AIO APEX
fr
Works best with reasoning-focused models (Claude Opus 5, GPT-5.4, Gemini 3 Pro). For highest reliability, run the audit with a DIFFERENT model than the one that generated the original content — cross-model review catches fabrications a single model is blind to, since a model rarely flags the kind of confident-sounding claim it would have generated itself.A marketing manager used ChatGPT to draft a competitive analysis report due to the VP of Sales in 45 minutes. The report cites specific market share percentages, a quote from a named industry analyst, and three statistics about competitor pricing — none of which the manager has independently verified, and there's no time to research each one from scratch before the meeting.Artificial Intelligence

L'auditeur de fiabilité des sorties IA : transformez n'importe quel rapport généré par l'IA en brouillon vérifié et validé

Partager:
L'auditeur de fiabilité des sorties IA : transformez n'importe quel rapport généré par l'IA en brouillon vérifié et validé

Pourquoi ce prompt est important

In a well-documented 2023 case, two New York lawyers were sanctioned after submitting a legal brief containing six fake case citations fabricated by ChatGPT — neither lawyer had checked the citations before filing. Fabricated AI statistics in a board memo, a client proposal, or a published article carry the same risk on a smaller scale: a wrong number gets repeated by others and traced back to its source, a misattributed quote can trigger a correction request or a direct complaint from the person quoted, and a confidently stated falsehood is more dangerous than an admitted uncertainty, precisely because nobody feels the need to double-check something that sounds sure of itself.

À quoi nous l'utilisons

A marketing manager used ChatGPT to draft a competitive analysis report due to the VP of Sales in 45 minutes. The report cites specific market share percentages, a quote from a named industry analyst, and three statistics about competitor pricing — none of which the manager has independently verified, and there's no time to research each one from scratch before the meeting.

Prompt

Role: You are a rigorous fact-checking editor whose job is to catch errors before they reach a client, executive, or publication — not to praise or rewrite the work.

Context: I generated the following content using AI: [PASTE AI-GENERATED CONTENT HERE]. This will be used for: [DESCRIBE THE STAKES, e.g. "a client-facing report", "an internal board memo", "a public blog post"]. I do not have time to manually verify every claim, so I need you to systematically flag what needs checking before I send this.

Task: Go through the content section by section and produce a structured audit that:
1. Lists every specific factual claim (statistics, dates, names, quotes, citations, technical specs) as a separate item
2. For each claim, classify it as: VERIFIABLE (a specific source could confirm or deny this), UNVERIFIABLE AS STATED (too vague to check, e.g. "many experts agree"), or SUSPICIOUS (internally inconsistent, suspiciously precise, or matches a known pattern of AI-fabricated detail)
3. For any claim classified SUSPICIOUS, explain specifically why — e.g. "this statistic has no attributed source and doesn't match [known figure] from general knowledge", or "this quote is attributed to a real person but cannot be verified"
4. Flag any internal contradictions between different sections of the content
5. Note any claims stated with unwarranted confidence (absolute language like "always", "never", "proven" around claims that are actually contested or uncertain)

Constraints:
- Do not simply rewrite or improve the prose — your only job is verification triage
- Do not assume a claim is true just because it sounds plausible or specific — AI-generated fabrications are often highly specific and plausible-sounding
- If you genuinely cannot assess a claim without external lookup, say so explicitly rather than guessing
- Prioritize the claims that would be most damaging if wrong, given what's at risk: [DESCRIBE WHAT'S AT RISK, e.g. "client trust", "legal exposure", "public credibility"]

Output Format:
## Verification Priority List
[Ranked list of the 3-5 claims most critical to verify before sending, with why]

## Full Claim Audit
[Table: Claim | Classification | Notes]

## Internal Contradictions
[List, or "None found"]

## Overconfident Language
[List of phrases with suggested softer alternatives]

## Bottom Line
[One paragraph: safe to send as-is, safe with minor edits, or needs a fact-check pass before it goes anywhere]

Résultat

Verification Priority List

  1. 34% market share claim for Competitor X — This specific figure appears with no cited source and will likely be the first number the VP questions. If wrong, it undermines the entire report's credibility.
  2. Quote attributed to "Sarah Chen, Gartner analyst" — A named, verifiable person is quoted with a specific opinion. If this person didn't say this, or doesn't hold this role at Gartner, this is a serious misattribution risk.
  3. "Competitor pricing dropped 18% in Q3" — A precise, checkable pricing claim. Public pricing pages or press releases may confirm or refute this quickly.

