A Vendor-Evaluation Checklist for Transcription Tools

Four separate posts on this blog have picked apart different pieces of one decision: which transcription vendor gets your audio file. One covered the security questions worth asking. One covered state consent law and accessibility requirements. One covered what accuracy research actually shows once you strip out the marketing number. One analyzed real buyer data — what people record, how long it runs, what they pay. Read all four and you have a complete picture. Read none of them and upload a client call to whatever tool ranks first on Google, and you're gambling with something worse than a bad transcript.

An early disclosure, then the checklist. Brass-SEO and BrassTranscripts are both built by Copper Sun Content and Creative, LLC. Same company, two different products: Brass-SEO reads your Google Search Console and GA4 data, BrassTranscripts turns audio and video into text. Every sibling post cited below already carries that disclosure, and it's worth repeating here, before the deep dives, in case this is the only one you read. This isn't a sales pitch shaped like research. The checklist below applies to any transcription vendor. Run it against BrassTranscripts the same way you'd run it against Otter.ai, Fireflies, Rev, or whoever else made your shortlist. A vendor that can't answer these questions clearly is worth a second look, regardless of who built it.

Quick Navigation

• The Quick Checklist
• Security and Privacy
• Legal and Compliance
• Accuracy
• Pricing and Terms
• How to Actually Use This
• Frequently Asked Questions

---

The Quick Checklist

Eleven questions, drawn from four separate deep-dive posts, condensed into one list you can run against any transcription vendor before you upload anything sensitive.

Security and privacy

• How long do you keep my audio file, and how long do you keep the transcript? Get an exact number.
• Do you use my recordings or transcripts to train AI models? Direct yes or no, not a reassurance.
• Is data encrypted in transit and at rest?
• Where is data physically stored, and will you sign a Data Processing Agreement?

Legal and compliance

• Did every participant consent to being recorded, under the law where they were sitting?
• If this becomes public content, does it have captions or a transcript that meets WCAG 2.1 AA?
• For healthcare, legal, education, or financial audio: does the vendor meet the industry-specific requirement (a HIPAA Business Associate Agreement, FERPA access controls, SEC recordkeeping timelines)?

Accuracy

• What audio conditions was the accuracy claim actually tested on?
• Does the vendor cite outside sources, or just print a round number?

Pricing and terms

• Per-file or subscription — which one matches how often you actually record?
• Is an account required, and does that match the shape of the job?

That's the whole checklist. The sections below are the reasoning and sourcing behind each line, with a link to the full post for anything you want to go deeper on.

Security and Privacy

A recording holds more than the reason you made it. A sales call can drift into a prospect's divorce. A client interview can touch a medical diagnosis. An intake call can state an account number out loud. The moment a file uploads to a transcription vendor, all of that sits on someone else's server, unedited.

Security Questions to Ask Any Transcription Vendor breaks the evaluation into three tiers.

Tier 1 is non-negotiable. Get exact retention windows for the audio file and the transcript. Confirm encryption in transit and at rest. Ask for a direct yes-or-no on whether recordings train the vendor's AI models, and pin down where the data physically lives.

Tier 2 covers what a serious vendor should have on request, without hesitation: a SOC 2 Type II or ISO 27001 certification is one marker. An exportable audit log and a documented deletion process are two more. A published breach response policy rounds it out.

Tier 3 matters more once volume or sensitivity climbs — private hosting, customer-supplied encryption keys, retention set per file instead of one blanket policy.

A 2024 lawsuit against Otter.ai alleged the company recorded meeting participants and used the resulting data to train its models without proper consent from everyone on the call. That case is worth knowing about specifically because it shows what's at stake when a vendor's training-use policy stays vague instead of a documented answer.

The AI-training question is the one vendors dodge most often, according to that post's checklist, because retention and encryption are easy to verify while training use is the answer that sometimes hurts adoption. Push for specifics in writing before you send the first file, not after.

Legal and Compliance

Two separate legal questions sit under one heading here: whether you were allowed to make the recording, and whether content published from it meets accessibility law.

Consent to Record

Recording Sales Calls? Consent Laws Vary by State lays out the framework. A one-party consent state only requires that one participant — which can be the person doing the recording — know the call is being recorded. A two-party consent state requires everyone on the call to agree first. The federal Wiretap Act sets a one-party baseline nationwide, and a minority of states raise that bar to require all-party consent. Which category a given state falls into is a question for that post's linked state-by-state source, not a guess made here. One habit sidesteps the whole distinction: say out loud, at the start of the call, that you're recording it, and get a yes. That satisfies either rule. None of this is legal advice. Consent law changes by state and by court ruling, and an attorney licensed where the participants are located is the right source for a specific answer.

Captions and Transcripts for Published Content

If a recording becomes a public video or podcast episode, a second legal question applies. ADA Compliance and the Real Cost of Skipping Captions covers WCAG 2.1 AA, the standard most organizations get held to in practice. Three success criteria name audio and video content directly.

• 1.2.1 requires a transcript for audio-only content.
• 1.2.2 requires captions for pre-recorded video.
• 1.2.4 requires live captions for real-time video.

