All articles
RELIABLE WORK

How to tell a sourced claim from an assumption

6 minutes read

How to tell a sourced claim from an assumption
Summary

In an AI draft, a line backed by a real source sounds just as confident as a line the model filled in. Here is how to tell the two apart before review.

Share this article

A sourced claim in an AI draft comes from a specific document, dataset, or approved source you could open and check right now. An assumption is something the model filled in because the source material did not say it directly. On the page the two look the same, and the hard part is separating them before a reviewer signs off.

Why the two are hard to tell apart in an AI draft

Fluency tells you nothing about sourcing, and a draft that reads well gives you no way to know which one you are looking at. A model often completes a sentence so that it sounds like a confident answer, whether or not the evidence supports it. In the finished prose, "the source states this" and "this is a reasonable guess" look identical, and a reader cannot recover the difference afterwards.

This happens with any tool whenever a draft sounds sure regardless of the evidence behind it. The firm that publishes the draft carries the risk, because the client sees a finished paragraph and not the reasoning behind it.

What separates a sourced claim from an assumption

Run each claim through these checks before it goes to review.

CheckSourced claimAssumption
Can you open the source?Yes, it points to a specific document or datasetNo source, or a vague reference like "industry data"
Does the source say this exact thing?The source states the claim, not something adjacentThe claim extends or generalises beyond what the source says
Is it marked?Shown as sourced, with the source attachedShown as an assumption, flagged for the reviewer to resolve

The middle row matters most. A real source can still be stretched. A report says usage grew in one region, and a paragraph later the draft claims usage grew everywhere. The citation is genuine, but the claim attached to it is not what the report said.

What a properly flagged assumption looks like

An assumption is not a problem by itself. Every draft fills gaps the source material does not cover, and a firm's own judgement often belongs there. The problem is an assumption dressed up as a fact, which leaves the reviewer unable to tell the difference without redoing the research.

To flag an assumption, write down plainly what is assumed and why, next to the claim itself, not in a footnote or left implicit. "Assuming the client keeps its current headcount through Q3" is a sentence a reviewer can approve, adjust, or reject in seconds. If a paragraph relies on the same assumption without naming it, the reviewer has to reconstruct the reasoning from scratch, which takes longer than writing the section themselves.

A worked example

Take a competitive-landscape section in a market research draft. The source material is a set of company websites and two industry reports the team approved for this engagement.

A sourced claim, drafted correctly: "According to [Report X], three of the five largest vendors in this category added a mobile app in the past year." The sentence names the source and says only what the report says. A reviewer can open the report, find the line, and move on.

An assumption, drafted correctly: "Assuming smaller vendors follow the same pattern within 12 to 18 months, based on how this category has adopted features historically." Nothing here is presented as a fact from the report, because it is not one. It is a forecast the team chooses to make, named as such.

Most failures fall between the two. Suppose the draft says "the market is rapidly moving toward mobile-first products". That sounds more confident than either version above, yet it has neither a source nor a named assumption. A reader takes it as coming from the report, and the report never said it. A reviewer has to catch this sentence, and it is easy to miss when the draft does not mark which lines are citations and which are forecasts.

A short check before a draft goes to review

Before any AI-assisted draft moves to review, run it through this sequence:

  1. Pull every specific claim: a number, a date, a named fact, a comparison.
  2. For each one, ask whether it traces to a real source you can open, or whether it is filling a gap.
  3. Confirm the source says what the claim says, not a nearby or broader version of it.
  4. Check that assumptions are marked as assumptions, in the text, not just in your head as the reviewer.
  5. Resolve or flag anything that fails steps 2 through 4 before it reaches the client.

The check is quick if the draft was built with it in mind and slow if it was not. When no evidence is attached, a reviewer has to do this from memory or from a fresh search, for every claim, every time. Most firms are already paying that cost without naming it.

Where this fits in a firm's workflow

None of this replaces expert review, but it changes what the reviewer does. If the evidence is attached to each claim, they check it. If they start from an unsourced paragraph, they do the research again, with the model's confident phrasing in front of them instead of their own. The reliable client work guide covers the full loop: approved sources in, visible evidence throughout, expert review before delivery.

Once a firm has a review pattern that works for one type of deliverable, turn it into something repeatable instead of rebuilding it on the next similar brief. That is the subject of reusable AI workflows.

See Hebno on your kind of work.

A short walkthrough, built around what your team delivers.