feel from my observations is that AI is looking at "is this irreplaceable?" more than "how big is this?"
Averi.ai (citing BrightEdge GEO Benchmark Report 2025) reports that "even sites with domain ratings of 30–40 can achieve meaningful citation rates when they have genuine expertise and content depth." A 50-customer survey, if it's information that exists nowhere else, has value. Uniqueness — not scale — is the condition for AI citation.
→ Why Is Industry Research So Often Cited by AI?
Why Most Companies Have Primary Sources but Don't Share Them
This is my hypothesis, but I think there are three main reasons.
The first is "thinking it's obvious." The information a company observes and collects every day feels so natural to them that they don't realize "this is worth publishing."
The second is "not thinking of it as data." Spreadsheet numbers aren't the only data. What your frontline staff has observed in the field is also legitimate primary source information.
The third is "buried inside the company." The information exists — it just hasn't been shaped into content. It hasn't been published.
My Perspective
I think primary source information is something you "find" — not something you "create."
Most companies already have it. They just haven't noticed. Inquiry logs, customer interviews, frontline observations, your own tool's usage data — organizing these and publishing them is what I think the essence of a primary source strategy looks like.
My reason for starting the Grok observations wasn't grand. It was just "I was curious whether my account would show up." But accumulating that record gives it meaning as an observation of how AI behaves.
→ What Is Original Research?
→ What Is a Case Study?
Summary
Primary source information is not the same as large-scale surveys.
What AI evaluates is uniqueness. That's why I believe publishing "experiences, observations, and data that no competitor holds" is what matters.
If you feel like "we don't have data," look around inside your company first. Conversations with customers, patterns in your inquiries, what you feel in the field — that's already primary source information.
The next question is: "what should companies do when they genuinely have no data at all?" That's what I'll think through in the next article.
References
- Averi.ai / BrightEdge, "AI Search Citation Benchmarks 2026" (citing BrightEdge GEO Benchmark Report 2025: even DR 30–40 sites can achieve meaningful citation rates with genuine expertise and content depth)