Full Claim Audit

ClaimClassificationNotes
"Competitor X holds 34% market share"SUSPICIOUSNo source cited; figure is suspiciously precise for a private company with no public market-share disclosures. Cross-check against IDC or Gartner public reports before using.
Quote from "Sarah Chen, Gartner analyst"SUSPICIOUSCannot confirm this person exists at Gartner or made this statement. Models frequently generate plausible-sounding attributed quotes that are fabricated. Verify via Gartner's published research or remove the attribution.
"Competitor pricing dropped 18% in Q3"VERIFIABLESpecific and checkable against the competitor's public pricing page or press releases.
"The market is rapidly consolidating"UNVERIFIABLE AS STATEDToo vague to check as written; would need a specific timeframe and metric to become verifiable.
"Competitor Y has been the clear leader for years"SUSPICIOUS (overconfident)Absolute language for a market described elsewhere in the same report as "highly competitive" — internally inconsistent framing.

Internal Contradictions

The report calls the market "highly competitive with no dominant player" in the intro paragraph, then later states Competitor Y has been "the clear leader for years." These two framings can't both be accurate — pick one and verify it before sending.

Overconfident Language

  • "proven to be the top choice" → "reported by multiple sources to be a leading choice"
  • "will definitely continue" → "is likely to continue, based on current trends"

Bottom Line

Not safe to send as-is. The market share figure and the Gartner analyst quote are both high-risk fabrication candidates that would be embarrassing if wrong in front of the VP of Sales — verify those two first, even if nothing else gets checked. The pricing claim is quick to confirm via public sources. Budget fifteen minutes for these three checks before sending.

La plupart des personnes qui demandent à une IA de « revérifier ceci » reçoivent en retour un message indiquant que le contenu semble correct — car une demande générale de vérification n'est pas une vraie tâche, c'est une demande de réassurance, et c'est précisément ce qu'un modèle de langage est enclin à fournir lorsque l'instruction ne l'oblige pas à faire autrement.

Ce Prompt remplace cette demande vague par une véritable procédure d'audit : classer chaque affirmation factuelle selon sa vérifiabilité, expliquer précisément pourquoi les plus risquées le sont, et conclure par un verdict concret plutôt que par un confort générique.

Pourquoi classer les affirmations plutôt que de simplement les « vérifier »

Demander à un modèle de « rechercher des erreurs » ne lui donne aucune tâche concrète à accomplir — il ne peut que relire le texte et signaler une impression vague. Imposer une classification en trois catégories (vérifiable, invérifiable tel quel, suspect) transforme la vérification en une tâche de tri mécanique que le modèle peut réellement exécuter affirmation par affirmation, plutôt qu'en un jugement ouvert qu'il risque d'avaliser sans réfléchir. La catégorie SUSPECT cible spécifiquement le mode d'échec qui compte le plus : les détails fabriqués par l'IA ont tendance à être très précis et plausibles — un analyste nommé, un pourcentage exact, une date spécifique — ce qui les rend d'autant plus dangereux, et c'est exactement ce qu'une vérification vague du type « est-ce que ça semble correct ? » ne détectera pas.

Pourquoi la vérification inter-modèles est particulièrement importante ici

Un modèle invité à auditer sa propre sortie, ou la sortie d'un modèle aux données d'entraînement et aux tendances similaires, sera moins enclin à signaler une fabrication qui correspond à ses propres schémas habituels de spécificité confiante. Effectuer l'audit avec un modèle différent — idéalement issu d'un autre laboratoire — signifie que l'auditeur ne partage pas les mêmes angles morts que le générateur. C'est le levier le plus efficace disponible : le Prompt fonctionne avec n'importe quel modèle, mais il fonctionne mieux entre différents modèles qu'au sein d'un seul.

La contrainte la plus importante : pas de réécriture

L'instruction interdit explicitement au modèle d'améliorer la prose. Sans cette contrainte, les modèles cèdent à leur instinct le plus fort — améliorer le texte — ce qui produit un brouillon soigné plutôt qu'un rapport de risque, et noie discrètement le signal de vérification dont vous avez réellement besoin sous des modifications stylistiques que vous n'avez pas demandées. Séparer « est-ce vrai » de « est-ce bien écrit » permet à l'audit de rester concentré sur la seule tâche pour laquelle il existe.

L'adapter

Le champ « enjeux » dans la section Contexte joue un rôle concret : nommer ce qui est réellement en jeu (confiance d'un client, exposition juridique, décision d'un conseil d'administration) indique au modèle comment prioriser sa liste de vérifications, car une statistique erronée dans un document de brainstorming interne a bien moins d'importance que la même erreur dans un communiqué de presse public. Remplissez-le précisément plutôt que de le laisser vague — plus les enjeux sont clairs, plus le triage sera efficace.

prompt-engineeringAI reliabilityfact-checkingai-hallucination
Partager: