Why citations stay manual — and the three-skill pipeline that fixes it
An executive brief makes dozens of factual claims. Each one needs a source. Doing it by hand is slow and error-prone. Here's how three small AI Skills in SharePoint turn that grind into a single conversation.
Takeaway 1
Citation work is mechanical — perfect for an AI assembly line.
Takeaway 2
An AI Skill bundles a complex prompt into one reusable verb.
Takeaway 3
Three skills in sequence model the whole research workflow: list, extract, cite.
The citation problem
An executive brief is full of confident assertions — "the global athletic apparel market will reach $350 billion by 2027" — and each one needs a source. Tracking down the backing data, opening the right report, copying the exact passage, and pasting it next to the claim is the kind of slow, mechanical work that erodes a whole afternoon.
Citation work is high-effort, low-judgment. That makes it the ideal first target for automation: the human still owns the argument, but the assembly-line steps move to AI.
The video opens with a real brief about a company called Zava. It has many assertions. The backing data already exists — it just lives in a folder of separate files. The author already knows where the evidence is. The work is connecting claim to source, one by one.
People assume the bottleneck is finding sources. In this workflow, the bottleneck is the mechanical lookup — and AI in SharePoint is built for exactly that kind of step.
Source material lives in a folder, not the brief
The brief never contains its own evidence. Backing data sits in a separate SharePoint folder containing the supporting reports, market studies, and reference documents. The brief asserts; the folder substantiates.
Keeping these two surfaces separate is the trick. The brief stays clean and persuasive. The folder stays as a research repository. The pipeline's job is to bridge them: read the brief, scan the folder, and write the matches into a third place — a tracking list.
Treat the source folder as the "evidence library." Anything you want the citation finder to cite has to live there. If a fact isn't in the folder, it won't get cited.
What an AI Skill is
An AI Skill in SharePoint is a reusable building block. It wraps a complicated prompt, a target shape, and the steps needed to act on a file or list, then exposes that bundle under a short, memorable name.
Once a skill exists, the user calls it like a verb: "use the assertion extractor skill on this file." The model still does the work, but the heavy prompt engineering — the schema, the constraints, the example outputs — is hidden inside the skill. Anyone on the team can invoke it without re-thinking the prompt.
Skills aren't magic plugins or hard-coded macros. They're authored prompts plus context that the SharePoint AI runtime knows how to execute. You can edit them, swap them, share them.
The three-skill pipeline
Three skills, run in order, handle the whole job:
1. List Creator — provisions a SharePoint list with the columns the rest of the pipeline expects (assertion, citation, source file). Without this, the next steps have nowhere to write.
2. Assertion Extractor — reads the brief and creates one list record per assertion, leaving the citation column empty.
3. Citation Finder — walks every record with an empty citation, searches the source folder, and writes the matching passage back into the record.
Each skill does one job, leaves the result in SharePoint, and hands off to the next. The list is the shared workspace; the skills are the workers.
A common assumption is that one giant prompt could do all of this in a single shot. The decomposition matters: each skill is testable, observable, and re-runnable on its own.
Flashcards — Foundations
01 · Foundations
AI Skill
tap to reveal →
A reusable building block in SharePoint that bundles a complex prompt, context, and target output under a short verb. Invoked by name on a file or list.
← tap to flip back
01 · Foundations
Assertion
tap to reveal →
A standalone factual claim inside a document that needs a source. The unit the extractor breaks the brief into — one record per assertion.
← tap to flip back
01 · Foundations
Citation Problem
tap to reveal →
The slow, mechanical work of matching claims in a document to passages in source files. High effort, low judgment — ideal for automation.
← tap to flip back
01 · Foundations
Source Folder
tap to reveal →
The SharePoint folder holding all backing data. The Citation Finder only searches inside this folder — facts outside it cannot get cited.
← tap to flip back
01 · Foundations
Three-Skill Pipeline
tap to reveal →
List Creator → Assertion Extractor → Citation Finder. Each skill does one job, writes to SharePoint, and hands off to the next.