---
id: ins_sourav-mohanty-ai-quotes-require-constraints-to-be-trustworthy
title: 'AI quotes are only as good as the instruction that produced them.'
operator: Sourav Mohanty
operator_role: Not yet a Product Marketing Leader | B2B Positioning & Messaging | ARR is not my son but I’ll raise it
source_url: https://www.linkedin.com/feed/update/urn:li:activity:7434483009837477888/
source_type: thread
source_title: 'AI quotes are only as good as the instruction that produced them.'
source_date: 2026-04-10
captured_date: 2026-05-03
domain: [research-discovery, pmm, content]
lifecycle: [research]
maturity: applied
artifact_class: framework
score: { originality: 3, specificity: 3, evidence: 2, transferability: 3, source: 3 }
tier: A
related: []
raw_ref: raw/linkedin/reactions/linkedin-reactions-2026-04-10.md
---

# AI quotes are only as good as the instruction that produced them.

## Claim
When using AI to extract quotes from transcripts, 'verbatim' is too vague, AI fills gaps with its own judgment, stitching together sentences and trimming hedges, making the output look sourced but potentially fabricated.

## Mechanism
AI reads transcripts relative to the job given, not neutrally. To get trustworthy quotes, lock specific constraints: copy one uninterrupted customer turn only, no merging or cleanup, preserve hedges/fillers, include speaker label + timestamp + context, and explicitly mention any changes.