Source tiers: from primary docs to rumors
Goal: Tier every claim in a brief before it mixes into external materials—no tier, no ship.
Best for: Research briefs, competitor notes, class reading, anything AI-assisted.
Not all links are equal. Primary docs beat random blogs. Rumors are not “Label every claim by source strength before you” Teach the model your scale, then force every claim to wear a tier.
Why beginners skip this
Skipping source tiers feels light—rumor and primary papers get cited the same. Tag the tier first.
Step 1 — Four tiers
Learn the four tiers first; untiered claims do not mix into external materials.
| Tier | Examples | How to treat |
|---|---|---|
| 1 Primary | Official docs, contracts, filings, your analytics | Can decide from these |
| 2 Reliable secondary | Reputable reporting that cites sources | Useful; still open links |
| 3 Weak | Random blogs, undated posts, marketing pages | Hypothesis only |
| 4 Rumor/meme | Unsourced social claims, screenshots without provenance | Do not ship as fact |
Step 2 — Ask the model to tier
Have the model tier only the links you pasted—do not let it invent sources from memory.
For each factual claim in your brief, label tier 1–4 and name the source.
If you cannot name a source I provided or a URL I can open, mark tier 4 / Unknown.
Do not invent citations to climb tiers.
Example
Filled prompt
Grade: (1) gov PDF (2) vendor blog (3) Twitter screenshot of a chart.
What may enter client brief? Forum/Twitter never primary.
Label A/B/C beside each link; C-tier ≠ A-tier policy.
What good output looks like
- Prices from the page sit at tier 1
- News interpretation may be tier 2
- Fancy unsourced stats drop to 4
- You can argue from the table in a meeting
Step 3 — Pair with the matrix
Feed tiers into the claim–evidence matrix status column; Contested often means cross-tier sources fighting.
Example
Filled prompt
Claim set: pricing page screenshot (tier1?), blog roundup (tier?), anonymous forum post (tier?).
Grade each source tier and say what I am allowed to put in an external brief.
Ban treating forum posts as primary.
What good output looks like
- Claim pack lists sources for “Label every claim by source strength before you de”
- Primary tab opened before share
- Unknowns labeled; no vibe citations
- Pass/fail: one primary URL saved
Keep it real
Run this guide’s Practice now on work you will actually ship this week—not a fictional scenario. Dated notes beat perfect theory.
Label three links today
A/B/C on links you already have. Refuse Tier C as policy.
Do not bookmark-only
Finish Practice now, then tick this guide on your path. Titles unread are not skills.
Reality check
Open this guide beside a live task. If you cannot name the task, pick one before reading further.
Close the loop
Write one line: next time a claim enters the body with no source tier, I refuse.
Example
Filled prompt
Bundle: (1) court PDF (2) news rewrite of court PDF (3) Discord screenshot.
Grade tiers for an external client brief. What is banned?
Label A/B/C beside each link; C-tier ≠ A-tier policy.
What good output looks like
- Primary tab opened before share
- Unknowns labeled; no vibe citations
- Pass/fail: one primary URL saved
- Claim pack lists sources for “Label every claim by source strength before you de”
Red flag
Acting on C-tier social as policy; untiered claims in a brief—tier before decide.
Save this card
Card: Tier every claim A/B/C before deciding—a C-tier social post is never A-tier policy. Keep this beside the tool; read before you start.
Wear-test: show your Save card to a teammate for thirty seconds—if they cannot restate the refuse line, tighten the card tonight and date it (Asia/Shanghai).
Before you start
Open a blank claim pack. One claim about “Label every claim by source strength before you” plus a slot for a primary URL. Chat citations are not enough. Guide: research-source-tiers. Keep the real job visible the whole time; a blank box means stop and fetch the artifact first.
How to use this week
This week write a source tier (primary / secondary / rumor) beside five real claims; rumors stay out of the body.
Close the week by quoting aloud the refuse line you actually used once. If you cannot, rerun Practice now and date the pass/fail (Asia/Shanghai).
Pitfalls
| Pitfall | Fix |
|---|---|
| Viral = reliable | Tier by provenance |
| Model-invented “tier 1 papers” | Only sources you can open |
| Skipping tiers on AI summaries | Tier every claim |
Example
Filled prompt
Temptation: skip the habit for research-source-tiers to save five minutes.
Job still: Label every claim by source strength before
Smallest correct pass: Done visible · checkable artifact · Unknowns [?].
Boundaries: draft only; no silent send; no invented metrics.
Return: artifact + refuse-line said aloud.
What good output looks like
- Names the shortcut refused (specific to research-source-tiers)
- Still delivers checkable output for “Label every claim by source strength before”
- Unknowns labeled; draft-only
- Pass/fail: refuse-line dated (research-source-tiers #4)
Practice now
Tier legend: paste tier definitions on Save once (Asia/Shanghai).
Upgrade try: spend 5 min upgrading one secondary to primary (Asia/Shanghai).
