What makes content "unique" in Google's definition?

Google's Search Central documentation (updated March 2026) defines unique content as information "not available elsewhere, providing independent value to users." This operates at the data layer, not the prose layer — rewriting a competitor's analysis in different words doesn't clear the bar. Original proprietary data (internal transaction logs, first-party benchmarks, models built on non-public inputs) is the clearest path to content Google treats as independently valuable.

How many data points do you need before publishing a "median"?

Below 15 observations, label the output "illustrative" rather than "median." A median from 8 deals is statistically unstable enough that a single outlier shifts it by 1–2x. At 20+ closed transactions in a defined sector and date range, a median EV/EBITDA figure is defensible — especially if you disclose the N, the date window, and your EBITDA normalization methodology.

Does duplicate content mean copied text, or something broader?

Broader. Google treats content as duplicate when the underlying informational substance is not meaningfully distinct from what's already indexed — even if the words differ. Two pages both summarizing the same Pitchbook report with different phrasing compete for the same slot, and neither wins on uniqueness grounds. Substantive uniqueness comes from data or analysis not reproducible from other indexed sources.

What should I include in a methodology footnote for published financial data?

At minimum: N (number of observations), the date range in absolute terms, what was excluded and why, how the core metric was calculated (e.g., "LTM EBITDA at close, pre-IFRS 16, pre-SBC adjustments, excluding transactions above 20x"), and the data source (internal model, public S-4 filing, third-party provider). That structure is what distinguishes a citable data point from a claim marketing invented — and it's what AI systems check before extracting your number into an overview.

Why do pages with original data earn more backlinks?

Ahrefs' 2024 analysis of 920,000 pages found pages with original proprietary data earned 2.3x more backlinks than topic-matched pages without original data. The mechanism is straightforward: journalists, analysts, and other writers need a citable source for a specific number. If your page is the only place that number exists with a methodology attached, it becomes the citation. Pages that repackage widely available data have no such pull — anyone citing the same statistic links to the primary source instead.

SEO Unique Content: What Google Actually Measures

For finance teams sitting on deal logs, model outputs, and proprietary transaction history, this matters. The gap between what you have and what ends up published is usually editorial erosion, not a data shortage.

What Google Considers "Unique" vs. What Gets Penalized

The distinction Google draws isn't between original and copied — it's between independently valuable and derivative. John Mueller stated in a 2023 Search Central office hours: "We're not looking for unique words, we're looking for unique value."

The practical split:

Type	What it looks like	Google's treatment
Thin content	Summary of publicly available data, no original analysis	Indexed but not cited; no ranking signal
Duplicate content	Same page content at multiple URLs, or near-identical to a competitor	Filtered from results; only one version shown
Derivative content	Repackaged data without a sourced methodology	Competes on authority, not content — and loses
Genuinely unique	Proprietary data, original methodology, or analysis not reproducible elsewhere	Citation magnet; AI overviews pull from this first

Ahrefs' 2024 analysis of 920,000 pages found that 60% of published pages get zero organic traffic, with thin and derivative content as the leading cause. Pages with original data — proprietary benchmarks, internal transaction analysis, first-party survey results — earned 2.3x more backlinks on average than pages covering the same topic with no original data point. The Princeton GEO study (arxiv 2311.09735) found that adding verifiable statistics increased AI citation rates by 37%.

Gary Illyes confirmed the directional trend at SMX Munich 2024: crawl budget increasingly concentrates on pages that demonstrate original research signals.

What the Data-Layer Test Actually Checks

Google's systems don't read your prose for freshness. They check whether the underlying claims exist elsewhere. Run the test yourself: paste your article's 3 most specific claims into a search query. If the top 5 results say the same thing with different words, you're derivative.

The threshold that changes outcomes is a sourced, original data point that can't be reproduced without access to your internal systems. An "industry median EV/EBITDA of 11.2x" sourced to your own closed-deal database is unique. The same number sourced to a Pitchbook PDF is not — it's accessible to anyone with a subscription.

