Hallucination Corrections

Overview

Hallucination Corrections helps you identify and address factual inaccuracies that AI models state about your brand. When an AI model makes an incorrect claim — such as wrong founding year, incorrect pricing, or misattributed features — this feature generates correction templates tailored to each AI provider and tracks your submissions through to resolution.

The correction workflow integrates with AEO Optima's accuracy checking system. When accuracy checks flag inaccurate attributes, you can generate and submit corrections directly.

Supported Providers

Correction Workflow

Each correction follows a four-stage lifecycle:

1. Draft

A correction template is generated with the following details pre-filled:

• Brand name and contact context
• Incorrect claim — The exact text the AI model stated incorrectly
• Correct information — The verified fact from your brand facts
• Authoritative source — A URL the provider can reference for verification
• Date observed — When the error was captured in a snapshot

The template is formatted as a professional correction request letter.

2. Submitted

After you submit the correction through the provider's feedback channel, mark it as submitted. The system records the submission date for tracking purposes.

3. Resolved

When subsequent snapshots show the error has been corrected, mark the correction as resolved. The resolution date is recorded.

4. Reappeared

If a previously resolved error resurfaces in new snapshots, the status can be set to "reappeared" to track recurring issues.

Provider-Specific Instructions

Each generated template includes specific instructions for the target provider:

• OpenAI — Use the thumbs-down button on the specific ChatGPT response, then paste the template in the feedback text area.
• Google — Use the "Report a problem" link in Gemini responses and paste the template as the detailed description.
• Anthropic — Use Claude's in-interface feedback button. For persistent issues, contact support with the template.
• Perplexity — Click the "Report" button next to incorrect answers and provide the template as context.
• Microsoft — Use the feedback button in Copilot/Bing Chat. For formal corrections, submit through Bing Webmaster Tools.

How to Use

1. Run an accuracy check or review Entity Analysis results to identify inaccurate claims.
2. Navigate to Hallucination Corrections in the sidebar.
3. Create a new correction by specifying the incorrect claim and the correct information.
4. Generate templates for one or all providers. The system creates a tailored correction request for each.
5. Submit the correction through the provider's feedback channel using the provided instructions.
6. Update the status as the correction progresses through the workflow.
7. Monitor subsequent snapshots to verify the error has been corrected.

Correction Statistics

The dashboard tracks aggregate stats across all corrections:

• Draft — Corrections generated but not yet submitted.
• Submitted — Corrections sent to providers, awaiting action.
• Resolved — Errors confirmed as fixed in AI responses.
• Reappeared — Previously resolved errors that have returned.

Tips for Effective Corrections

• Include authoritative sources — Corrections with URLs pointing to official documentation or press releases are more likely to be acted upon.
• Be specific — Quote the exact incorrect claim and provide the precise correct information.
• Submit to all relevant providers — If ChatGPT gets your founding year wrong, check Gemini and Claude too.
• Follow up — If a correction is not addressed within 4--6 weeks, resubmit with updated snapshot evidence.
• Keep brand facts current — The accuracy checking system relies on your brand facts to detect errors automatically.

Plan Requirements

Hallucination Corrections is available on Pro-SME and above.