Every few weeks somebody asks me the same question, and I still don’t have a clean answer for it.
“What kind of page are we supposed to build so the AI quotes us?”
There’s no shortage of advice out there. Publish listicles. Publish more blog posts. Build out the resource hub.
But whether any of that applies to your company is a completely separate question, and almost nobody stops to check.
Handing the same checklist to a B2B software vendor and a chain of cosmetic clinics always struck me as a bit lazy. (One size of uniform for the entire school, and we call that a dress code.)
So here’s a study that goes at the question with 25,000+ real citations pulled from five different AI engines. The answer flips by industry, which is the fun part.
Ninety Days, Five Engines, 25,337 Citations
The analysis comes from DeltaV Digital, an agency working in AI search (the study).
They ran it from April 14 to July 13, 2026 — 90 days. In that window they collected 21,075 AI answers and pulled 25,337 citations out of them.
The engines were ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode. Eight industries in all — B2B technology services, consumer automotive, higher education, medical aesthetics, and four more.
Two metrics do the work here:
- : what percent of all citations went to a given page type
- : how often the AI actually quoted a page, per time it fetched it
Citation rate is basically a batting average. The model showed up, read the page, and then either used it in the answer or walked away.
One thing to hold onto before the numbers: DeltaV sells AI search services, and the measurement runs on an AI visibility platform (Peec AI). So “you should invest in AI visibility” is the conclusion they’re set up to land on. Read with that in the back of your mind.
The Average Says “Articles And Lists”. The Hit Rate Doesn’t.
Start with all eight brands pooled together. Sorted by citation share, the top of the table looks exactly like the standard advice:
- Articles: 23.7%
- Listicles: 19.6%
- Product pages: 16.3%
So “write articles and lists” survives contact with the data, at least on this view.
But re-sort by citation rate — the batting average — and the picture rearranges itself.
The top slot goes to comparison pages, at a citation rate of 1.87, roughly 45% above the overall average.
And comparison pages account for just 4.1% of all citations.
That’s a strange pairing. The pages that get used most reliably are the pages almost nobody is making. Call it an empty shelf.
Break It Out By Industry And The Lead Changes Hands
Here’s the part that actually changes what you’d do on Monday.
Split the same data by industry, and the page type collecting the citations swaps cleanly from one to the next:
- B2B technology services: listicles, 61%
- Multi-location medical aesthetics: the homepage, 55%
- A health-information nonprofit: articles, 54%
- Higher education: degree and program pages, 53%
- Consumer automotive: articles plus listicles, 64% combined
DeltaV calls this an industry “citation fingerprint”, and the name fits: every industry leaves a different pattern.
A clinic group whose single most-cited asset is the homepage isn’t something I’d have predicted. But it means that before anybody spins up a content calendar, the higher-leverage move might just be fixing the front page.
Which makes the pooled average nobody’s actual pattern.
Some Industries Own Their Citations. One Owns Zero.
There’s a second number here that points straight at where the money should go.
The share of citations landing on the brand’s own domain splits wildly:
- Higher education: 74.7%
- Specialty food: 18.9%
- Consumer automotive: 7.4%
- B2B technology services: 0.0%
Higher ed is fighting on home turf. B2B tech services isn’t on the field.
In that B2B set, the most-cited domain was LinkedIn, with 736 citations. And Reddit turned up among the top citation sources for seven of the eight brands.
So “publish more on our own blog” is the wrong instinct for some categories. You’d want to go where the third parties already are.
One more distinction worth carrying around: getting read by an AI and getting cited by one are separate events. On the same health queries, WebMD gets fetched constantly and sits at a citation rate of 0.33. Mayo Clinic hits 1.79. The CDC, 1.85.
The model shows up, reads, and leaves. Retrieval and citation want separate columns in the report.
How Much Can Eight Brands Actually Tell You?
Now the part that keeps this honest.
The study covers eight brands total. One per industry.
So “B2B technology services is 61% listicles” is really one company’s observation, and treating it as the industry’s number would be a stretch. Eight data points don’t make eight benchmarks.
And like I said, the publisher sells AI search services.
Which is why I don’t think the takeaway is “B2B should write listicles”. Turning this into a new universal checklist just adds one more universal checklist to the pile — the exact problem the study documents.
Maybe I’m wrong about how far these particular percentages travel. But the narrower claim holds up fine: the page type that gets cited genuinely changes by industry, and now there’s data showing it. That’s the argument for measuring your own category instead of borrowing somebody else’s.
Conclusion: Measure Your Own Category Before You Copy Anyone’s Playbook
So when AI-visibility advice doesn’t seem to fit, the advice isn’t necessarily wrong. It’s an average being handed out as a prescription, in a market where the citation pattern moves by industry.
An industry where 74.7% of your citations sit on your own site and one where that number is 0.0% are not the same problem. Setting a budget without knowing which one you’re in is mostly guesswork.
Two things worth doing. First, get a repeatable read on which page types the AI cites in your category — not somebody else’s case studies, your own measurement. Second, publish one comparison page, since that shelf sat empty across every industry here: a 1.87 citation rate against 4.1% of supply.
Once you know the shape, what to build stops being an argument. Mass-producing content while you’re still guessing is the expensive version.
Same models, different industry, completely different move. I find that oddly satisfying.
Sources
- DeltaV Digital, “AI search citations study: what 25,000+ citations reveal”, 2026, deltavdigital.com (90-day window from April 14 to July 13, 2026; 21,075 AI answers collected and 25,337 citations extracted across ChatGPT, Perplexity, Gemini, Google AI Overviews and Google AI Mode. Eight brands across eight industries; for each brand the top 30 most-cited URLs and top 30 domains were classified, for 240 URLs and 240 domains in total. Citation share by page type: articles 23.7%, listicles 19.6%, product pages 16.3%. Citation rate led by comparison pages at 1.87, whose citation share was 4.1%. By industry: B2B technology services listicles 61%, medical aesthetics homepage 55%, health-information nonprofit articles 54%, higher education program pages 53%, consumer automotive articles plus listicles 64%. Share of citations on the brand’s own domain ranged from higher education 74.7% down to B2B technology services 0.0%. The most-cited domain for B2B technology services was LinkedIn with 736 citations; Reddit appeared among the top citation sources for seven of the eight brands. Citation rates on the same queries: WebMD 0.33, Mayo Clinic 1.79, CDC 1.85. Data collection ran on Peec AI. Limitations: eight brands, one per industry, so the figures are not industry benchmarks; the publisher is an agency selling AI search and SEO services.)