Auditing Your Deal Log for Publishable Content

The workflow most finance teams should run but don't: before handing numbers to marketing, audit your existing deal log column-by-column for what's actually publishable as proprietary content. This gives marketing a spec with methodology attached — not a quote they'll paraphrase into something unverifiable.

A representative deal log has columns like Company, Close Date, Status, Enterprise Value, LTM EBITDA, Multiple, Sector, and Source. Here's how to evaluate each:

Column	Publishable as-is?	What's missing	Fix
Company name	No	NDA / disclosure risk	Anonymize to sector + revenue tier
Close date	Yes	None	Keep — date range anchors credibility
Status	Yes (as filter only)	None	Use to restrict to "Closed"
Enterprise Value	Conditional	Source (public filing vs. model estimate)	Footnote: "per S-4 filing" or "management estimate"
LTM EBITDA	Conditional	Normalization methodology	Add "pre-IFRS 16, pre-SBC" or specify adjustments
EV/EBITDA	Yes	Methodology note strengthens it	"LTM EBITDA at close, pre-synergies" — this is publishable as-is
Sector	Yes	None	Keep
Source	No	Reveals proprietary sourcing channel	Remove or generalize to "PE-backed buyout"

Five checks to run before the handoff:

Does any row name a client or counterparty? Anonymize to a size/sector descriptor (e.g., "mid-market SaaS, $40M ARR").
Are your multiples from public filings or your own model? Flag the difference. "EV/EBITDA per S-4 filing" and "EV/EBITDA per management model" are not interchangeable.
Did you normalize EBITDA? Define the adjustments. "Adjusted EBITDA" without a definition gets rewritten by marketing into something meaningless — and then someone questions the number in a comment.
Is the date range defensible? 18-month windows (e.g., Q1 2023–Q2 2024) hold up better than cherry-picked quarters. Below 15 observations, label the output "illustrative," not "median."
What's the N? Lead with it. "Based on 22 closed transactions in the B2B software sector, Q1 2023–Q4 2024, median EV/EBITDA was 11.2x" is citable. "Multiples ranged from 7x to 14x" is noise.

To pull only closed deals from the last 18 months into a clean export tab:

=QUERY('Deal Log'!A:H,
  "SELECT B, D, E, F, G
   WHERE C = 'Closed'
   AND B >= date '2023-10-01'
   ORDER BY B DESC", 1)

Where column B is Close Date, C is Status, D is EV/EBITDA, E is Sector, F is LTM EBITDA, and G is the methodology note. Add a column H that outputs a pre-written footnote string and you've got a publishable extract marketing can actually use without paraphrasing your numbers into something unverifiable.

ModelMonkey can generate QUERY formulas like this from a plain-English description if you'd rather not hand-write the date arithmetic each time.

Building Unique Content From Internal Data: Version A vs. Version B

The difference between publishable and generic isn't the data — it's the methodology disclosure. Compare:

Version A (derivative): "Median EV/EBITDA multiples in software M&A were approximately 11–14x over the past two years."

Version B (publishable): "Median EV/EBITDA was 11.2x across 22 closed B2B software transactions in Q1 2023–Q4 2024, measured on LTM EBITDA at close, pre-synergies, excluding outliers above 20x."

Version A is on 40 other pages. Version B is only yours. The Princeton GEO research shows AI systems extract and cite Version B at substantially higher rates because the methodology makes it verifiable.

The spec format worth internalizing before handing anything to marketing:

N: number of observations
Date range: absolute, not relative ("Q3 2023–Q4 2024," not "the past 18 months")
Exclusions: what was filtered and why
Methodology: how the metric was calculated (LTM EBITDA at close, pre-IFRS 16, etc.)
Data source: internal model, public S-4, third-party data provider

That's the IC memo format applied to content. It also happens to be exactly what AI systems need to justify citing your page instead of a competitor's.