FREE COURSE · BEGINNER
Source checking

Check an AI answer before you trust it

A checked paragraph with each important claim linked to evidence or marked unconfirmed.

About 20 minutes4 practical stepsFree to read
A website displayed on a laptop at a bright wooden desk.

Before you start

A notes app and the two fictional source extracts in this lesson. No AI account or paid research tool is required.

Stay on the free route

This is a source-checking exercise, not a promise that every answer can be verified. The practice sources are invented and labelled. Use real primary sources when checking a real claim.

Original practice lesson · Official references checked 17 October 2026 · Button names may differ between app versions.

Your progress

0 of 4 steps complete.

Start first step ↓
STEP 1 · ABOUT 4 MIN

Break the answer into claims

An answer may mix supported facts with guesses. Check individual claims rather than giving the whole paragraph a single confidence score.

  1. Read the fictional answer below and split it into three separate factual claims.
  2. Mark words such as always, unlimited and free, because they can hide conditions.
  3. Create a table in your notes: Claim | Source | Supported, contradicted or unconfirmed.
Practice example
Fictional answer to inspect: “SketchPad Mini allows five exports each month, includes transparent backgrounds on the free plan and lets you cancel a paid plan whenever you want.”

Claim 1: five exports per month
Claim 2: transparent backgrounds on free
Claim 3: cancel whenever you want
STEP 2 · ABOUT 6 MIN

Read what the source actually says

A link or citation is a starting point. The source must support the specific claim and the right plan or version.

  1. Compare each claim with the two fictional source extracts. These describe an invented product, not a catalogue entry.
  2. Record which extract supports or contradicts each claim. Do not turn missing information into confirmation.
  3. For a real product, open the provider’s relevant plan or help page and check the date, region, tier and billing term. A search snippet alone is not enough.
Practice example
Fictional source A — SketchPad Mini plan card, practice version 1:
Free: five PNG exports per month, standard background only.
Paid: transparent-background export is included.

Fictional source B — account help, practice version 1:
A paid subscription renews at the end of its billing period.
This extract does not explain cancellation.
STEP 3 · ABOUT 5 MIN

Decide what can stay

Keep supported facts, correct contradicted ones and label the rest as unconfirmed.

  1. Check your results against the worked review below. If yours differs, identify the exact sentence that supports your decision.
  2. Rewrite the answer so a reader can distinguish the free and paid features.
  3. Do not ask the same AI to reassure you that its original answer is right. Ask for evidence, then inspect that evidence yourself.
Practice example
Worked review:
Five exports: supported by source A.
Transparent background on free: contradicted by source A.
Cancel whenever you want: unconfirmed; source B does not say.

Checked version: “The practice free plan includes five standard-background PNG exports per month. Transparent backgrounds require the practice paid plan. Cancellation terms are not confirmed by these extracts.”
STEP 4 · ABOUT 5 MIN

Leave a useful evidence trail

A small record helps you repeat the check when the underlying information changes.

  1. Save your revised paragraph next to the claim table. For a real check, also save the exact source URL and the date you read it.
  2. Choose one thing that would make you recheck it, such as a new plan, changed feature or old source date.
  3. Write “not enough evidence” when that is the honest outcome. Then try the research courses with your own non-sensitive topic.
TRY IT YOURSELF · ABOUT 5 MIN

A source supports less than the answer claims

A fictional plan card says “Free: five standard PNG exports per month. Paid: transparent backgrounds.” An answer says “Free users get five transparent PNGs each month.”

Original fictional scenario · choose an answer, then read the explanation.

Which conclusion matches the source?

See a worked example and review checklist

Supported: five standard PNG exports per month. Contradicted: transparent backgrounds on free. Checked wording: “The practice free plan includes five standard-background PNG exports each month.”

  • I separated the allowance from the feature claim.
  • My wording keeps the plan distinction.
  • I have not added a claim the source does not support.

Try it with a new example: For your next real check, record the exact source, relevant plan and date. Mark anything the source does not answer as unconfirmed.

This checks one practice decision, not your real work. Answers save only in this browser when storage is available.

MAKE THE LESSON YOURS

Keep your practice notes

Record what you made, what you checked and what you want to improve. These notes stay in this browser; they are not sent to us or synced to another device. Clearing browser data removes them.

Optional · up to 2,000 characters.

YOUR NEXT SMALL STEP

Put your practice to work

Repeat the task with a fresh brief. Check the result before using it for a real project.

Choose your next skill →

Part of Get confident with AI · Research with evidence.

READ THE BACKGROUND GUIDANCE

NIST: Generative AI Profile

Section 2.2 explains confidently incorrect output and invented citations. Our fictional evidence exercise is independent teaching material.

