We analyzed 25,000+ AI citations across 5 engines. Every industry has its own citation fingerprint.
DeltaV Digital’s 2026 AI search citations study. Study window: April 14 to July 13, 2026.
There is no single best content format for AI search. Between April 14 and July 13, 2026, we tracked 21,075 AI engine responses across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode, then analyzed 25,337 citations to the most-cited pages across 8 industries. No page type dominated. Every industry showed a distinct citation fingerprint instead: a specific mix of page types that AI engines rely on when they answer questions in that category.
The fingerprints are not subtle. Listicles captured 61% of citations in B2B technology services. Homepages captured 55% for a local services brand. Program pages captured 53% in higher education. The same engines, the same 90 days, and three completely different answers to “what should we publish?” The practical takeaway for marketers: copy the fingerprint of your industry, not a generic AI content best practices checklist.
This is DeltaV Digital’s first original research release on generative engine optimization (GEO), built entirely from first-party tracking data across our client portfolio. It is the dataset behind how we scope AI GEO engagements, and we are publishing it because most advice about “content AI loves” is written without citation data. This report is the data.
Key findings at a glance
- Listicles earned 61% of AI citations in B2B technology services. In healthcare, they earned 0%, and articles earned 54%.
- No single page type dominated portfolio-wide: articles earned 23.7% of citations, listicles 19.6%, and product pages 16.3%.
- Comparison pages posted the highest citation rate of any page type at 1.87 citations per retrieval, 45% above the portfolio average of 1.29, yet earned only 4.1% of total citations.
- AI engines retrieved Healthline, WebMD, and Medical News Today constantly for health queries, then cited them at just 0.55, 0.33, and 0.24 citations per retrieval. CDC (1.85), Mayo Clinic (1.79), and NIH (1.66) were both retrieved and cited.
- Own-domain citation share ranged from 0% for a B2B technology services firm to 74.7% for a university.
- Reddit appeared among the top cited domains for 7 of the 8 brands studied.
How we ran the study
We collected 21,075 AI engine responses (“chats”) daily between April 14 and July 13, 2026 via Peec AI, an AI search visibility platform. The dataset spans 8 client brands, deduplicated to one tracking project per organization, across consumer automotive, B2B cybersecurity, B2B technology services, higher education, healthcare nonprofit, specialty food CPG, government scholarship programs, and multi-location medical aesthetics.
Five engines were tracked: ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode.
The unit of analysis is the top 30 most-cited URLs and top 30 most-cited domains per brand over the window. That yields 240 URLs (25,337 citations) and 240 domains (67,657 citations) analyzed. Page types (listicle, article, product page, comparison, homepage, and so on) and domain types (editorial, corporate, institutional, UGC, reference, and so on) are Peec AI’s automated classifications.
Two metrics carry this report:
- Citation share is a category’s percentage of all citations in the analyzed set.
- Citation rate is citations earned per retrieval. Retrieval means the AI engine pulled the page while researching. Citation means it referenced the source in its answer. Citation rate is the closest thing to a trust metric: it separates pages AI reads from pages AI relies on.
Findings are based on the top 30 cited sources per brand and on prompt sets designed around each brand’s category, so results describe commercially relevant AI queries rather than all AI usage. Client identities are anonymized; results are reported as portfolio aggregates and industry categories.
Limitations. The top-30 cut captures the bulk of citations in a power-law distribution but not the long tail. Prompt sets skew commercially relevant and are not a random sample of all AI usage. Page and domain classifications are automated and imperfect at the margins; a manual spot check found them directionally reliable. 8 brands in 8 industries is a strong directional sample, not a census, so findings are presented as portfolio-level patterns. One brand (the healthcare nonprofit) began tracking in June 2026, so its window is roughly 30 days rather than 90.
How to cite this study: DeltaV Digital, “AI search citations study,” 2026. Analysis of 21,075 AI engine responses via Peec AI, April 14 to July 13, 2026.