Social last: social sources cannot be sole evidence (Asia/Shanghai).
Primary hunt: 10 min Asia/Shanghai to find one primary or park (Asia/Shanghai).
Downgrade drill: demote one source a tier with a one-line why (Asia/Shanghai).
Tier drill: label three sources primary/secondary/social for one claim (Asia/Shanghai).
Thin-spot drill: rewrite the Save card with real nouns from your work (repo, doc, inbox). Generic placeholders fail the drill.
Tonight write a source tier (primary / secondary / rumor) beside five real claims; rumors stay out of the body.
Ship-check: artifact path ___ · refuse-line dated Asia/Shanghai ___ · never-do ___
One sentence takeaway
Primary / secondary / rumor three tiers; rumors stay out of the body.
Next
来源层级:从一手文档到传闻
目标: 给简报里的每条主张标来源层级;没有层级的主张,不许混进对外材料。
适合: 定价页、新闻、口头数字搅在同一段里的调研与竞品跟踪者。
定价页价格通常是 1 级;新闻解读可能是 2 级;无来源的花哨统计掉到 4 或 Unknown。不在你粘贴材料里的,就标 Unknown,别硬补链接。
为什么新手跳过
来源不分等级像省事;一级传闻和一级论文会被同等引用。先贴等级。
步骤 1——四级
先认四级来源;没有层级的主张,不许混进对外材料。
| 层级 | 例子 | 如何对待 |
|---|---|---|
| 1 一手 | 官方文档、合同、备案、你的分析后台 | 可据此决定 |
| 2 可靠二手 | 有出处的靠谱报道 | 有用;仍要点开 |
| 3 弱 | 随机博客、无日期帖、营销页 | 仅作假设 |
| 4 传闻/梗 | 无来源社交主张、来路不明截图 | 不要当事实发出 |
步骤 2——让模型分级
让模型只对你粘贴的链接分级;别让它凭记忆补来源。
对简报里每条事实主张,标层级 1–4 并点名来源。
如果无法对应我提供的来源或我能打开的 URL,标 4 / Unknown。
不要编造引文来爬层级。
实例
填好的提示词
给这篇 Vendor X 定价简报重新分级。我粘贴的来源:定价页 + 一篇新闻。
没有层级的主张不许出现。不在我粘贴材料里的 → Unknown/4。
边界:不编造;未知标 [?];发送前停下。
完成检查:一分钟内能判过/不过。
像样的输出大概长这样
- 定价页上的价格标成 1 级,并写清「为什么够一级」的理由
- 媒体新闻解读标成 2 级,同时写明降级的具体原因
- 没有任何来源的花哨统计直接掉到 4 级或 Unknown
- 整张表拿去会上可以跟人争论「依据到底够不够」
- 规则守住:粘贴材料里没出现的数字 → 一律 Unknown
步骤 3——与矩阵配对
把层级喂进主张–证据矩阵的状态列;Contested 常常是跨层级来源在互相打架。
实例
填好的提示词
把这篇指南用到我的真实情况:[research source tiers]。边界:不编造事实。请返回产物,以及我应跑的 3 项检查。
完成检查:一分钟内能判过/不过。
像样的输出大概长这样
- 产物贴合你刚刚粘贴的真实分级任务,不另起炉灶编场景
- 边界仍然写着「不要编造事实或补假链接」
- 文末附三项出门检查:有没有缺层级?缺链接?口头数字?
- 检查项短得像塞进口袋的清单,开会前扫一眼就行
- 读完能立刻打开自己的简报改一版
- 一分钟内能对照清单判过/不过;缺项标红再继续
保持真实
用本周真要交付的工作做「现在就练一次」,别编假场景。带日期的实操笔记,比完美理论有用。
今天给三个链接分级
对已有链接标 A/B/C。拒绝把 C 级当政策。
不要只收藏
先做完「现在就练一次」,再在学习路径上打勾。只收藏不练,标题变不成技能。
现实核对
把这篇指南放在真实任务旁边。说不出任务,就先挑一件再往下读。
闭环
写下一行:下次主张不写来源等级就进正文,我拒绝。
红旗
来源不分等级,一级传闻和一级论文会被同等引用;先贴等级。
保存这张卡
记下:一张主张–证据或来源分级表的核对日。标上日期。
开始前
聊天前用一句话写下主张。
本周怎么用
本周对五条真主张各写来源等级(一手/二手/传闻);传闻不进正文。
常见坑
| 坑 | 改法 |
|---|---|
| 无层级主张混进简报 | 不在材料里 → Unknown/4 |
| 新闻当 1 级 | 定价页等一手更高 |
| 花哨统计无来源 | 直接掉到 4 |
现在就练一次
今晚就练:对五条真主张各写来源等级(一手/二手/传闻);传闻不进正文。
一句话带走
一手/二手/传闻三档;传闻不进正文。
下一步
← Back to home · Learn · Tools · Compare · Start