Skip to content
Science-backed data on why AI says what it says
Citation signals

Numeric signals

Numeric signals are text patterns that BLURSOR recognises as possible quantitative statements, such as percentages, prices, ranges, or proportions. They are detected formats, not verified statistics or a measure of research quality.

Checked against BLURSOR’s measurements ·

What it means

A number becomes useful when a reader understands what it describes. For a percentage, that can mean the underlying population and time period. For a price, it can mean the currency, package, and conditions. A count of number-shaped text cannot supply that context.

When publishing a figure based on data, identify its origin and explain transformations that affect the claim. W3C's data guidance treats provenance as context that helps others interpret and assess data. Adding more numbers without that context does not make a page better supported.

Illustrative example

Hypothetical example: explaining a process figure

Suppose a fictional workshop writes '18 of 20 repair requests were answered within one working day during our June trial.' The fraction, period, and process are clearer than a standalone claim of '90% faster'. The example's figures are invented solely to illustrate wording.

Even the clearer sentence would need a real record and a definition of 'answered' before publication as a factual claim. It would not establish that the same result applies to every month, location, or type of request.

How BLURSOR measures it

In the report: Numeric patterns

BLURSOR scans cleaned raw-page text for selected patterns: percentages, certain currency amounts, comma-grouped large numbers, magnitude suffixes, plus-suffixed counts, multipliers, numeric ranges, and expressions such as '18 of 20'. Ordinary digits and every possible local number format are not universally counted.

The metric counts pattern occurrences, not distinct facts or independent studies. Repetition can increase it. A price, marketing claim, and properly sourced research result can all match the same formatting rules. The checker does not validate calculations, consult the source dataset, or establish statistical significance.

These citation metrics need at least five extracted readable blocks. Too little usable content or suppressed page evidence produces an unknown result rather than zero. The scan covers retained raw content, not every later browser interaction.

What to do next

  1. For important figures, state what was counted, the relevant period, units, and conditions; give readers a source or a clear explanation of your own method.
  2. Have the content owner check percentages and comparisons against the underlying records, including the denominator and the comparison baseline.
  3. Keep only numbers that help answer the visitor's question. Do not invent statistics or repeat them to raise a detector count.

What it cannot tell you

  • A zero can mean no tested format matched even though the page contains useful numbers or qualitative evidence.
  • No numeric count or threshold proves credibility, suitability for citation, or improved visibility in an AI answer.

Sources and further reading

The definitions below describe the underlying concepts. The measurement section describes BLURSOR’s own checks.

Related explanations