Koppla logo

Fact Checking

Verify clinical claims in your notes against peer-reviewed research with AI-powered fact-checking and detailed verdicts.

The Fact check tab in Notis analyzes clinical claims in your note and checks them against published research. Each claim receives a verdict, an explanation, and source citations. You can jump to any claim in your note and have Chat draft a rewrite of the ones that need work.

Accessing Fact Check

Click Fact check in the editor header while the AI panel is closed, or choose the Fact check tab if the panel is already open.

Running a Fact Check

Write or Finalize Your Note

Fact checking works best on complete or near-complete content. Finish drafting before running a check.

Open the Fact Check Tab

Click Fact check in the editor header, or choose the Fact check tab in the AI panel.

Click Run Check

Click Run check to begin analysis.

Review Results

Claims appear as color-coded cards with verdicts, explanations, and sources.

A check usually takes 15–30 seconds, and you can keep writing while it runs. Notis reads up to about 30,000 characters of your note and checks up to 12 claims per run. If you switch to another note before it finishes, the results are saved to the note you checked.

If a check fails, Notis shows what went wrong and keeps the Run check button available so you can try again.

How Fact Checking Works

The fact-check process uses a multi-step approach to verify claims:

  1. Claim extraction — The system reads your note and lists its factual, verifiable claims (statistics, treatment descriptions, research findings, clinical assertions)
  2. Prioritizing — If there are more than 12, it checks the 12 most important first: safety claims, then treatment, statistics, diagnostic criteria, and general points last
  3. Verification — Each claim is searched for and checked against published sources, including claims that state well-known facts
  4. Results — All claims appear in one list, in the order they appear in your note
The system automatically expands abbreviations and preserves organizational attributions (e.g., "According to the APA...") to ensure accurate verification.

Understanding Verdicts

Each verified claim receives one of six verdict types. These help quickly identify which statements are well-supported and which need attention.

A claim marked Not checked has no verdict yet. See Claims Not Checked Yet.

Evidence Grades

Claims rated Supported, Misleading, Outdated, or False also get an evidence grade: Strong evidence, Moderate evidence, Limited evidence, or Minimal evidence. A one-line note under the explanation names the kind of source and its main limitation.

The grade starts from the type of source:

  • Clinical guidelines, systematic reviews, and randomized trials can be graded Strong. A guideline counts as Strong only when its recommendation is strong. A conditional recommendation, such as "consider…", counts as Moderate.
  • Observational studies, such as cohort studies and surveys, are usually graded Moderate.
  • Expert opinion and guidance from organizations are graded Limited.
  • Case reports are graded Minimal.

Serious concerns, such as small samples or studies that disagree, lower the grade.

The verdict and the grade answer different questions. The verdict says whether the claim is accurate. The grade says how strong the evidence behind it is. "Supported · Limited evidence" means the claim is accurate but rests on weak sources. On a False claim, the grade describes the evidence against the claim, for example Strong evidence against.

Definitions, such as DSM criteria, and descriptions of what an organization or resource offers don't get a grade.

Reviewing Results

Claim Cards

Each claim appears as a color-coded card in the results panel:

  • Verdict badge — Color-coded label (Supported, Misleading, False, etc.)
  • Evidence grade — How strong the evidence is, shown next to the verdict when the claim is graded
  • Quoted text — The claim wording from your note. This may be edited during fact-checking for clarity but preserves the original meaning
  • Explanation — Why the claim received this verdict, with clinical context
  • Evidence note — The kind of source behind the grade and its main limitation
  • Sources — Click the source count to see links to the research that informed the rating

Counts Summary

A summary bar stays at the top of the results while you scroll. It shows the number of claims and a count for each verdict type (e.g., "7 Supported", "2 Misleading"), starting with the verdicts that most need attention. Click a verdict count to show only those claims, and click it again to show all of them.

If any accurate claims are worded more firmly than their evidence allows, the bar also shows a worded too strongly count. Click it to show only those claims.

Claims Not Checked Yet

A claim shows as Not checked when:

  • Your note had more than 12 claims, so the lower-priority ones wait until you check them.
  • The check failed for a technical reason. If the search service was busy, the card says so: "The search service was busy, so this claim wasn't checked…"

To check them, click Check on a single card. To check up to 12 at once, click the button in the row under the summary bar, such as Check 12. The row also says how many claims haven't been checked yet.

