Belief 6 of 7 from the VizzEx AI Citation Study mid-study findings
One of the sites in our study is a reliability engineering publisher. On Google’s AI Overviews, it was tracked on “what is FMEA?” for ten checks, through day 56 of the study. It lost all ten.
On the same site, on the same topic, it was tracked on “What are the main shortcomings of using RPN to prioritize risk in an FMEA?” It was cited at nine of the ten checks.
Same site, same subject, same weeks. The head term went to someone else every time. The question a practitioner would actually ask stayed with the site that answered it.
The belief
“Head terms are the prize worth chasing.”
What is a head term?
A head term is the short, broad search that names a whole topic, like “types of malware” or “what is FMEA.” It usually has the most searches and the most competition. SEO tools also call it a head keyword or a short-tail keyword.
Chasing them is the SEO habit carried into AI search: find the big keyword, build the best page for it, win it. Most AI-visibility studies and dashboards test exactly that, with prompts like “best CRM for mid-market,” and call it buyer discovery. If you want to understand what most AI-visibility dashboards actually track under the hood — and what they leave out — that’s worth reading before you trust the numbers.
Does ranking on Google still get you cited?
Ranking on Google for the keyword no longer gets you cited in its AI Overview the way it used to. Ahrefs found that only 38% of Google AI Overview citations come from pages ranking in the top 10, down from 76% in mid-2025. Their explanation is query fan-out: Google answers a long list of related searches, not just the one typed. Google has confirmed it does this. But Ahrefs didn’t test whether fan-out is what caused the drop, and couldn’t, because Google doesn’t show which related searches it ran.
We tested the belief itself: does the head term go to the site that chases it?
How we tested it
The VizzEx AI Citation Study tracks 10 live websites, including two famous marketing publishers we left untouched as a control group, on about 1,200 real buyer questions across Google AI Overviews, ChatGPT, Claude and Perplexity. One of the ten sites is ours, vizzex.ai. Every answer is stored, so we can see which site held each question at every check, and what the page that held it was.
Who holds the head terms
On the head terms, AI engines mostly cite video sites, discussion forums, encyclopedias and dictionaries, not competitors. Across one client’s questions alone, YouTube appeared 140 times among the sources holding the answer. After YouTube come Wikipedia, Reddit, the dictionaries and, on technical topics, the big platform vendors and standards bodies. In our Belief 5 post, on citation decay, the citations our sites lost with no content reason went to the same kind of source: NIST, Merriam-Webster, healthcare.gov, the large security vendors. We tracked the citations our sites lost with no content reason closely enough to see the pattern: the losses weren’t random, and the destinations weren’t either.
That’s fenced ground. An engine answering “what is a SOC?” wants a definition from a source it already relies on, a video or a discussion thread. A better blog post doesn’t change that.
On specific questions, blog posts take the platforms’ place
On the specific questions, the platforms mostly fade into the background, and blog posts and articles take their place. Through day 56, Claude cited YouTube, Reddit, LinkedIn or another big platform in 32% of its answers to questions of three words or fewer, and in 11% of its answers to questions of eight words or more. Over the same questions, blog posts and articles grew from 30% of the sources Claude cited to 51%. On Perplexity, the platforms stayed at about 8% of the sources whatever the length, while blog posts and articles grew from 22% to 41%.
93% of the wins on big topics were specific questions
The sites using VizzEx won on big topics anyway, just not on the head terms:
- 93% of their wins on big topics were specific questions, not the head term itself (196 of 210).
- Outside one site, they won 2 of the 16 head-term phrasings they tracked. The exception is the reliability engineering publisher, whose definitional depth won most of the rest.
One caveat: we tracked few head terms to begin with, so part of that 93% reflects what we chose to track. That means the 93% alone doesn’t prove the point. The rest of the evidence in this post does.
Winning doesn’t make you the only source
Winning a specific question on a big topic doesn’t make you the only answer, either. When the sites using VizzEx won one, they were almost never the only source:
| Engine | Won questions where the site was the only source cited |
|---|---|
| Google AI Overviews | 2 of 85 |
| Perplexity | 0 of 145 |
| Claude | 0 of 78 |
| ChatGPT | 5 of 10 |
ChatGPT cites few sources per answer, so when it cites you, you’re often the only one.
