Each industry gets cited differently by AI engines, and the winning page type is not the same across verticals. In a study published 13 July 2026 by Brandon Kidd of DeltaV Digital, covering 25,337 citations across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode, listicles captured about 61% of citations in B2B technology services, homepages about 55% for a local medical-aesthetics brand, and program pages about 53% in higher education. The same engines, over the same 90 days, cited a completely different mix of page types depending on the category being asked about.
That pattern has a name in the report: a citation fingerprint — the specific blend of page types an engine relies on when answering questions in a given vertical. This piece lays out the data, attributed and hedged, then explains why a generic "publish listicles and FAQs" checklist under-serves most brands. The companion how-to on matching your industry's fingerprint turns it into a workflow.
What is a citation fingerprint?
A citation fingerprint is the distinctive mix of page types AI engines cite when they answer questions in one industry. It is a per-vertical pattern, not a per-engine one: it describes what kind of page wins the citation (a listicle, a homepage, a product page, a program page), rather than which engine does the citing.
The DeltaV study measured it two ways: citation share (what percentage of an industry's total citations each page type earned) and citation rate (citations per retrieval, i.e. how often a fetched page was actually kept). The headline finding is that these blends diverge sharply between industries — so the "right" content format is a property of your category, not a universal rule.
Which page types get cited most overall?
Across all eight brands combined, three formats carried more than half of citations. The portfolio baseline, per the DeltaV data:
| Page type | Citation share (all industries) | Citation rate (cites per retrieval) |
|---|---|---|
| Articles | 23.7% | 1.43 |
| Listicles | 19.6% | 1.45 |
| Product pages | 16.3% | 1.22 |
| Comparison pages | 4.1% | 1.87 (highest) |
Source: Brandon Kidd, DeltaV Digital, "AI search citations study" (25,337 citations, 21,075 AI responses, 14 Apr–13 Jul 2026). Comparison pages were a small share of citations but the most efficient format — when one was retrieved, it was the most likely to be kept and cited. This corroborates the corpus view that neutral comparison content earns trust and that listicles are a heavily-cited format — but, as the next section shows, the averages hide most of the story.
How much does the cited page type change by industry?
A lot. This is the core finding. The dominant page type flips from vertical to vertical:
| Industry | Dominant cited page type(s) | Share |
|---|---|---|
| B2B technology services | Listicles | 61% |
| Consumer automotive | Articles + listicles | 64% combined |
| Medical aesthetics (local) | Homepages | 55% |
| Healthcare | Articles | 54% |
| Specialty food CPG | Listicles + homepages | 54% combined |
| Higher education | Program pages | 53% |
| B2B cybersecurity | Articles + how-to guides | 51% combined |
Source: DeltaV Digital, July 2026. The reading: a B2B software brand lives or dies by earning placement in third-party listicles, while a local medical-aesthetics practice is cited mostly through its own homepage, and a university through its program pages. Publishing a pile of listicles would do little for the local practice or the university. The same engines, the same 90 days, and three different answers to "what should we publish?"
Can your own website win the citation, or must it be earned?
This is the second under-discussed cut, and it is the one that decides your content budget. The DeltaV study reported each industry's own-domain citation share — how much of its cited content came from the brand's own site versus third-party sources:
| Industry | Own-domain citation share |
|---|---|
| Higher education | 74.7% |
| Specialty food CPG | 18.9% |
| Government scholarship | 11.5% |
| Healthcare nonprofit | 8.3% |
| B2B cybersecurity | 8.2% |
| Consumer automotive | 7.4% |
| Medical aesthetics | 3.0% |
| B2B technology services | 0.0% |
Source: DeltaV Digital, July 2026. In higher education, the brand was the citation — three-quarters of the answers pulled from its own domain. In B2B technology services, none of it did: every cited source was a third-party listicle, review site, or directory. That maps cleanly onto the owned-versus-earned distinction: some verticals reward publishing your own authoritative pages, and others force you to earn placement in independent lists because engines will not cite your site directly for those queries.
One consistent thread across all eight brands: Reddit appeared among the top-cited domains for seven of them, and LinkedIn was the single most-cited domain for the B2B technology-services brand (736 citations). That matches the corpus finding that Reddit's AI citations are real but concentrated on specific surfaces, and reinforces that user-generated and professional-network content sits in most industries' fingerprints whether or not a brand cultivates it.
How does this square with "listicles get cited" and other corpus findings?
It sharpens those findings rather than reversing them. Three reconciliations:
- Listicles still matter — but not universally. Why AI loves listicles holds on the average and in list-heavy verticals like B2B tech. The fingerprint data adds the caveat: in local services, healthcare, and higher education, other formats win, so "publish listicles" is category-dependent advice.
- This is a different axis from content edits. The Princeton GEO experiments tested what changes to a page lift citations (adding statistics, quotes, sources). The fingerprint data tests which page type gets cited in a vertical. You need both: the right format, then the evidence density inside it.
- This is page-type concentration, not domain concentration. It is distinct from how few domains AI engines concentrate citations on and from per-vertical brand concentration. Fingerprints describe formats; those describe which sites and brands recur. Keep the two separate when you plan.
What are the caveats on this data?
Several, and they are load-bearing:
- Eight brands, one per vertical. Each "industry fingerprint" is essentially a single brand's citation profile. Read the per-industry percentages as illustrations of how much variance exists between categories, not as audited benchmarks for the whole industry.
- Single-vendor, single-window. The data comes from one measurement pipeline over one 90-day window (Apr–Jul 2026). Engines change their retrieval behaviour, so treat the numbers as a dated snapshot and cite "DeltaV Digital, July 2026" with the date when you reuse a figure.
- Citation share is not causation. A format winning citations in a vertical does not prove that publishing more of it will earn you citations — the engines may be citing incumbents. Use the fingerprint to prioritise where to look, then measure your own results.
None of these undo the durable, corroborated point: the page types AI engines cite differ enough between industries that a one-size-fits-all content checklist is the wrong starting point.
Every industry has a distinct AI citation fingerprint. Copy the fingerprint of your category, not a generic best-practices checklist — the format that wins B2B software citations is not the one that wins for a local clinic or a university.
What should you actually do about it?
Work from your category's fingerprint outward:
- Find which page types get cited for your category's questions. Look at the actual AI answers for your buyers' queries and note the format of each cited source, per the companion how-to.
- Check your own-domain share. If engines rarely cite sites like yours for your queries, prioritise earned placement in independent lists; if they cite owned pages (as in higher ed), invest in your own extractable, well-structured pages.
- Match the format, then raise the evidence density inside it. Once you know the winning page type, apply the content edits that lift citations within it.
- Track it as a trend, per engine. Fingerprints shift as engines change, so watch your citation coverage and share of voice over time rather than acting on one snapshot.
The through-line is measurement: knowing which page types and sources AI engines cite for your category — and whether that includes your own domain — is exactly what Buffy Intel is built to show, engine by engine and snapshot over snapshot, so you can spend on the format your vertical actually rewards instead of a generic checklist.