Import contacts

A spreadsheet becomes useful when your team can find its people in the workspace and trust their details. Import a CSV or pasted list into a review batch, check the proposed matches and save the contacts you intend to keep.

Before you start: use a workspace you are authorized to update. Map columns prepares CSV, TSV and flat JSON without a model key, AI credits or a model request. AI-assisted is an optional choice for free-form lists and requires an enabled model and budget. Check Setup if that optional path is unavailable.

1. Prepare your list

Use clear column names. This fictional CSV is a starting example:

fullName,email,phone,companyName,website
Alex Morgan,alex@example.test,,Example Company,https://example.test
Jamie Chen,jamie@example.test,,Example Company,https://example.test

A reviewed contact needs a name and at least an email address or phone number. An email must be valid. A supplied website must be a valid domain or HTTP(S) address. Include a business website when you want to identify a company reliably.

Give every CSV or TSV column a distinct header and keep the same number of columns on each row. A batch accepts up to 500 contact rows, 40 columns, 100 characters per header, 250,000 source characters, and 2,000 characters per source cell. Uploaded files also have a 250 KB limit, so split a larger export into smaller files before preparing it.

For JSON, use flat contact objects or a list of those objects. Numeric values keep their original digits as source text, which protects long identifiers and phone numbers. Paste a free-form list as one person per line, or upload it as a TXT file, and keep any analysis guidance within 1,000 characters.

2. Prepare the review

Open Contacts, choose List import, and upload the file or paste its contents. Map columns is selected initially. Choose Review columns to inspect suggested column matches. Choose the contact field for each source column you want to import. Choose a full-name column, or separate first and last names, and at least email or phone. Each source column can supply only one contact field. Unmapped columns stay in Original source.

Choose Build contact review. This path copies mapped source values, validates them and checks existing identities. It makes no model request, including when AI is paused or no provider is connected. A suggestion is only a column-name match; review it before proceeding.

For example, map fullName to Full name, email to Email, and companyName to Company in the sample above. An unfamiliar heading such as Customer can be selected manually. If your source changes, choose Review columns again so older mappings cannot apply to a different file.

For notes or a free-form list, explicitly select AI-assisted, add any cleanup context, and choose Analyze with AI. This sends the source to the configured model under the workspace's cost controls. An unavailable model does not trigger a hidden fallback to another import method. Analysis cannot approve or import contacts.

Both paths preserve every accepted source row in its original order. Invalid or ambiguous rows remain visible and start excluded. When AI omits a row, its original source still appears so you can complete or exclude it.

3. Review the rows

Check names, email addresses, phone numbers, company information, and each row's import decision. Correct errors and exclude rows you do not want. Inspect possible matches rather than assuming similar names identify the same person.

Large batches show 50 source rows per page. Use Next page and Previous page to review the rest. Row numbers, selected counts, saving and approval still cover the complete batch.

If a row has no proposed contact, open it and fill in the name and contact method only when the source supports them. Edited fields have length limits; shorten an oversized value when the review tells you which field exceeded its limit.

Company information can remain unlinked when there is not enough evidence to identify the business. Resolve warnings deliberately; do not invent missing fields.

4. Confirm the import

Save your review changes, then choose Review approval. Inspect the exact snapshot and choose Approve and import only when it is correct.

A save commits every reviewed row and its history together. A failed save leaves the previous review and approval unchanged. A successful edit clears the earlier approval. If another operator saved or started importing the batch, reload the latest review before saving. Already imported rows keep their original receipt and cannot be edited through a retry review.

5. Verify the result

Check the batch result and open a sample of the accepted contacts. Failed rows remain available with a reason.

What to check

Compare the source row count with the review row count before approval; they should match. If analysis rejects repeated headers, broken quotes, missing columns, nested JSON, or control characters, correct the export and try again. Split a batch when its complete source exceeds the analysis budget, even if the file fits the upload limit. The AI context budget applies only to AI-assisted preparation. Inspect the existing batch before resubmitting the same file. Use separate actions when you want to send email, enroll contacts in a campaign, or create an opportunity.

Correct a row from its original source

Each review row keeps its source row number. Open Original source to compare the untouched source fields with the proposed contact. It opens automatically for rows needing review, including rows the AI omitted. Correct the fields, choose whether to include the row, and save the review before approving its exact snapshot.

Uploads must contain valid UTF-8. An incorrectly encoded file is refused with a correction message, rather than replacing unreadable characters. Save it as UTF-8 and upload again. Recent import history is cached across workspace navigation; if a refresh fails, retry history while keeping the current review available.

Try the fictional demo

Column mapping uses your pasted fictional data and saves its review in this browser session. Approving new, distinct contacts creates simulated contacts only. Existing-contact updates start excluded and require a connected workspace. The demo makes no model request and sends no data to live business APIs.