What should you remove from a call transcript before reuse?

Which details identify a person even without a name?

Combinations of ordinary details do. A job title, a company size, and a city can narrow a quote to one person after the name is gone.

The direct identifiers are the easy part: every name said on the call, emails, phone numbers, the employer, and anything that looks like an account, bank, or passport number. Search for each one instead of reading for them, because a colleague's name said once in minute 34 is easy to skim past.

Indirect details take longer. A line such as "I run revenue at a 40-person freight company up north" names nobody, and a reader who works in freight may still know exactly who said it.

The UK ICO's anonymisation guidance, published on 28 March 2025, asks whether identifying someone is "reasonably likely" and checks that against a "motivated intruder", someone who sets out to work out who is behind the data. For a marketing team the intruder is easy to picture, usually the speaker's competitor or manager.

Is a pseudonym the same as anonymous?

No. Swapping a name for "Customer A" is pseudonymisation, and the ICO treats it as a weaker step than anonymisation that stays under data protection rules.

The difference matters because a marketing team almost always holds the way back to the person. The recording sits in the call tool and the CRM shows which account the call belonged to. So a quote credited to "a customer" is pseudonymous inside your company, and at best anonymous to readers outside it.

Underneath sits a second question. The ICO says personal information can be reused for a new purpose only if that purpose is compatible with the original one, and a lawful basis is still needed for the new purpose. How that applies to your calls is a question for your data protection adviser, and permission from the people on the call is covered in our post on recorded sales call consent.

Can you trust automatic redaction?

Trust it as a first pass and nothing more. Amazon's own documentation says its redaction may miss sensitive data and tells you to review the output.

Amazon Transcribe's PII redaction documentation is unusually direct about the limits. It warns that the feature "may not identify and remove all instances" of sensitive data, recommends reviewing what comes out, and says it does not meet HIPAA de-identification requirements.

Dialpad's PII redaction, announced on September 28, 2023, removes party names, email addresses, bank details, passport numbers, social security numbers, and credit card numbers from the transcript and the call summary. A job title and a city are not on that list, and neither is the story only one customer could tell.

In Corvan, calls stay private unless you say otherwise, and when our agents write from one, names and figures are left out and any that slip through are flagged before you approve the draft.

What should you keep, and where should the raw transcript live?

Keep the cleaned excerpt with a note of who cleaned it, and keep the raw recording in one place that only the people who need it can open.

Copies multiply fast. Someone pastes the transcript into a doc to pull quotes, the doc lands in a shared drive, and an AI writing tool gets a copy too, each with every name in it. We would pick one home for raw files, limit who can open it, and work from cleaned excerpts everywhere else.

Speaker identification deserves its own line in that process. On 13 August 2026 a federal judge in the Northern District of California let federal wiretap, California privacy, and Illinois biometric privacy claims against Otter.ai proceed, as UC Today reported. The plaintiffs allege the assistant captured voiceprints through speaker identification, and those are allegations, not findings. A voiceprint does not live in the transcript text, so a redaction pass over the words never reaches it.

How do you check a quote before it is published?

Read it the way a motivated intruder would, and cut until nothing in it narrows the field to one person.

Teams skip this step when a post is due.

1. Search the excerpt for every name and number in the full transcript, including ones the redaction tool left alone.
2. Read it for combinations of role, company size, region, and any event specific enough to have made the news.
3. Read the caption too, since "from a call with a fintech CFO last week" puts back what the quote took out.
4. Ask a colleague who knows the market who said it, and if they can guess, cut more.
5. If the quote only works because of the detail that identifies the speaker, drop the quote.