The portfolio baseline: no single format wins AI search
23.7% of all analyzed citations went to articles. Listicles earned 19.6%, and product pages earned 16.3%. The next-closest category reached 11.8%. Across 25,337 citations, the leading format holds less than a quarter of the market.
| Page type | Citation share | Citation rate (citations per retrieval) |
|---|---|---|
| Article | 23.7% | 1.43 |
| Listicle | 19.6% | 1.45 |
| Product page | 16.3% | 1.22 |
| Other | 11.8% | 1.13 |
| Homepage | 10.8% | 1.11 |
| Category page | 5.2% | 1.06 |
| How-to guide | 4.5% | 1.32 |
| Comparison | 4.1% | 1.87 |
| Profile | 3.8% | 1.11 |
| Discussion | 0.5% | 1.32 |
This table looks flat on purpose. If you stopped reading here, you would conclude that articles win AI search by a modest margin and that a balanced content mix is the safe play. Both conclusions are wrong, and the portfolio average is why. It blends 8 industries whose citation patterns have almost nothing in common. The average is not the finding. The variance is.
Note one number in that table for later: comparison pages sit near the bottom on citation share at 4.1% and at the top on citation rate at 1.87. We return to that gap in the trust section.
The core finding: every industry has a citation fingerprint
61% of AI citations in B2B technology services went to listicles. 55% of citations for a multi-location medical aesthetics brand went to homepages. 53% of citations in higher education went to program pages. Each industry concentrates its citations in a different format, and the concentration is strong enough to dictate strategy.
| Industry | Winning page type | Share of citations |
|---|---|---|
| B2B technology services | Listicles (“top 10 firms” roundups) | 61% |
| Multi-location medical aesthetics (local services) | Homepages | 55% |
| Healthcare / medical information | Articles | 54% |
| Higher education | Program (product) pages | 53% |
| Consumer automotive | Articles (41%) + listicles (23%) | 64% combined |
| B2B cybersecurity compliance | Articles (38%) + how-to guides (13%) | 51% combined |
| Specialty food CPG | Listicles (28%) + homepages (26%) | 54% combined |
| Government scholarship programs | Listicles (25%) + directory-style pages | mixed |
Listicles earn 61% of AI citations in B2B services. In healthcare, they earn 0%.
B2B technology services: third-party roundups own the answer
In B2B services, AI answers to “who should I hire” queries are assembled almost entirely from third-party “top 10” roundups, LinkedIn, and Reddit. The single most-cited source set for the B2B technology services brand was roundup listicles, with citation rates of 3 to 7 citations per retrieval. One retrieval of a ranked list feeds many lines of the AI’s answer, which is why a single well-placed roundup outperforms a year of company blog posts in this vertical.
Local services: AI cites the homepage, not the blog
55% of top citations for the medical aesthetics brand went to provider homepages. AI engines skipped the content marketing layer entirely for local queries. Local answer engine optimization (AEO) looks closer to classic local SEO, meaning a clear, complete homepage, than it does to blogging. It is the clearest case in the data of AEO by industry diverging from generic best practices.
Healthcare: authority is the format
Articles won healthcare with 54% of citations, but not just any articles. Citations concentrated in articles from institutional sources, a pattern we unpack in the trust gap section below. In a category where wrong answers carry real consequences, the engines behaved like strict editors.
Higher education: official pages win outright
AI sent prospective students straight to official program pages, which earned 53% of citations in the category. The university’s own domain earned 75% of all top-30 citations in its category at a 5.09 citation rate, the highest single-domain trust score in the study.
For a university, 75% of AI citations pointed to its own program pages. For a local med spa, 3%.
Blended fingerprints
Four industries split their citations across two formats rather than one. Consumer automotive combined articles (41%) and listicles (23%) for 64% of citations. B2B cybersecurity compliance combined articles (38%) and how-to guides (13%) for 51%. Specialty food CPG combined listicles (28%) and homepages (26%) for 54%. Government scholarship programs leaned on listicles (25%) and directory-style pages with a mixed remainder. A blended fingerprint is still a fingerprint: two formats to build, not ten.