Open official learning ↗

Sources and lesson notes

The practical brief and exercises are ours. Official references support the product guidance; we do not host the providers’ course content.

Checked 17 October 2026. Tool plans can change; consult the provider before choosing an upgrade. These beginner exercises have not been presented as certified or exhaustive product testing.

Progress saves in this browser. No site account needed.

THE OTHER FEW

Find the few things worth knowing.

Explore AI tools, software and technology through official sources, clear trade-offs and practical comparisons, so you can find the few that matter to your task.

Popular searches:

🔥 Fresh Few

Recently added to our catalogue, with a reason to explore each one.

💎 Hidden Gems

Editorial alternatives with a specific reason to consider them.

📊 Popular Categories

More from The Other Few
Personal Few

Build the few tools that work together.

Describe the outcome, not the software. The Other Few turns the job into a practical stack, keeps overlaps visible, and explains the trade-offs.

Stack suggestions are task-fit decisions, not product rankings. Commercial relationships must never change compatibility.
READY

Tell us the outcome.

We'll break it into jobs, choose the fewest sensible tools and show the trade-offs.

Evidence-first: current pricing, feature availability and platform support should be shown only when verified from an appropriate source.
THE OTHER FEW · INTELLIGENCE LAYER

From task to decision.

The shortlist now explains the fit, the trade-offs and what to do next.

PERSONALISED FIT
✦

Task understanding

Translate natural-language goals into task signals before matching products.

GoalUse caseConstraints
◎

Compatibility

Show why a tool fits this search instead of presenting a permanent product score.

PlatformExperiencePriority
↗

Decision path

Move naturally from shortlist to profile, alternatives, compare and saved tools.

ExploreCompareSave
DISCOVERY

Go beyond the shortlist.

Browse editorial collections when you want to explore rather than solve one task.

Fresh FewOnly products with a genuine freshness signal.Explore →
Hidden GemsInteresting products that deserve a closer look.Explore →
Popular CategoriesStart with the job, then narrow to the Few.Browse →
VERIFIED SOURCES · EDITORIAL FIT · NO PAID RANKING
RECOMMENDATION GRAPH

See why the Few connect.

The recommendation layer connects tasks, constraints, products, alternatives, workflows and evidence without turning the catalogue into a popularity ranking.

GRAPH ACTIVE
YOUR TASKgoal + constraintsTASKFITTOOLEVIDENCE intentexperienceplatformpriorityproductalternativesourcefreshness
EXAMPLE DECISION PATH
1 · GoalBuild a website for a small business.
2 · ConstraintsBeginner · web · low-code · ease.
3 · Candidate setReplit · Lovable · Claude — narrowed before presentation.
4 · Trade-offLovable favours simplicity; Replit offers more control; Claude is broader rather than purpose-built for the build.
Compatibility is task-specific. It is not a permanent product rating.

Graph signals

These are the relationships the engine can reason over as the catalogue grows.

Goal / intentConstraintTask fitAlternativeWorkflow overlapEvidenceFreshnessSaved context
Important: the graph does not invent evidence. Unsupported claims remain unsupported until verified.
PERSONAL DECISION LAYER

Make the decision smaller.

Bring your goal, saved tools, workflows and fresh changes together. The Other Few narrows the decision without pretending there is one universal winner.
● Decision context ready
Launch a businessBuild a websiteCreate videosResearch & decide
GoalWaiting
ConstraintsWaiting
Your FewLocal saves
ChangesFreshness ready
Start with the outcome.

We'll turn it into a smaller, clearer decision using the intelligence already built into The Other Few.
Fit is contextual. Editorial judgement, evidence and commercial relationships remain separate.Private on this device.
Intelligence Core
Make the catalogue smarter, not just bigger.
The intelligence layer now treats products, tasks, constraints, alternatives and workflows as connected evidence. Expansion is staged into verification lanes rather than filling the catalogue with unverified entries.
18 core records · verification-first
18Core records
7Categories covered
14Relationship types
0Fabricated records

Connected catalogue

Core products are connected to the tasks and categories they are already structured to serve.

Expansion lanes

These are research lanes, not unpublished product claims. A lane becomes catalogue data only after source verification and taxonomy review.
Editorial rule: catalogue breadth never outranks evidence quality. A product can enter the research queue without entering the public Few.
Intelligence Core

One brain. Every decision.

The catalogue, Finder, Decision Engine, workflows, Your Few and freshness signals now have a shared conceptual layer. The goal is not more results — it is better decisions with fewer tools.
● CORE MODEL READY

Decision graph

The same product can play different roles depending on the user's outcome and constraints.
18core products
7active categories
0saved tools
0watched tools
GoalOutcome
IntentTask
ConstraintsFit limits
ProductsCatalogue
DecisionFew