If you edit your note after the check, these buttons turn off and the row says Re-run to check the edited note.

The Fact check tab label also shows a count of the claims rated False, Misleading, Outdated, or Unsupported, so you can see how many need attention without opening the tab.

When Your Note Changes

If you edit your note after a fact check, a warning at the top of the results says Your note has changed since this check. While the warning shows, the Re-run button is highlighted. The warning stays when you reload the page or open the note on another device.

Fact checks run before this update don't show the warning. Run the check again to turn it on for that note.

Running the Check Again

Click Re-run in the summary bar to check the note again after you revise it. A new check replaces the previous results.

Taking Action on Results

Find a Claim in Your Note

Click the locate icon (Find in note) on a claim card. Notis selects the claim in your note and scrolls to it. If the fact check reworded the claim, Notis selects the paragraph that matches it most closely. On phones and small screens, the AI panel closes so you can see the highlighted text.

If Notis can't find a close enough match, it shows Couldn't find this claim in the note. This usually means the text was edited after the check, or the claim was reworded when it was extracted.

Fix a Flagged Claim

Every checked claim that isn't rated Supported has a Fix button:

Click Fix

Notis selects the claim in your note and opens the Chat tab.

Review the Suggested Rewrite

Chat receives the claim, its verdict, and the explanation, and drafts a rewrite that's consistent with current evidence. Read it carefully and compare it with the sources on the claim card. Ask follow-up questions if you need a different version.

Swap It In

Choose Add to note → Replace selection on the reply to replace the original claim. Don't click in your note first, or the selection moves and the reply replaces the new selection instead.

If Notis couldn't find the claim in your note, Chat still drafts a rewrite. Select the original text in your note yourself before choosing Replace selection.

Soften a Claim Worded Too Strongly

A Supported claim can still be worded more firmly than its evidence allows. For example, it may state as settled something only a few studies suggest. These cards have a Soften button.

Soften works like Fix: Notis selects the claim and opens Chat. Chat rewords the claim with a single qualifier, such as "can help" or "research suggests", so it matches the evidence.

Revising Claims Yourself

You can also revise flagged claims by hand. Read why the claim was flagged, use Find in note to jump to it, and apply appropriate changes:

  • Add qualifiers to provide context ("research suggests..." instead of definitive statements)
  • Update statistics with current figures
  • Include citations from the provided sources
  • Remove claims that cannot be adequately supported

Using Results in Chat

Your latest fact-check results are available as context in the Chat tab while the Fact check button at the bottom of the chat box is on. Chat also sees each claim's evidence grade and uses it to keep suggested wording no firmer than the evidence allows. Claims that weren't checked aren't included. After running a fact check, switch to Chat and ask:

  • "How should I revise the misleading claims?"
  • "Can you suggest better wording for the flagged statistics?"
  • "What qualifiers should I add to make the unsupported claims more accurate?"

When to Run Fact Checks

Best timing for fact checks:
  • After completing the first draft
  • Following major content revisions
  • Before sharing or publishing content
  • After adding new statistics or research findings

Best Practices

Write complete drafts first: The system works best with substantial content. Running checks on partial notes may miss important context.

Pay attention to statistics: Numbers, percentages, and prevalence figures are the most common source of misleading claims. Always verify these carefully.

Don't ignore misleading verdicts: These often reveal opportunities to add important nuance that makes content more credible and clinically responsible.

Review every rewrite: Fix and Soften give you a draft, not a verified answer. Check it against the sources before you replace the original claim.

Consider the audience: Some claims that are technically unsupported may be appropriate as clinical observations in certain contexts. Use professional judgment when deciding how to address flags.

Combine with Research: If a claim is flagged, use the Research tab to find current sources that can help you revise the claim accurately.

Limitations

The fact-check tool analyzes claims based on available academic literature:

Works best with:
  • Factual statements about research findings
  • Clinical statistics and outcome data
  • Therapeutic technique descriptions
  • Evidence-based treatment approaches
May have difficulty with:
  • Very recent research published within the last few months
  • Proprietary therapeutic approaches with limited published evidence
  • Organizational claims (e.g., specific statistics attributed to PSI or MHA websites)
  • Claims requiring deep contextual knowledge of specialized therapeutic modalities
Professional judgment should always be applied when reviewing results. The fact-check tool compares claims with published sources. It doesn't guarantee accuracy or replace your clinical judgment, and clinical expertise is essential for interpreting results in context.