Choose one of three approaches. The steps for that option appear below.
Choose an option above. On a phone, swipe the tabs left or right to see all options →
Claude runs normalisation code; you approve its rules before grouping and check every proposed survivor.
Collect the contact CSV
Normalise contact fields with code
Check and approve rules
Draft duplicate groups and survivor
Check groups and record decisions
Use: You
Have ready: contact CSV, for example 6,000 contacts with name, email, phone and company; retain original IDs.
Use: Claude
Ask Claude for normalisation rules, per-field change counts, changed-row samples and a working table retaining original IDs.
What to paste into the tool
Task: Normalisation report: rules applied, per-field change counts, sample of changed rows; Normalised working table with original IDs retained. Using code, normalise the fields for matching: trim spaces, case-fold the codes and strip stray punctuation. Keep every original ID and value alongside the normalised form - change nothing in the source data. Show me the exact normalisation rules you applied, per-field change counts, and a sample of changed rows.
Use: You
Check the rules report and changed rows; approve or adjust the rules before requesting groups.
Data owner approves the normalisation rules before any grouping; original IDs must be retained and nothing deleted. Check the normalisation rules on 20 sample contacts so that distinct people are not made to look the same.
Use: Claude
Ask Claude for a downloadable review file: members, matching fields, reason, confidence tier and proposed survivor per group.
What to paste into the tool
Task: Duplicate-group review file: group members, matching fields, match reason, confidence tier, proposed survivor per group. Using the approved normalised table and rules, group likely duplicates. For each group list the member rows with original IDs, the fields that matched, a plain-language match reason, a confidence tier (high or review), and a proposed survivor with the reason. Produce this as a downloadable review file. Merge nothing and delete nothing.
Use: You
Compare matched fields and proposed survivors with source facts; record approvals or rejections.
Check matched fields and proposed survivors against the source CSV, covering every high group and a sample of the rest. Approve or reject each group before any merge happens. Do not delete or merge records in the source system; original IDs are retained throughout.
You hand-write sample answers; Vizly normalises and groups contacts, then you export the review file and code.
Write expected values for 20
Normalise contacts
Check sample results before grouping
Draft groups
Export files and record group
Use: You
Have ready: contact CSV, for example 6,000 contacts; hand-write normalised names, emails and phones for 20 contacts, including near-miss distinct people.
Use: Vizly
Ask Vizly to normalise the CSV, preserving original IDs, and show its rules and changed-row samples.
What to paste into the tool
Task: Normalisation rules applied; Normalised working view with original IDs retained. I uploaded a contact CSV where codes stand in for names, emails and phones. Normalise the values for duplicate matching: trim and case-fold them, strip punctuation. Keep every original ID and original value next to the normalised form. Show me the rules you applied and a sample of changed rows. Delete nothing.
Use: You
Test Vizly's rules against your 20-contact sample; approve or return them for adjustment.
Approve the rules before any grouping.
Use: Vizly
Ask Vizly for matching fields, reasons, confidence tiers, survivor proposals and generated analysis code.
What to paste into the tool
Task: Duplicate-group review with matching fields, match reasons, confidence tiers, proposed survivor per group; Generated analysis code for the record. Using the approved normalised view, group likely duplicates. For each group list member rows with original IDs, matching fields, a plain-language match reason, a confidence tier (high or review), and a proposed survivor with the reason. Also give me the generated analysis code so the logic is on record. Merge nothing and delete nothing.
Use: You
Export the review file and generated analysis code promptly; record approve/reject decisions and leave the source system untouched.
Export the review file promptly: the vendor page stated, as of 2026-10-28, that uploaded data is deleted permanently after one hour of inactivity, with a footnote that Premium subscribers can opt into longer retention windows. Approve or reject each group before any merge happens; do not delete or merge records in the source system.
You write and test the rules; ChatGPT applies them and proposes groups for your decision.
Write and test the matching
Apply the rules and propose
Check groups and record decisions
Use: You
Have ready: contact CSV, for example 6,000 contacts with name, email, phone and company; write matching rules and confidence-tier definitions.
You define the matching rules; original IDs must be retained and nothing deleted. Test rules on 20 sample contacts.
Use: ChatGPT
Ask ChatGPT to apply only your rules and produce the review file with original IDs, matching fields, reasons, tiers and survivor proposals.
What to paste into the tool
Task: Duplicate-group review file: group members with original IDs, matching fields, match reason, confidence tier, proposed survivor per group. Attached is a contact CSV where codes stand in for names, emails and phones, and here is my matching rule set. Apply only these rules: normalise as specified, group records that meet the rules, and for each group list member rows with original IDs, the fields that matched, the rule that fired as the match reason, the confidence tier from my definitions, and a proposed survivor. If a group fits no rule cleanly, put it in a 'needs review' bucket instead of forcing it. Merge nothing and delete nothing.
Use: You
Check every group against your rules and facts; record approvals or rejections in the final file.
Do not delete or merge records in the source system; original IDs are retained.
Best when: You want code-run normalisation with rules and change counts.
What it does: Claude normalises and groups; you approve rules and groups
Strong at: Runs normalisation as code and returns a downloadable review file.
Limit: Record matching lacks operation-specific evidence; check every group and survivor.
Free plan: Yes, limited usage windows
Best when: You want to test normalisation against a hand-checked 20-contact sample.
What it does: Vizly normalises and groups; you test samples and record decisions
Strong at: Returns generated Python or R alongside the duplicate-group review.
Limit: Duplicate detection unconfirmed; export before one-hour inactivity deletion unless retention enabled.
Free plan: Yes, limited to 10 messages a month
Best when: You want matching decisions traceable to rules you wrote.
What it does: ChatGPT applies rules; you write rules and approve groups
Strong at: Applies your rules and separates groups that fit no rule.
Limit: Bad owner-written rules still need checking; operation-specific application is unconfirmed.
Free plan: Yes, limited uploads and data analysis
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between ChatGPT and Claude for this? See the comparison.
Data owner approves the normalisation approach before grouping.
Data owner approves or rejects each group before any merge happens.
Original IDs are retained and nothing is deleted.
No deletion or merge in the source system.
Check the result against your original inputs. Approve it before sharing or acting.