The trust gap: AI reads everything, cites selectively
Comparison pages are the most trusted format almost nobody builds. They earned only 4.1% of citations but posted the highest citation rate of any page type at 1.87 citations per retrieval, 45% above the portfolio average of 1.29. When an AI engine retrieves a comparison page, it leans on it hard. The starkest example came from the B2B cybersecurity data, where a “GCC High vs GCC vs Commercial” comparison earned 4.09 citations per retrieval, the highest of any URL in that industry’s top 30. Comparison content is underproduced relative to how much AI engines trust it, which makes it the most actionable single recommendation in this report.
The same retrieval-versus-citation lens explains which domains AI engines actually trust:
| Domain type | Citation rate |
|---|---|
| Brand’s own domain (“You”) | 2.93 (1.51 excluding the higher-ed outlier) |
| Editorial | 1.43 |
| Reference (Wikipedia, directories) | 1.21 |
| UGC (Reddit, YouTube, LinkedIn) | 1.16 |
| Institutional (gov, edu, medical) | 1.12 |
| Corporate (third-party commercial) | 1.07 |
| Competitor | 1.01 |
The starkest example is consumer health media. AI engines retrieved Healthline, WebMD, and Medical News Today constantly for diabetes-related queries, then declined to cite them: citation rates of 0.55, 0.33, and 0.24 respectively. Meanwhile CDC (1.85), Mayo Clinic (1.79), and NIH (1.66) were both retrieved and cited. AI engines made a visible quality judgment between lookalike health content and primary medical authorities.
AI engines retrieved WebMD constantly. They cited it 0.33 times per read. Mayo Clinic: 1.79.
YouTube showed the same split by vertical. It was the single most-retrieved domain for the automotive brand and converted respectably there at 1.33. For higher education, YouTube converted at just 0.27. Video gets read everywhere; whether it gets cited depends on the vertical.
Where AI visibility is actually won: the 0% to 75% range
74.7% of top-30 domain citations in higher education went to the brand’s own website. In B2B technology services, that number was 0%. Between those poles sits every other industry in the study, and the spread is the most strategically important table in this report.
| Industry | Own-domain citation share |
|---|---|
| Higher education | 74.7% |
| Specialty food CPG | 18.9% |
| Government scholarship program | 11.5% |
| Healthcare nonprofit | 8.3% |
| B2B cybersecurity | 8.2% |
| Consumer automotive | 7.4% |
| Medical aesthetics (local) | 3.0% |
| B2B technology services | 0.0% |
The 0% is the headline within the headline. One B2B services firm’s own website did not crack the top 30 cited domains for AI queries in its own category. Not underrepresented. Absent.
One B2B firm’s own website didn’t crack the top 30 domains AI cites about its own category.
In most industries, AI visibility is won on other people’s websites: editorial reviews, roundups, Reddit, and LinkedIn. That is a digital PR and third-party placement strategy, not an on-site content strategy. Higher education is the exception that proves the rule. When queries are program-specific, official pages win.
There is a nuance worth holding onto from the domain-type table above. Own domains that do get retrieved are cited at the highest rate of any category, 1.51 even excluding the higher-ed outlier. The problem is not that AI distrusts brand websites. The problem is getting retrieved at all.
UGC is a first-class AI source
Reddit appeared among the top cited domains for 7 of the 8 brands, every industry except the healthcare nonprofit. LinkedIn was the number 1 most-cited domain for the B2B technology services brand, with 736 citations, ahead of every corporate site, analyst firm, and news outlet. arXiv, the academic preprint server, was the number 3 cited domain in that same B2B category: AI answers about “agentic AI companies” cite academic papers alongside vendor roundups.
If your AI visibility plan has no answer for Reddit, LinkedIn, or the reference layer of your category, it has a hole where a large share of citations live.
What this means for your business
The data supports five prescriptive moves, each keyed to a business type.
