The Gatekeeper Call: Auditing a Marketing Asset List Before RAG Ingestion

Marketing Academy · Field Work●AI in Marketing

All projects·From: Custom GPTs & Internal Marketing Knowledge Bases

MiniAudit· 25 minutes

Awfis Space Solutions

Objective: Given a raw list of 20 candidate documents pulled from a shared drive, decide which should be ingested into a RAG knowledge base as-is, which need cleanup first, and which must be rejected outright.

You're the content lead at Awfis Space Solutions, the flexible workspace operator, scoping the first internal GPT for the marketing team. Someone dumped 20 files from a shared drive into a folder and asked you to 'just upload it all.'

Sort the 20 files into Ingest Now, Clean First, and Reject using the lesson's freshness and relevance rules, not gut feel.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets

FreeSort and tag the 20-file audit list

Free, no account friction, filters and columns are all this audit needs

The process

1 step

Step 01 of 01

Auditing marketing assets for RAG-readiness before ingestion

The lesson's asset-prep stage says RAG quality depends on what you feed it: remove outdated brand books and failed campaign reports, keep current style guides, high-performing copy, and verified case studies, and tag everything with metadata so retrieval finds the right context.

Given filename, document type, and last-modified date for 20 files, which ones are safe to ingest today, which need cleanup, and which should never enter the knowledge base?

Google Sheets— Paste the file list into Sheets with columns: filename, type, last modified, status.

Procedure

Import the 20-row file list and freeze the header row
Flag anything over 18 months old or superseded by a newer version as Reject
Flag failed-campaign postmortems and legacy product sheets as Reject regardless of age
Flag current style guides, verified case studies, and top-performing copy as Ingest Now
Flag anything Ingest-worthy but missing a channel/date metadata tag as Clean First

Sample output

AWFIS ASSET AUDIT (20 files)

INGEST NOW (9)
  brand-voice-guide-2026.pdf
  top-10-linkedin-posts-q2-2026.docx
  customer-case-study-verified-fintech-client.pdf
  ...6 more

CLEAN FIRST (6)
  meta-ads-swipe-file.xlsx  (missing channel/date tags)
  ...5 more

REJECT (5)
  brand-guide-v3-2022.pdf  (superseded, 3 versions old)
  diwali-campaign-postmortem-2023-FAILED.pdf  (failed campaign)
  ...3 more

Healthy

Reject pile is dominated by outdated brand books and failed-campaign reports, not recent work.

Unhealthy

A 2022 brand guide and a flagged-as-failed campaign report both sit in the Ingest Now pile.

What this means

Freshness and outcome, not file size or polish, decide what a RAG system is allowed to learn from.

So what do I do about it?

Symptom	Action	Effort
An outdated brand book slipped into Ingest Now	Add a hard age cutoff column and auto-flag anything past it for manual review	5 min

YouYou can do this yourself, no engineering access required.

Final deliverable

A triaged 20-file list split into Ingest Now / Clean First / Reject, with a one-line reason for every Reject.

See a reference example

Sample output

Zomato, marketing knowledge base pre-ingestion audit (excerpt)

INGEST NOW
  2026-brand-voice-and-tone-guide.pdf
  top-performing-swiggy-comparison-ad-copy.docx

CLEAN FIRST
  influencer-brief-template.docx  (add channel + audience metadata)

REJECT
  ipl-2023-sponsorship-recap-UNDERPERFORMED.pdf  (failed campaign, do not train on it)
  brand-guidelines-v2-2021.pdf  (superseded by 2026 version)

Success criteria

You're done when you can:

Every Reject has a specific, non-vague reason tied to age or outcome
Nothing tagged Ingest Now is a failed campaign or a superseded document version