Study Says Long-Form 'Complete Guides' Get Cited Once by AI Search, Then Disappear

AI search engines are ruthless editors. A new study says most of what they cite gets used exactly once, then thrown away.

Somantra AI, an Answer Engine Optimization platform based in Sydney, tracked 2,437,107 citation records across 28,725 domains on ChatGPT and Google AI Overviews from November 2025 through July 2026, according to a report the company published Monday, August 17. The subject was Australian insurance websites. The finding: 57.2% of domains got cited in exactly one month of the seven-month window and were never surfaced again. Only 2.7% of domains, about 779 sites, earned a citation every single month.

Somantra calls the gap between those two numbers the "content graveyard." The company's headline isn't the churn rate. It's what separates the survivors from the graveyard.

Format, Not Brand, Decides Who Survives

According to Somantra founder Arun Prasad, domain authority and brand recognition explained almost nothing about which sites kept getting cited. Format explained a lot.

Pages built around comparison tables, FAQ structures, and discount or pricing language got cited persistently at roughly twice the rate of one-and-done domains, the study found. Government explainer sites and comparison sites dominated the list of domains cited across all seven months.

Meanwhile, content labeled a "complete guide" — the sprawling, 3,000-to-12,000-word format that dominated Google SEO strategy for the past decade — was 3.5 times more common among domains that got cited once and vanished than among domains cited every month.

"If a content team is still producing long-form guides because that is what won on Google in 2020, the data suggests those pages are being cited once and then left behind," Prasad said. "Discount pages, comparison pages and FAQs are the formats holding their citations. Brands that make that switch keep their visibility. Brands that do not lose it one citation cycle at a time."

Somantra itself flags the limits of its own data. The company states plainly that the study establishes correlation across a single category and a seven-month window, not causation, and that it's consistent with, not proof of, how conversational AI systems assemble answers. That caveat matters because Somantra sells AEO optimization services. A study that concludes "switch your content format" is also a study that sells the exact service Somantra offers. None of that makes the underlying numbers wrong. It does mean readers should treat the 2.4-million-citation dataset as real evidence and the prescriptive advice as a vendor's pitch built on top of it.

Outside Data Points in the Same Direction

Other reporting backs up the broader shift even if it doesn't verify Somantra's specific numbers. The BBC reported, according to Design Rush, that HubSpot lost an estimated 140 million visits in a single year as AI Overviews and algorithm changes cut into click-through rates. HubSpot CMO Kipp Bodnar told the BBC that searches showing an AI Overview see click-through rates roughly 60% to 70% lower than standard results.

HubSpot's response was to break long-form product content into smaller, extractable chunks. AI now delivers 7% to 12% of the company's site visitors in a typical month, according to that BBC reporting. MKM Building Supplies saw AI-referred traffic grow from near zero to a double-digit share of total traffic within a year. Digital director Andy Pickup told the BBC those visitors converted at a higher rate than traditional search traffic. Data from Mersel cited in that same reporting put AI-referred conversion at 14.2% versus 2.8% for organic Google search.

Google's own share of information discovery dropped from 89.3% to 57.6% between December 2022 and December 2025, per that reporting, as more research queries shifted to AI assistants.

What the Coverage Leaves Out

The wire-style writeups of the Somantra study, including versions carried by GlobeNewswire, dailyguardian.ca, and Yahoo Finance Singapore, are near-identical press-release text and don't independently verify Somantra's methodology or citation-tracking tool. Tech Times' writeup adds useful framing by noting the finding "inverts a central assumption of the SEO era" but likewise doesn't interrogate how Somantra defines a "citation" or whether its tracking captures the full universe of AI search queries versus a sample skewed toward insurance-related prompts.

Search Engine Journal's reporting adds a separate wrinkle: Google says its AI Overviews and AI Mode rely on retrieval-augmented generation and "query fan-out," pulling from multiple sources and not documenting how it resolves conflicting information between them. That opacity is the real problem for any business trying to optimize for AI citation. There's no equivalent of a Google Search Console ranking report for ChatGPT, and Google folds its own AI Overview clicks into general search totals, making them hard to isolate.

For any brand deciding whether to gut its content strategy based on one company's single-industry, seven-month study, the open question is simple: does this pattern hold outside Australian insurance, and does it hold on assistants other than ChatGPT and Google AI Overviews? Nobody has published that yet.