On ChatGPT, the question decides
ChatGPT made the split sharpest. Across the eight sites using VizzEx, through day 56:
| Question length (brand names excluded) | Questions tracked | Questions ChatGPT ever cited the site on |
|---|---|---|
| Three words or fewer | 86 | 1 |
| Eight words or more | 879 | 46 |
The one short question was a local search (“IT services” plus a town name), cited once in the study’s first week, on the homepage, and never again.
Same site, same engine, different question
The clearest case is one client, a managed security provider:
- 63 questions on big institutional topics (zero trust, firewalls, SIEM, penetration testing and other high-volume security topics): ChatGPT never cited it.
- The 315 questions its own blog posts answer, including its frameworks, its customer incidents and how it runs its service: ChatGPT cited it on 22.
Same site, same engine, same weeks. The difference wasn’t the company. It was the question.
Where NIST or Wikipedia already owns the answer, no amount of structure gets you in. Where only you can answer, you can win it. And as our data on what actually earns citations in AI search shows, domain authority doesn’t open the fenced ground — only owning the answer does.
“Best X” is a head term too
“Best X” questions look like buyer questions, so they’re what most AI-visibility dashboards track. In our data they behave like head terms. “Best” asks the engine to rank a whole category, and nothing in the question says what this buyer needs. So the engine answers the category.
Through day 56, the sites using VizzEx tracked four plain “best X” questions, such as “best managed security service provider for mid-size companies.” Across Google, ChatGPT, Claude and Perplexity, none of the sites was cited even once, in 111 checks. One site ranked fourth on Google for its “best” question and was still never cited. That’s the same pattern our control group confirmed: ranking on Google for the keyword no longer gets you cited in AI answers. The answers to the managed security question named nine of the client’s direct competitors. One of them was in every answer, on all three engines, from day 21 on.
Add the buyer’s need, and the result changes
Add what the buyer needs, and the result changes. A healthcare client tracks “best [provider] for member digital experience?” and “which [provider] has the best support for switching mid-year without disruption?” Through day 56:
- Google cited the site at 7 of 9 and 8 of 10 checks.
- Perplexity cited it at 7 of 10 and 9 of 10, and named the client in 3 and 4 of those answers.
“For mid-size companies” didn’t help the security question. A company size isn’t a need. “Without disruption” is.
That’s six questions, so take it as a pattern to check on your own questions, not a law. But it points the same way as everything else here: don’t chase “best X.” Answer the need inside it.
If the head term is fenced, what gets in?
We took every position the sites in the study held and sorted them by what the page actually is. We found eight kinds of page, across seven sites. Four of them get the company named in the answer. Four get it cited and leave it anonymous.
| Kind of page | Cited | Named | |
|---|---|---|---|
| 1 | You next to a named rival | Yes | Yes |
| 2 | A term you coined, defined on your own page | Yes | Yes |
| 3 | A framework or model you named | Yes | Yes |
| 4 | A named customer’s story | Yes | Yes |
| 5 | A category you defined | Yes | Sometimes |
| 6 | The conversation around a head term | Yes | No |
| 7 | Your buyer’s situational question | Yes | No |
| 8 | A narrow method your practitioners use | Yes | No |
Held positions sorted by hand at mid-study, with the numbers below re-checked through day 56. “Held” means cited in Google’s AI Overview at every check where one appeared.
The four that get you named
In each of these, the company’s name is part of what the question is about.
1. You next to a named rival
This one is ours, so take it with the disclosure from Belief 4: vizzex.ai is one of the sites in the study. We track nine questions that name VizzEx and a rival: Link Whisper, MarketMuse, InLinks.
- Held at every Google check, 9 of 9.
- Cited on 26 of 27 checks across the other AI engines, and named on all 27.
- The rival’s own site appeared in the sources once in 27.
Nobody but us has written a page that compares both, so we hold the position by default. The day a rival writes theirs, they’re in the set too. (Belief 2 showed the other side of this page: whether the answer says our name depends on whose name is in the question.) If you haven’t read that one yet, Belief 2 is the right place to start before going further here.
2. A term you coined, defined on your own page, with your name on it
One of our partners, Genoo, a marketing automation company, has coined terms: the self-segmentation loop, the 42-day memory window, the Goldilocks data equation. We tracked 10 questions about them. None of the questions contains the word Genoo.
- ChatGPT cited the site on 9 of the 10 and said Genoo on all 9.
- On Perplexity, the answers named Genoo on 1 of the questions. Four days after one small change to the site, they named it on 5.
