ai-fanout.comOpen the lab
Menu

AI Answer Evidence Lab · browser-local

Map what stays. Mark what changes.

Compare public AI answers, visible source URLs and user-defined coverage criteria. Export the evidence before you call a pattern a trend.

Open the Evidence Lab Build a protocol
  • No signup
  • No provider call
  • No input upload
Illustrative interface previewVisible source recurrence
Not research data
Q

What should a team record before comparing public AI answers?

Source domainRun ARun BRun C
method.example●●●
schema.example●—●
rights.example—●—
reference.example●——
recorded as visible not recordedCalculation rules
01

From pasted answer to inspectable package.

  1. 01Record

    Add the question, public surface, date, answer and visible source URLs.

  2. 02Compare

    See source recurrence, literal coverage, answer overlap and comparability notes.

  3. 03Export

    Keep the observations, method version and boundaries together in JSON or CSV.

02

A result you can audit—not a visibility score.

Every measure exposes its inputs and limit. The lab never ranks providers, judges factual accuracy or guesses the private searches behind an answer.

Analyze your observations
Source recurrenceWhich visible hosts return?

Presence by observation, normalized at the host level.

Answer overlapHow much vocabulary is shared?

Mean pairwise Jaccard overlap of normalized content-word sets.

Literal coverageWhich supplied phrases appear?

Transparent case-insensitive phrase matching—never semantic scoring.

ComparabilityWhat changed around the answer?

Surface, model label, locale, route, timestamp and missing-source warnings.

Tool 01 · analyze

AI Answer Evidence Lab

Bring one to five observations. Compare visible sources and answer properties entirely in the browser.

  • JSON and CSV exports
  • Synthetic demo included
  • No automatic storage
Open the lab
Tool 02 · plan

Observation Protocol Builder

Freeze the question set, route, repetitions, missing-data rule and rights decision before collection.

  • JSON and Markdown exports
  • Readiness signals
  • No scheduling or collection
Build a protocol
03

Understand every field in the output.

The maintained library supports the tools with definitions, measurement contracts, data standards and field procedures.

  1. M-01
    Source diversity

    Count visible source domains with an explicit denominator.

  2. M-02
    Citation persistence

    Measure which visible citations recur across observation windows.

  3. M-03
    Answer stability

    Compare observable answer properties without manufacturing a trend.

  4. F-01
    Source audit field guide

    Run a documented audit of links shown in public AI answers.

Browse all 13 references
04A

Observable

  • Public final answers supplied by the user
  • Visible cited or linked source URLs
  • Declared surface, route, locale and time
  • User-defined literal coverage criteria
04B

Not exposed

  • Hidden fan-out queries or system prompts
  • Private retrieval traces or ranking internals
  • Model reasoning or intermediate steps
  • Factual accuracy or causal effects

Public tools · research claims gated

Useful tools now. Provider benchmarks only after proof.

The website is indexable, but no provider collection has started and no public provider dataset exists. The tools and synthetic example do not turn a draft protocol into a research finding.

0of 3 comparable batches

Inspect research status