As of 13 August 2026, AI can only partly clean up your sales 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 1 hour until you can act on the result.

   How to actually do it

   1. Open your CRM export tool and download a copy of the relevant accounts, contacts, opportunities, owners, stages, emails, phone numbers, activity dates and record IDs without altering the live CRM.
   2. Open your CRM field settings and internal data rules, then gather the definitions for duplicate handling, account ownership, pipeline stages, required fields, formatting and retention.
   3. Remove fields that the chatbot does not need, protect confidential information where possible, and paste the export and rules into a chat using the supplied prompt.
   4. Ask the chatbot to produce the proposed cleaning table, keeping every source record ID and separating confirmed formatting fixes from possible duplicates and judgement calls.
   5. Export the proposed changes and have the CRM owner compare each merge, deletion, ownership change and stage change against the source record and recent sales activity.
   6. Apply only approved formatting fixes and other approved changes to a test list or sandbox first, then compare the resulting records and pipeline totals with the original export.
   7. Send the approved change list to the CRM administrator for the live import, and retain the original export, approval notes and rejected changes.

   Prompt

   Act as a CRM data-cleaning assistant. I will provide an exported sales CRM file and, where available, our field definitions, pipeline stages and data rules. Do not invent values, delete records, merge records, change ownership, or change opportunity stages unless the evidence and rules explicitly support it. Do not infer sensitive personal information.
   
   Analyse the data and return:
   1. A summary of the fields, missing values, inconsistent formats and likely duplicate records.
   2. A proposed cleaning table with the record ID, field, current value, proposed value, reason, confidence, and required human approval.
   3. Separate lists for exact duplicates, possible duplicates, invalid or incomplete records, conflicting company names, inconsistent phone or email formats, stale-looking opportunities, and records needing a CRM administrator.
   4. A safe cleaned version only for formatting changes that are explicitly supported by the supplied rules. Keep record IDs and all original values in separate columns.
   5. A short import plan that explains which changes can be made in bulk and which must be checked individually.
   
   Preserve the source data. Treat a possible duplicate as a warning, not a confirmed merge. Treat a missing value as missing, not as permission to guess. Flag any change that could affect ownership, reporting, consent, customer history or pipeline forecasting. Before producing the final recommendations, state which assumptions you used and which decisions require a salesperson or CRM administrator.
   
   CRM export:
   [PASTE EXPORT OR ATTACH FILE]
   
   Field definitions and data rules:
   [PASTE DEFINITIONS]
   
   Pipeline and ownership rules:
   [PASTE RULES]

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