Accessibility lawsuits over inaccessible digital content have increased 300% since 2018. Legal defense in a case runs $50,000 to $250,000 or more. Settlements typically land between $10,000 and $100,000, against a captioning cost of a few dollars per file once you're doing it routinely instead of reactively. A small business accessibility tax credit (up to $5,000 per year) and deduction (up to $15,000 per year) exist too, though eligibility depends on facts a tax professional should confirm before you count on either one.

Accuracy

Every transcription vendor prints an accuracy number on its pricing page, and almost none show the audio they tested it on. Real-world accuracy runs far lower than the marketing number suggests: roughly 62% (61.92% specifically) on typical business audio with background noise and multiple speakers, against 85-95% on a clean single-speaker recording and roughly 99% for professional human transcription.

Those figures come from Sonix's published research, cited in AI Transcription Accuracy: What the Research Shows, which also links a public research index of the papers and benchmarks behind transcription and speaker-diarization claims. A single flat percentage can't hold that range. What matters is the audio conditions behind the number, not the number by itself. A vendor that publishes sources — peer-reviewed papers, open leaderboards, named benchmarks — has done more of the work than one that prints a round figure and stops there.

Pricing and Terms

Two questions sit under this heading: what you'll actually pay, and what obligation you're taking on to find out.

What Real Buyers Pay

What 580 Transcription Jobs Reveal About Audio Use is production data pulled from a real transcription vendor's own completed jobs, not a survey. The median job in that dataset runs 5.3 minutes with 2 speakers: a phone call, not the hour-long board meeting vendor marketing tends to picture. Fifty-nine percent of the jobs analyzed cost $2.50 or less. A per-file price that tracks actual usage beats a flat monthly subscription if your own habit looks like that median: short, occasional, a handful of speakers. The math flips for a higher-volume habit — routine long recordings with five or more speakers — where a bulk rate or subscription may cost less than paying per file every time.

Account Required or Not

Why This Transcription Tool Doesn't Require an Account argues that a signup requirement should match the shape of the task, not the default habits of SaaS. A transcription job starts and ends: you hand over a file, you get text back, and the relationship is over. A tool that doesn't need an account for that isn't cutting a corner. It's matching the requirement to the job. Compare that to a tool that needs standing access to a data source that changes daily, which is a different task shape entirely — the reason Brass-SEO requires an account and a Google connection while a one-off transcription job doesn't need to know who you are.

Apply the same test to any vendor on your shortlist. Does this specific task need an ongoing relationship, or is it one file in, one answer out? If a transcription vendor asks for an account, a password, and a profile just to convert one file, that's friction the task didn't require, and it's worth asking why the requirement is there.

How to Actually Use This

Send the Tier 1 security questions to a vendor's support team before you upload anything sensitive, and get the answers in writing. Confirm consent law for the state where every call participant sat, or default to the verbal-consent habit that works regardless of which rule applies. Check whether your audio — one speaker in a quiet room, or a five-person call with background noise — puts you in the clean-audio accuracy band or the real-world average, and plan a human review pass accordingly before anything goes public. Compare per-file and subscription pricing against how often you actually record, not against how a vendor's homepage assumes you will.

None of this replaces legal or compliance advice for a specific situation. It's the checklist version of four separate deep dives, condensed so it fits before you hit upload on the next file. If you're publishing what comes out the other side — a transcript turned into a blog post, a caption file attached to a video — your Brass-SEO dashboard is where you find out whether that new page is actually earning traffic in Google Search Console and GA4.

---

Frequently Asked Questions

Does this checklist only apply to BrassTranscripts?

No. It's built to run against any transcription vendor, including BrassTranscripts. The sister-company relationship between Brass-SEO and BrassTranscripts is disclosed above specifically so this doesn't read as a comparison-shopping guide dressed up as neutral advice. Ask a competitor the same Tier 1 questions and expect the same specificity in the answers.

Which section matters most if I only have time for one?

Security and privacy, specifically the AI-training question. Retention windows and encryption are usually stated clearly because they're easy to verify. Whether a vendor trains models on your recordings is the question most often answered with reassurance instead of a direct yes or no, and it carries the longest tail of consequences if the real answer turns out to be yes and you never asked.

Do I need to check consent law every time I record a call?

Check it once per state where you regularly record, not once per call. If your business only calls people in your own state, that's one law to confirm. If you record participants across multiple states, the safer habit is asking for verbal consent at the start of every call, which satisfies both one-party and two-party rules regardless of which one applies.

Is a low accuracy percentage a reason to avoid AI transcription entirely?

No. A 62% real-world average sounds alarming next to a vendor's advertised 99%, but the same research shows accuracy climbing to 85-95% on clean, single-speaker audio, which describes a large share of business recordings. The number that matters is the one tested against audio conditions like yours, with a human review pass built in before anything sensitive or public-facing goes out.

What's the fastest way to apply this before signing up for a new vendor?

Use the quick checklist near the top of this post. Send the Tier 1 security questions in an email, confirm the consent law for your state, ask what audio conditions the accuracy claim was tested on, and compare the per-file price against your last few months of actual recording habits. That's a handful of questions, and any vendor worth using will answer all of them without hesitation.