Proprietary data: top asset for AI citations

Publish original numbers — according to Kevin Indig and Amanda Johnson of Growth Memo, that is the most reliable lever for making a page more original. The most defensible figures are a byproduct of the business itself, not data assembled to feed a content calendar. On Search Engine Land, the authors explain why proprietary first-party data is becoming the most valuable citation asset in the era of AI search and generative engine optimization — and why owning the number alone is not enough.

Paying for loosely tied surveys is fading. Almost every product generates publishable data today, and you do not need a research team. The bar to stand out is lower than many teams assume.

First-party data and information gain

On-Page.ai's recent information gain study scored 150 top-3 Google pages across 50 keywords in ten verticals on how much each adds beyond its ranking cohort. The scale runs from 0 to 100 and measures meaningful contribution, not wording.

The median score was 52. Original data correlated with that score more than any other page-level trait — including length. Pages with at most one unique figure averaged 40.2 points, while pages with 15 or more unique metrics averaged 62.1. The score climbed steadily at every step in between.

The good news: the bar is low. Top organic results contain only about four unique data points on average. Publish a page with more than four real original claims, figures, or answers and you gain another lever for increasingly competitive organic visibility.

The analysis also found that in almost every search, adjacent unanswered questions remain open. On-Page worked with synthetic reader questions closely related to each query topic. The result resembles query fan-out: new pages can close these gaps and stand out.

AI citations and comprehensive content

Growth Memo found a similar pattern in a ChatGPT citation analysis: a single evergreen page covering more than ten query intents delivers more AI citation reach than ten single-intent pages. The ROI of comprehensive content is front-loaded — one well-built page compounds citation reach over time. The long tail exists, but the top five percent of pages capture a disproportionate share of ongoing citation activity.

Brands should therefore monitor high-intent prompts across the buyer journey. The Reasoning Lift model divides the path into Problem, Exploration, Comparison, Validation, and Selection. Answering these questions with knowledge only your brand can provide keeps you competitive.

However, the study focuses on classic rankings. Information gain comes from Google patent language. AI citations, AI Mode, and AI Overviews are not included — a relevant blind spot for GEO teams.

Caveat: primary source does not automatically win

Most proprietary data advice skips a crucial point: few test whether AI rewards the brand that originated a number or the page that presents it most readably.

• DATE and NUMBER are the entity types that most strongly predict ChatGPT citations, according to Growth Memo.
• Highly cited pages are dense with specific entities: methodology, precise statistics, named comparisons.
• Even when other sources are cited, external authority signals can still strengthen your brand.

Entity richness and balanced sentiment also matter. Generic advice feels vague; proprietary data creates verifiable entities. Publishing proprietary content is necessary but not sufficient: LLM extraction structure and trusted sources in the topic decide who gets cited — even when your brand owns the data.

Structuring data for extraction

Growth Memo analyzed 18,012 verified ChatGPT citations and found a ski-ramp distribution: 44.2 percent of all citations come from the first 30 percent of a page. The middle third earns 31.1 percent; content buried deep is roughly 2.5 times less likely to be cited. In every vertical, AI reads hardest in the 10–20 percent band after the title block.

Five rules for extractable data studies

Kevin Indig and Amanda Johnson derive five concrete structure rules from the citation data. First, place the strongest metric in the first 30 percent of the page, ideally right after the title in the 10–20 percent band: figure, comparison, implication — all in the first screen. Second, define immediately: one sentence on what the number measures and which population it covers. Undefined statistics are harder for AI systems to extract with confidence.

Third, make methodology visible in a short labeled block: sample size, time window, collection method. Attribution confidence is part of citability. Fourth, rank secondary findings by strength and place them up front. The 20-paragraph narrative buildup is a human retention pattern that costs machine citations. Fifth, skip the suspense close: AI reads like a busy editor, not a patient reader.

Owning data and structuring it for extraction wins visibility in classic search and AI answers. Aggregators with better-structured pages can collect citations your original research earned. Brands sitting on product, usage, or pricing data should therefore think about structure, off-site authority, and prompt tracking together.