Through day 56, the one term without a home on the site was the exception. AI engines used its idea with no link and no name: absorbed. More on that in Belief 7.
3. A framework or model you named
A client, a managed security provider, named its own operational resilience framework.
- The question about it was held at every Google check, 6 of 6, and cited and named on all three of the other AI engines.
- The same client’s partnership maturity model: held 8 of 8, named on two engines.
- Genoo’s ExpertMesh Strategy: held at both of its Google checks, and cited and named on all three engines.
- Our own Symmetry Gate: held 6 of 6, named on all three.
When the question is about your framework, the answer can’t describe it without saying whose it is. A name isn’t a guarantee of a citation, though. The same security client’s other framework missed Google’s first two checks before holding from day 28. It’s named on all three engines, but only one of them cites the page.
4. A named customer’s story
The same client tracks four questions about one enterprise customer. Three were held at every check, and the fourth at 5 of 6. That fourth one, “how did the provider help that customer recover from its network crisis?”, was cited and named on all three engines. A second story, a zero-day exploit the client handled without charging its customer, was named on all three.
A story with your customer in it is something only you can tell, and the answer has to say who told it.
The four that get you cited, without your name
These four get the site cited, but the answer rarely says whose page it is.
5. A category you defined
This one is ours too. VizzEx coined “horizontal content analysis”: looking at how all the posts on a blog relate to each other, instead of examining one post at a time. People do search for it, and vizzex.ai ranks first on Google for it. We track “What is horizontal content analysis?” Through day 56:
- Google’s AI Overview cited vizzex.ai at every check where it showed an answer, 9 of 9.
- Perplexity cited vizzex.ai at all 10 checks and never named VizzEx.
- Claude cited it at 6 of 9 and never named VizzEx.
- ChatGPT never cited it.
The category travels. The company behind it doesn’t always.
6. The conversation around a head term
This one shows up on six sites in six industries, because it has the same shape everywhere:
| Site | The head term | A specific question on the same topic |
|---|---|---|
| Managed security | “types of malware”: lost 2 of 2 (tracked from day 49) | “How does encrypted malware get past antivirus?”: held 11 of 11 |
| Reliability engineering | “what is FMEA”: lost 10 of 10 | “What are the main shortcomings of using RPN to prioritize risk in an FMEA?”: cited at 9 of 10 |
| Healthcare | “deductible health insurance”: lost 9 of 9 | “What does implementation look like when switching PBMs?”: held 9 of 9 |
| Local government IT | “virtual CISO Illinois”: lost 10 of 10 | “Why do former employees still have system access after leaving a municipality?”: cited at 8 of 10 |
| Marketing automation (Genoo) | “communication cadence”: lost 11 of 11 | “Does the Rule of Seven mean every nurturing sequence needs exactly seven emails?”: cited at 9 of 11 |
| vizzex.ai (ours) | “citation half-life”: lost 8 of 8 | “What hidden content patterns does horizontal blog analysis reveal?”: held 6 of 6, cited on all three engines |
Google AI Overviews, every check through day 56 where an AI Overview appeared.
You don’t win the head term. You win the specific questions your buyer asks about that topic.
7. Your buyer’s situational question
These name a role in a situation. For example:
- A security provider’s questions on managed detection and response for a HIPAA-covered organization: held 6 of 6.
- The same provider’s business case for consolidating on Microsoft E5: held 7 of 7.
- A managed IT firm for local governments, on email spoofing in municipalities: cited at 7 of 8 checks.
- A managed IT provider’s question on what a CFO can do to keep a cyber insurance claim from being denied: cited on two of the three engines.
These are the questions on your buyer’s mind, in their words. Know your ideal customer and their buying journey, and write for that.
8. A narrow method your practitioners use
The reliability engineering publisher held 15 of its 120 tracked questions at every check through day 56: detection rankings, FMEA causes, detection controls versus recommended actions, pre-control chart rules, and more. That made it the third most-held site in the study. The answers didn’t say its name on any of them, except one of the two questions that already contain it.
The split
The four named kinds put your name inside the thing being asked about: next to a rival, inside a term, a framework or a story. The answer can’t describe it without saying whose it is.
The four cited kinds are your buyer’s own questions, specific enough that the site that answers them holds the citation for weeks. But nothing on the page binds the name to the answer, so the answer says “according to one source.”
