AI Citation Study: Comparison Pages Convert Best, But Rarely Get Picked

AI models retrieve comparison pages less often than articles or listicles, but when they do, they quote them at the highest rate of any content type: 1.87 citations per retrieval, 45% above average. The same research shows institutional sources like government and university sites get cited far more consistently than commercial publishers, even in categories where those publishers dominate search traffic. For businesses chasing AI visibility, the takeaway is simple: getting pulled into an AI's research is not the same as getting credited in its answer.

What changed?

Digital marketing agency DeltaV Digital published a new AI search citations study analysing how ChatGPT, Perplexity and Google AI decide which pages to cite. The team collected 21,075 AI engine responses daily between 14 April and 13 July 2026 using Peec AI, an AI search visibility platform, producing a dataset of 25,337 citations across eight brands.

The headline finding challenges a common assumption in GEO circles that one content format beats all others. No single page type dominated across the portfolio: articles earned 23.7% of citations, listicles 19.6%, and product pages 16.3%.

The more interesting number sits further down the table. Comparison pages posted the highest citation rate of any page type at 1.87 citations per retrieval, well above the portfolio average of 1.29, yet earned only 4.1% of total citations. When AI models do pull from a comparison page, they trust it and quote it often. They just don't pull from comparison pages very often in the first place.

The study also exposed a gap between being retrieved and being cited, particularly in health content. 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 respectively. By contrast, the CDC (1.85), Mayo Clinic (1.79) and NIH (1.66) were both retrieved and cited at much higher rates. The pattern suggests AI models increasingly treat commercial health publishers as background reading, while leaning on institutional sources for the actual quote.

Brand-level variation was just as stark. Own-domain citation share ranged from 0% for a B2B technology services firm to 74.7% for a university. One community platform kept showing up almost everywhere: Reddit appeared among the top cited domains for 7 of the 8 brands studied.

This lands alongside separate research that reinforces the same theme from a different angle. A May 2026 meta-analysis by Cyrus Shepard of Zyppy, covered by Digital Applied, scored the evidence behind 23 proposed AI citation factors by synthesising 54 experiments, patents and case studies into a single ranking, published on Zyppy Signal on 7 May 2026. Two findings stand out: brand web mentions correlate roughly three times more strongly with AI visibility than backlinks, and cited content runs about 25.7% fresher than organic top-10 results across nearly 17 million citations.

Together, the two studies point the same way: authority earned through mentions elsewhere, and trust built through institutional standing, now matter more to AI citation than page format or backlinks alone.

Why does this matter for your business?

If your GEO strategy has focused purely on publishing comparison guides because they feel like natural AI bait, this data suggests you're optimising for the wrong metric. Comparison pages convert into citations at a high rate once retrieved, but most brands will never get the volume of retrievals that articles and listicles receive by default.

The health sector findings are a warning for any commercial publisher competing with institutional or government sources. If AI models are willing to read your content but prefer to quote a .gov or .edu domain instead, your traffic from AI referrals will underperform your visibility. You'll show up in the model's research, but not in the answer the user actually sees.

The Reddit finding matters too. If Reddit appeared among the top cited domains for 7 of the 8 brands studied, then a brand's presence, or absence, in genuine Reddit discussion is now effectively part of its AI visibility profile, whether or not marketing teams have ever treated Reddit as a channel worth managing.

The Zyppy meta-analysis adds a practical prioritisation signal: brand web mentions correlate roughly three times more strongly with AI visibility than backlinks. For years, SEO teams have built link-building programmes as their primary off-site lever. This data suggests unlinked brand mentions, in press coverage, forums, comparison sites and review platforms, deserve at least equal investment now.

What should you do now?

Start by auditing which content types your AI citations are actually coming from, not which types you assume should be working. If comparison content isn't getting retrieved often, the fix might be distribution and internal linking, not more comparison pages.

Check whether your industry has institutional competitors your commercial content can't out-cite on trust alone. In regulated or health-adjacent categories, that may mean partnering with, or being cited by, recognised bodies rather than trying to outrank them directly.

Audit your presence on Reddit and other community platforms where genuine discussion happens. This isn't about posting promotional content; it's about making sure accurate, current information about your brand exists where AI models are already looking.

Shift some budget from pure backlink acquisition toward earned brand mentions: quotes in trade press, inclusion in review roundups, and citations in forums and Q&A sites, given the three-times correlation strength the Zyppy analysis found.

Finally, track your own citation rate against your retrieval rate if you can. Being mentioned in an AI model's source list without being quoted is a measurable gap, and closing it is now a distinct discipline from simply getting indexed. If you haven't checked where your brand currently stands, running a free audit with Sited is a straightforward way to see which AI platforms already recognise you and which ones don't.

Frequently asked questions

What's the difference between being "retrieved" and being "cited" by an AI model?

Retrieval means the AI system pulled your page into its research process when answering a query. Citation means it actually quoted or linked to your page in the final answer. Healthline, WebMD and Medical News Today were retrieved constantly for health queries but cited at rates as low as 0.24 citations per retrieval, showing the two are not the same thing.

Should I stop publishing comparison pages if they only get 4% of citations?

No. Comparison pages posted the highest citation rate of any page type at 1.87 citations per retrieval, 45% above the portfolio average. The issue is retrieval frequency, not quality, so the fix is better distribution and internal linking to those pages rather than abandoning the format.

Does this mean backlinks no longer matter for AI visibility?

Backlinks still matter, but the evidence suggests they're not the strongest lever anymore. Brand web mentions correlate roughly three times more strongly with AI visibility than backlinks, so unlinked mentions in press, forums and review sites deserve renewed attention alongside traditional link building.

Why does Reddit keep showing up in AI citations?

Models appear to treat Reddit as a proxy for genuine, unfiltered user opinion, which is valuable for answering comparison and recommendation queries. Reddit appeared among the top cited domains for 7 of the 8 brands studied in the DeltaV Digital research, making it a channel worth monitoring even without an active posting strategy.

How was the DeltaV Digital study conducted?

The team collected 21,075 AI engine responses daily between 14 April and 13 July 2026 using Peec AI, an AI search visibility platform, then analysed the resulting citations across eight brands in different industries to compare page-type and domain-level performance.