- B2B services firms: shift budget from blogging to placements. With listicles earning 61% of citations in B2B services and one firm’s own domain earning 0%, the highest-return work is getting named in third-party “top 10” roundups, building a credible LinkedIn presence, and earning mentions in the communities AI already cites. This is the core of a generative engine optimization program, and it looks more like digital PR than content marketing.
- Local service businesses: make the homepage complete before writing anything else. Homepages earned 55% of citations for the local services brand. A clear homepage that states what you do, where you do it, and why you are credible outperforms a blog for local AI answers. This extends what already works in local SEO rather than replacing it.
- Healthcare and other trust-sensitive categories: publish at institutional grade or partner with institutions. The engines cited Mayo Clinic at 1.79 and WebMD at 0.33 for the same query space. Lookalike health content gets read and skipped. Articles with real clinical authority behind them are the only format that converts retrievals into citations here.
- Higher education and product-led organizations: invest in official program and product pages. Program pages earned 53% of higher-ed citations, and the university’s domain converted at 5.09 citations per retrieval. When buyers ask about a specific named offering, AI engines prefer the official source. Make those pages complete, current, and specific.
- Everyone: build comparison pages. At 1.87 citations per retrieval, 45% above the portfolio average, “X vs Y” pages are the most trusted format in the study and among the least produced at 4.1% of citations. Publishing honest comparisons in your category is the cheapest citation-rate win available.
One more framing note: these citations increasingly happen inside answers that never send a click, which is why we treat AI visibility as part of the same system described in our work on zero-click marketing. Winning the citation is the new impression, and in many categories it happens on domains you do not own.
Citation-ready findings
Writing about AI search? These findings are free to cite with attribution to the DeltaV Digital 2026 AI search citations study, linked to this page.
- Listicles earn 61% of AI citations in B2B services. In healthcare, they earn 0%.
- Comparison pages are cited 45% more often per retrieval than the average page. Almost nobody builds them.
- AI engines retrieved WebMD constantly. They cited it 0.33 times per read. Mayo Clinic: 1.79.
- One B2B firm’s own website didn’t crack the top 30 domains AI cites about its own category.
- For a university, 75% of AI citations pointed to its own program pages. For a local med spa, 3%.
Frequently asked questions
What types of pages does ChatGPT cite?
Across 25,337 citations from ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode, articles earned 23.7% of citations, listicles 19.6%, and product pages 16.3%. The mix shifts sharply by industry: listicles earned 61% of citations in B2B technology services, homepages earned 55% in local services, and program pages earned 53% in higher education. Comparison pages earned the highest citation rate of any page type at 1.87 citations per retrieval.
Do AI engines cite Reddit?
Yes. Reddit appeared among the top cited domains for 7 of the 8 brands in our 2026 study, every industry except the healthcare nonprofit. UGC domains as a group, including Reddit, YouTube, and LinkedIn, converted at 1.16 citations per retrieval, and LinkedIn was the single most-cited domain for the B2B technology services brand with 736 citations.
How do I get my website cited by AI?
Match the citation fingerprint of your industry rather than following a generic checklist. In our data, local businesses earned citations through complete homepages (55% of citations), universities through official program pages (53%), and B2B services firms through third-party roundups and LinkedIn rather than their own sites. Across all industries, comparison pages earned the highest citation rate at 1.87 citations per retrieval, and own domains that were retrieved converted at 1.51 even excluding the higher-ed outlier, so the bottleneck is usually retrieval, not trust.
Request an AI visibility audit
This study tells you where citations concentrate in 8 industries. An audit tells you where they concentrate in yours: which prompts your buyers run, which domains the engines cite when they answer, and how far your current pages sit from your industry’s fingerprint. Request an AI visibility audit and we will map your citation footprint against this dataset.
About DeltaV Digital
DeltaV Digital is an integrated digital marketing agency connecting SEO, paid media, and web development into unified growth systems. The team has run organic and AI search programs for 12+ years across healthcare, beauty, technology, professional services, finance, education, and ecommerce, with 300+ engagements and a 95% client retention rate. This study is the first release in the agency’s ongoing AI search research program; the dataset refreshes quarterly.