As of day 56 of the study, only three of the eight sites using VizzEx had written any of the four kinds of page that get a company named: vizzex.ai, Genoo and the security client. The other five were cited, but only on the kinds that leave them anonymous, and their names almost never appeared in those answers.
What doesn’t get in
- A comparison of two things that aren’t you, like pen testing versus bug bounties. Pages like that were named in 0.6% of the answers that cited them.
- A head term an institution owns. The security client tracks 63 questions on big topics like zero trust, firewalls and SIEM, where standards bodies and the large vendors hold the answer. ChatGPT never cited the client on any of them.
- Marketing pages, on questions without your brand in them. They were only 8% of the citations we measured there. Marketing pages get cited when the question already names you.
Verdict: Did not survive
You don’t win the head term. You win the specific questions your buyer asks about that topic.
Nobody wins “cybersecurity.” The winners hold the hundreds of specific questions a buyer asks on the way there, and the ground nobody else can answer because it’s theirs: their expertise, their Unique Information Delta (UID).
Write the specific question only your buyer asks, and you get cited. Put your name on the thing, and you get named.
Watch Belief 6: Mid-Study Findings Webinar Segment
Kim Albee and Carolyn Holzman walked through this belief live in the Mid-Study Findings webinar, including all eight kinds of page. Watch the Belief 6 segment here, or the full hour for all seven beliefs.
What you can do Today
Three steps you can start on today, using the questions you already track.
1. Sort your question list three ways
- An institution already answers this. Stop funding it.
- Only we can answer this. That’s your ground.
- Our name could be in this: a rival, a term, a framework, a customer story. That’s the ground that names you.
2. Count which of the eight kinds you already have
As of day 56 of our study, only three of the eight sites had any of the four named kinds. Pick one of the four named kinds you don’t have yet, and write that page: your product next to a named rival, a framework you already use with a name, a customer story you can tell.
3. Put your name where the idea is defined
On the page that defines your term or framework, and at its first mention on the posts that teach it, say whose it is (“Genoo’s self-segmentation loop”) and link to the definition.
Want to learn to connect those pages with semantic relationship links? Our free quick course, What are Semantic Relationships and Why do they Matter?, walks through it.
What’s next
Belief 6 is one of seven things everyone says about AI search that we put on a scoreboard. One is left: whether more content is the answer.
The study’s final checkpoints run through the end of October. Then we re-score every row on the board against the final data, including the predictions we filed at mid-study, whichever way they go.
Kim Albee is co-founder of VizzEx with Carolyn Holzman. The VizzEx AI Citation Study is a controlled, 84-day study of how Google AI Overviews, ChatGPT, Claude and Perplexity choose, keep and name their sources.
Frequently Asked Questions
What is a head term and why doesn't chasing it work in AI search?
A head term is the short, broad search that names a whole topic, like 'types of malware' or 'what is FMEA.' It usually has the most searches and the most competition. On the head terms, AI engines mostly cite video sites, discussion forums, encyclopedias and dictionaries, not competitors. That's fenced ground. An engine answering 'what is a SOC?' wants a definition from a source it already relies on, a video or a discussion thread. A better blog post doesn't change that.
What kinds of content get a company named — not just cited — in AI search answers?
Four kinds of page get the company named in the answer: you next to a named rival, a term you coined defined on your own page, a framework or model you named, and a named customer's story. Four others get it cited and leave it anonymous: a category you defined, the conversation around a head term, your buyer's situational question, and a narrow method your practitioners use.
Do 'best X' questions work for getting cited in AI search?
'Best X' questions look like buyer questions, so they're what most AI-visibility dashboards track. In our data they behave like head terms. 'Best' asks the engine to rank a whole category, and nothing in the question says what this buyer needs. So the engine answers the category. Through day 56, the sites using VizzEx tracked four plain 'best X' questions. Across Google, ChatGPT, Claude and Perplexity, none of the sites was cited even once, in 111 checks.
How does question length affect whether ChatGPT cites a site?
Across the eight sites using VizzEx, through day 56, ChatGPT cited sites on 1 of 86 questions of three words or fewer, and on 46 of 879 questions of eight words or more. The one short question was a local search, cited once in the study's first week and never again.
Does ranking on Google still get a page cited in AI Overviews?
Ranking on Google for the keyword no longer gets you cited in its AI Overview the way it used to. Ahrefs found that only 38% of Google AI Overview citations come from pages ranking in the top 10, down from 76% in mid-2025. One site ranked fourth on Google for its 'best' question and was still never cited.