As of 13 August 2026, AI can only partly find and merge duplicate contacts in your CRM.

What to actually do

1. Use a tool built for this

   The route this page recommends

2. Hand it to a person

   Second choice

   A person who owns the outcome does this end to end, worth it when the failure is dear.

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open your CRM's export function and create a copy containing contact IDs, names, email addresses, phone numbers, companies, activity dates, lead sources, consent fields, notes and any field definitions.
   2. Export the data to a separate file, remove fields that are not needed for matching, and keep the original export unchanged as a backup.
   3. Write down your CRM's merge rules, including which record should normally be retained, how consent history must be preserved and whether shared email addresses are legitimate.
   4. Paste the prompt and the export data into a chatbot, then answer its questions about your CRM rules before asking it to produce the duplicate table.
   5. Sort the results into definite duplicates, possible duplicates and no-merge cases, then compare every proposed match with the original CRM record and its activity and consent history.
   6. Test the process on a small group of approved matches using your CRM's merge or import tools, and compare record counts, field values, notes, activity history and consent status with the backup.
   7. Merge only the approved matches in the CRM, save the AI's table as an audit record, and check a sample of merged contacts before using the data for a campaign.

   Prompt

   I need to identify possible duplicate contacts in a CRM export and prepare a safe merge plan. I will paste the data after this instruction.
   
   Use these fields if present: contact ID, first name, last name, email address, phone number, company, job title, address, last activity date, lead source, marketing consent status and notes.
   
   Rules:
   1. Never merge records automatically.
   2. Treat an exact matching email address as a strong duplicate signal, but flag shared addresses for manual checking.
   3. Use phone number, company and name as supporting evidence only. Do not treat a name or company match alone as proof.
   4. Separate definite duplicates, possible duplicates and records that should remain separate.
   5. Do not invent, normalise or alter contact data.
   6. Preserve the contact ID of the record you recommend keeping and list the IDs of records that could be merged into it.
   7. Recommend which field values to retain only where the source data supports that recommendation. Otherwise write "manual decision needed".
   8. Flag conflicts in consent status, email address, phone number, company, notes or activity history.
   9. Do not expose unnecessary personal data in the explanation. Use contact IDs wherever possible.
   
   Return:
   A. A short description of the matching rules used.
   B. A table with: confidence, keep contact ID, possible duplicate IDs, matching evidence, conflicting fields and proposed action.
   C. A separate list of records needing manual review, with the reason for each.
   D. A checklist for backing up the CRM, testing a small batch, checking the merge result and preserving an audit record.
   E. A list of cases where no merge should be made.
   
   Before making recommendations, ask me for any missing CRM rules that could change the result, including whether shared email addresses are allowed, which record should normally be retained and how consent history must be preserved.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.