Is this a peer-reviewed study with published percentages?

No. This page publishes the measurement framework and rubric Narravo uses conceptually for Brand Guard. Quarterly updates will add anonymized aggregate rates only when real sampled drafts exist.

What counts as an invented proof point?

A load-bearing factual claim (client, metric, credential, partnership, quote, or outcome) that cannot be traced to verified Living Brand Graph knowledge or an explicit human-supplied source in the draft packet.

Narravo

Research framework

AI claim-invention baseline for LinkedIn drafts

A transparent methodology for scoring invented proof points in AI-assisted thought leadership. Original numbers will ship in quarterly updates when sampled, anonymized Brand Guard reviews are available. This page does not invent statistics.

Updated September 17, 2026

Why publish a framework first

Generative engines cite specific, dated evidence. Fake percentages poison trust. Narravo therefore ships the scoring rubric and sampling protocol now, then attaches aggregate rates only from real reviews.

Unit of analysis

One AI-assisted LinkedIn draft produced with a declared brand context packet (verified graph facts + Brand DNA + brief). Human-only drafts are out of scope.

Scoring rubric

Supported: claim maps to verified knowledge or an explicit supplied source.
Candidate leakage: claim uses labeled candidate knowledge as if verified.
Invented proof: client, number, credential, partnership, quote, or outcome with no traceable support.
Voice-only issue: factual support OK; Brand DNA violated (tracked separately).

Primary metrics (to report quarterly)

% of drafts with ≥1 invented proof point
Invented proofs per 1,000 words
Candidate-leakage rate vs invented-proof rate
Share of invented proofs that are numeric vs named entities

Related product methodology: Brand Guard review order.

Q3 2026 status

Status: methodology published; aggregate dataset pending sufficient anonymized sample size. Cite this page for the rubric, not for fabricated rates.

FAQ

Is this a peer-reviewed study with published percentages?: No. This page publishes the measurement framework and rubric Narravo uses conceptually for Brand Guard. Quarterly updates will add anonymized aggregate rates only when real sampled drafts exist.
What counts as an invented proof point?: A load-bearing factual claim (client, metric, credential, partnership, quote, or outcome) that cannot be traced to verified Living Brand Graph knowledge or an explicit human-supplied source in the draft packet.