We Checked the Source on the Most-Cited AEO Statistics

The AI visibility advice space has a statistics problem. The same numbers circulate across agency blog posts, LinkedIn threads, and consultant decks -- often without a primary source. When we traced them ourselves, we found some that don't hold up. Others came back verified, with solid methodology behind them.

We did this work for a specific reason: building a research page means citing only what we can defend to a journalist. What followed was more systematic than we expected.

The "85%" freshness figure everyone quotes

The single most repeated AI citation statistic in AEO content is some version of: "85% of AI-cited URLs are under two years old."

In our May 18, 2026 investigation of commonly cited AEO statistics, we traced this figure to its apparent origin. The most credible primary study with the closest methodology is Seer Interactive's analysis of 5,000+ URLs cited across ChatGPT, Perplexity, and Google AI Overviews, drawn from 42 organizations between June 2024 and September 2025.

The Seer Interactive finding is **79%**, not 85%.

The 85% number appears to be rounding error in practitioner aggregation -- someone summarized the data as "roughly 80%," someone else rounded up, and the inflated figure spread. The distinction matters. 79% and 85% describe meaningfully different levels of freshness concentration, and citing the higher number in a recommendation makes a client's problem sound more urgent than the primary data supports.

The full Seer breakdown across thresholds: 65% of AI-cited content was published within the past year. 79% within two years. 89% within three years. 6% was older than six years. Freshness concentration is real and significant. The number has been overstated.

The "3.7x original research" claim

A second widely quoted figure: "pages with original research are 3.7x more likely to be cited by AI platforms."

We searched for the primary study behind this. We did not find one.

The figure originates from a 2026 practitioner synthesis that aggregated 23 AI citation studies. It is a secondary analysis -- a summary of other sources, not a direct measurement of citation behavior. We could not independently verify the underlying 23 studies or identify a primary dataset that produced the 3.7x figure specifically.

More importantly, the mechanism the figure is attributed to -- "original content" -- may be measuring something else. Two primary-sourced findings exist at comparable scale:

Domains listed on Trustpilot, G2, Capterra, Yelp, or BBB have a 3x higher ChatGPT citation rate than domains without such listings. Sites with 32,000+ referring domains are 3.5x more likely to be cited by ChatGPT than sites with fewer than 200 referring domains. Both findings come from SE Ranking's November 2025 primary study with disclosed methodology.

These are domain authority and third-party validation effects. Pages with original research tend to be published on high-authority domains with strong external citation profiles. The practitioner synthesis may be attributing to "original content" what is actually an authority infrastructure effect. The two signals are not easily separated without a controlled study.

The direction -- that publishing something distinctive and difficult to replicate raises citation probability -- is credible. The 3.7x multiplier is not verified at a primary level.

The "30% year-signal improvement" claim

Third in the commonly repeated set: including the current year in page titles and headings improves AI citation rates by approximately 30%.

No primary study found.

The mechanism is plausible. Perplexity's live retrieval weights metadata recency signals. A page titled "Commercial Plumbers in Hamilton 2026" signals currency in a way that "Commercial Plumbers in Hamilton" does not, and that signal is visible to the crawler before it reads the page content. For a platform explicitly doing live retrieval, this is a reasonable lever to use.

But plausible mechanism is not a verified finding. The 30% figure appears in practitioner roundups without a source URL. It likely spread through the same aggregation pattern as the 85% figure: one summary, widely repeated until the number acquires apparent authority.

We include recency signals in our fix recommendations because the mechanism is sound. We do not quote a specific lift percentage, because none has been verified.

What held up

Three sources produced primary, methodology-disclosed findings that we use in our own analysis:

**Seer Interactive (5,000+ URLs, 42 organizations, Jun 2024 -- Sep 2025):** The freshness breakdown cited above. Additionally: 35% higher organic click-through rate for brands cited in Google AI Overviews vs. brands not cited. Primary methodology disclosed and traceable.

**SE Ranking (November 2025):** The review platform and domain authority findings cited above. Domains on major third-party review platforms have a 3x higher ChatGPT citation rate. High-authority domains (32K+ referring domains) are 3.5x more likely to be cited by ChatGPT. Both are primary findings with disclosed sample methodology.

**Ahrefs (17 million citations, 7 platforms):** Average age of URLs cited by AI assistants: 1,064 days -- 25.7% fresher than organic SERP citations on average. Platform breakdown: ChatGPT in-text citations average 1,023 days old; Copilot, 1,056 days; Gemini, 1,118 days; Perplexity, 1,166 days.

Why the Perplexity age number is counterintuitive

That Perplexity figure -- 1,166 days -- is worth stopping on. Perplexity does live web retrieval on every query. You would expect it to have the freshest citation profile of any major platform. It doesn't, at least not in the Ahrefs data.

One explanation: Perplexity also uses training-data sources for certain query types, particularly broad informational queries where live retrieval may not be the dominant mechanism. The 1,166-day average may reflect this mixed behavior more than pure live-retrieval. This complicates the simple "Perplexity always rewards fresh content" frame.

Our May 2026 platform investigation (knowledge file updated 2026-05-17) found that Perplexity's freshness advantage is most pronounced for local and transactional queries -- "best accountant in Hamilton," "plumber near me" -- where live retrieval is clearly in use. For informational queries, the picture is less clear. Recommending that a client urgently update their FAQ pages to improve Perplexity performance is probably right for some query types and likely overstated for others.

What this means for decisions

The direction of the commonly cited statistics is largely right. Freshness matters. Directory presence matters. Domain authority matters. The problem is when imprecise multipliers drive specific prioritization decisions.

A client asking "will adding 2026 to my title tags improve my Perplexity score by 30%?" gets an accurate answer from us: the direction is probably right, the 30% is a practitioner claim without a source. A client asking whether getting listed on G2 will move their ChatGPT score gets a better answer: there is a primary study showing a 3x citation lift for domains with review platform presence. That is a different level of confidence.

The practical priority order that the verified data supports: third-party review platform presence (3x ChatGPT citation lift, primary sourced), domain authority from inbound links (3.5x ChatGPT citation lift, primary sourced), and content within the last two years (real signal, but the 79% concentration means you're competing inside a large group, not against an unusual threshold).

For local service businesses, the question is not whether freshness or directory presence matter -- they do -- but which gap is actually driving a low score on a specific platform. A business blocking PerplexityBot in robots.txt while updating page titles is solving the wrong problem first. A business with thin directory presence blaming content age for a low ChatGPT score is misreading the signal.

A Signal Check at sourcepull.ca shows the per-platform breakdown -- ChatGPT, Perplexity, Gemini, and Claude scored separately. When the gap is visible at that level, the fix path is a lot clearer than any single statistic can make it.