Research Stop Rules: End the Brief Before You Invent the Stat

Fake statistics rarely come from bad intentions. They come from open-ended research. A brief says “research the topic,” the writer researches until the deadline, and the draft gets filled with whatever was found last. That’s usually a number from a vendor blog, quoted by an aggregator, quoted by an AI answer, with the original study three hops away and possibly never existing at all.

The fix isn’t “research harder.” It’s to research less, more precisely: decide in advance which claims the piece needs, what evidence each one requires, and when to stop looking. That’s a stop rule, and it belongs in the brief, before anyone opens a search tab.

This guide covers the claim inventory, a three-tier source standard, the stop rule itself, and the four dispositions that settle every claim before drafting begins.

Why the brief is the right place to stop

By the time a draft exists, every unsupported claim has a sentence built around it, and cutting it means rewriting the paragraph. Writers resist that, editors miss it, and the stat survives. At the brief stage, a claim is one line in a table. Striking it costs nothing.

There’s also an AI-specific reason. Google’s guidance on generative content notes that these models work by predicting likely word sequences, can produce plausible-sounding but inaccurate output, and that manual fact-checking and review are critical (Google: using generative AI content). A brief with no claim boundaries invites a model, or a rushed human, to fill the gaps with fluent fiction. A brief with a claim inventory gives everyone a fence.

Step 1: Inventory claims before research

Before any research, list every claim the piece needs to make and sort each one by load:

• Load-bearing: the argument collapses without it. “Most service businesses lose leads to slow response times” is load-bearing in an article about response-time automation.
• Supporting: strengthens the argument, but the piece survives without it.
• Decorative: color, scale, or “did you know” material. Nice to have, never needed.

Most briefs give all claims equal research time. That’s backwards. Load-bearing claims deserve most of your effort. Decorative claims get almost none, and they’re the first to go when evidence is thin.

A good brief usually has two or three load-bearing claims, a handful of supporting ones, and decorative claims kept in reserve. If you list more than three load-bearing claims, the piece is probably trying to make two arguments. Split it.

Step 2: Set a source standard per tier

Define three tiers of evidence:

• T1, primary: the original study, dataset, official documentation, statute, or standard. The organization that measured or decided the thing.
• T2, credible secondary: reputable reporting or analysis that cites T1 directly and links to it.
• T3, everything else: aggregator stat pages, uncited blog posts, social posts, and AI-generated answers.

Then the rules:

T3 never supports a published claim on its own. It’s a lead: a pointer to where T1 might live. An AI answer that cites “a 2023 study” is a lead until you’ve opened the study.

Step 3: Write the stop rule

The stop rule has three parts, and all three go in the brief:

1. Time box per claim. Load-bearing: up to 30 minutes. Supporting: 10. Decorative: 3. These are illustrative defaults. Calibrate them to your rates and turnaround, but write numbers down.
2. Hop limit. Follow citations at most two hops from where you found the claim. If the trail goes cold at hop two, the claim is unsourced, however many pages repeat it.
3. Stop condition. Research for a claim ends at the first of three events: it meets its tier standard, the time box expires, or the hop limit is reached.

The time box isn’t there to save money, though it does. It forces a decision. Without it, a writer who can’t find the study keeps searching, finds a T3 page that “basically says the same thing,” and calls it sourced. With it, the writer reaches a clean fork: the claim is sourced or it isn’t, and the brief already says what happens next.

Three tests every source must pass

Meeting a tier standard is necessary, but not sufficient. Before you mark a claim Source, run three quick tests:

1. The quote test. Paste the exact sentence from the source that supports the claim. If you can’t find a sentence that says it, only one that “implies” it, the source doesn’t support the claim as written. Rewrite the claim to match the sentence, or downgrade it.
2. The scope test. Check who and what the source measured. A survey of enterprise IT buyers doesn’t support a claim about local service businesses, and a study of one country doesn’t speak for another. Scope mismatch is the most common way a real statistic becomes a false claim.
3. The date test. Record when the source was published and when you accessed it. Platform documentation, regulations, and market data change. A claim about how a crawler behaves needs a source from this year. A claim about how combustion works doesn’t.

These tests take a minute each, and they catch more misused statistics than any amount of extra searching.

Step 4: Assign a disposition to every claim

When research stops, each claim gets exactly one of four dispositions:

• Source: met its tier standard. Record the URL, the exact supporting sentence, and the date accessed.
• Soften: didn’t meet the standard, but the underlying observation is sound. Rewrite it as experience, a pattern, or a hypothesis: “In our audits, slow response times are a frequent theme,” not “78% of leads are lost to slow response.”
• Strike: decorative or unsupportable. Remove it, with no regret.
• Ask: the evidence exists, but only the client has it. Their own data, their own policy, their own process. Send a specific question with a deadline.

The Ask disposition is the most underused tool in content production. Clients are often sitting on the strongest evidence in the piece (their own response times, their own warranty terms, their own process steps) and nobody asks, because the writer assumes the internet will provide. A brief with three precise Ask questions produces better content than one with ten aggregator stats.

Step 5: Guard against claim creep

Claim inventories fail in one predictable way: new claims appear during drafting. The writer, mid-paragraph, adds “and this is why 60% of businesses…” because the sentence felt like it needed weight.

The rule: any claim that isn’t in the inventory gets flagged in the draft and goes through the same process (load, tier, time box, disposition) before the piece moves to editing. In practice, most creep claims are decorative and get struck in seconds. The point is that they get decided, instead of slipping through on momentum.

A useful check for the editor is to count the factual claims in the draft and compare that number with the inventory. If the draft has more, something crept in.

Worked example: an HVAC brief

Illustrative example. A brief for a regional HVAC company’s article, “When to repair vs. replace a furnace.”

Research time, roughly an hour, all of it spent on two claims. The finished piece carries two sourced facts, one client insight no competitor can copy, one honest conditional, and zero invented numbers. It’s also more persuasive than the version with five stats, because the client’s own pattern data is the most specific thing on the page.

The brief section to copy

Add this block to every brief, right after audience and angle:

CLAIM INVENTORY
# | Claim | Load (LB/S/D) | Tier needed | Time box | Disposition | Source URL + quote | Date
STOP RULE
Time boxes: LB 30 min / S 10 min / D 3 min. Hop limit: 2.
Stop at first of: standard met / time box expired / hop limit.
ASK LIST (to client, due before drafting)
1.
2.
CREEP RULE
Claims not listed here are flagged in the draft and inventoried before edit.

It’s a small block. It changes what research means: from “find things to say” to “decide what can be said.”

Accelerate the whole workflow

Organa Obscura’s Content-Brief Toolkit is built around this discipline. Its 13-page guide walks through a five-stage workflow with explicit research stop criteria and a handoff QA method. The workbook (Google Sheets or Excel) has a Client Intake tab that collects the evidence only clients have, a Research Checklist of 15 tasks, an eight-section Content Brief, an 18-check Handoff QA tab, and an annotated example for a fictional studio, Cedar & Kiln, that shows the whole flow filled in. Markdown templates are included if your briefs live in Notion, Obsidian, or Docs.

To speed up the research and QA steps, the SEO & Content Prompt Pack adds 120 fill-in prompts, including categories for keyword and topic research, briefs and outlines, and QA and fact-checking. It works with whatever AI tool you already use, and comes with a Notion-ready CSV.