| Key Takeaway | What it means for your newsroom |
|---|---|
| Free tiers are real, but they are metered | Every tool assessed here offers a genuine no-cost entry point. What separates them is how the limit is applied: monthly minutes, a cap on file length, a lifetime cap on uploads, or a daily allowance. Check the cap that matches your busiest week, not your average one. |
| Pricing model predicts long-term cost | Per-seat subscriptions suit small editorial teams with steady volume. Pay-as-you-go metering suits desks with spiky workloads. A 30-minute interview costs roughly £0.50 on per-minute pricing, so the model matters more than the headline rate. |
| Speaker separation is usually the paid feature | Riverside’s free tier and Good Tape’s free tier both withhold speaker labels. For a two-person interview that is workable. For a panel or a heated council meeting, it doubles your editing time. |
| Language coverage is uneven | Riverside and Good Tape both advertise 100+ languages while Otter.ai covers English, French and Spanish. Engine choice inside pay-as-you-go tools also changes coverage, and some South Asian languages appear in only two of three engines. |
| Data handling is an editorial decision | Good Tape states it does not train models on customer data and supplies a standard data processing agreement. Ask that question of any vendor before a sensitive source recording leaves your building. |
| Transcription belongs inside the publishing workflow | Tools such as Publishrs.com matter because the transcript is raw material for several outputs, and it needs to sit beside the published piece rather than in a separate account nobody can audit. |
An interview ends and the real work begins. A 45-minute conversation with a cabinet minister, a strike organiser or a championship manager has to become a news story, a longer feature, a newsletter item and three social posts, and it has to happen before a competitor publishes the same quotes. Transcription is the step that decides how quickly all of that can move.
That is why AI transcription has moved from a novelty to standard equipment in newsrooms of every size. A survey of the tools available in late 2026, published by Journalism.co.uk, found a crowded field with free tiers on almost every product. The interesting question is no longer whether a free option exists. It is which free option survives contact with a working editorial week.
Pricing has fragmented into three models, and the free tiers are generous in ways that flatter a first test and frustrate a real deadline. The tools that look identical in a demo behave very differently when you are uploading a 90-minute file at 22:00 with a page due at midnight. This guide sets out what the free and low-cost options actually include, how the limits bite, and where transcription should sit in your publishing platform rather than in a side account.
AI transcription pricing has split into three models
Why the pricing model matters more than the feature list
Look past the feature comparison table and the market divides cleanly. The first model is the per-seat subscription: a flat monthly fee for a named journalist, usually with a usage ceiling in minutes. Otter.ai’s Pro tier is £12.75 per user per month, or £6.25 per user per month paid annually, and it carries 1,200 in-app recording minutes. Good Tape’s Pro plan is £16.75 per month, or £13.50 per month billed annually, with 20 hours of transcription a month.
The second model is the metered engine: you pay for what you use, with no subscription and no seat licence. Convert Express charges £0.02 per minute, which puts a 30-minute interview at roughly £0.50 and a 90-minute lecture at roughly £1.50, topped up from a prepaid balance of £4.25. There is no monthly commitment, which makes it easy to trial on a single investigation.
The third model is the freemium funnel, where the free tier is genuinely usable but deliberately uncomfortable. Riverside offers unlimited use of its AI transcriber free of charge with no sign-up, but no speaker separation, no editing and no cloud storage. Done Typing lets you transcribe 10-minute files for up to 10 minutes a day as plain text without an account, then extends to 5 hours a month once you sign in, still free.
The practical implication is straightforward. A desk with five reporters producing steady daily output will usually do better on a per-seat plan, because the volume is predictable. A desk that publishes two long interviews a month and nothing else will waste money on seats and should use metering. Read the pricing shape before you read the feature list, because the feature list is identical everywhere and the cost curve is not.
What the free tiers actually include
Minutes, file length and speaker labels are the real limits
Free tiers are never limited by the number of tools. They are limited by four dials: how many minutes you get, how long a single file can be, how many files you can store, and which formats you can upload. Those dials are set very differently across products that market themselves in almost identical language.
Otter.ai’s Basic plan gives 300 monthly transcription minutes with a 30-minute limit per conversation, plus automated summaries for video meetings. The sting is a lifetime cap of three audio or video file imports, which rules it out as a file-based transcription tool after the first week. Good Tape’s free tier allows three transcriptions a month, each up to 30 minutes, storing three recordings at a time, and it warns that free transcriptions may queue rather than process instantly. Convert Express gives 10 free minutes every day with a preview before you pay.
- Riverside: unlimited free transcription across 100+ languages with no account, accepting MP3, WAV, MP4 and MOV, but no speaker separation or storage.
- Good Tape: three files a month at up to 30 minutes each, built specifically for journalists, with speaker labels reserved for the paid plan.
- Otter.ai: 300 minutes a month but only three lifetime file imports, best treated as a live meeting recorder rather than a file processor.
- Done Typing: browser-based with no install, 5 hours a month once signed in, but the unregistered tier caps files at 10 minutes and outputs plain text only.
- Convert Express: 10 free minutes daily, then £0.02 per minute with no subscription at all.
Add up what a normal Tuesday looks like and the limits start to matter. Two 45-minute interviews plus a 20-minute press briefing is 110 minutes, which clears Otter’s monthly ceiling in a single shift and uses more than half of Done Typing’s monthly allowance. Good Tape’s three free files would be gone by lunchtime.
The honest reading is that free tiers are excellent for evaluation and poor for production. They are how you decide which tool to pay for, not how you avoid paying. Budget for a paid tier and use the free version to test accuracy on your own recordings, in your own accents, before you commit.
Matching the tool to the editorial workflow
Interviews, podcasts and long meetings need different tools
The most common mistake is buying one tool for every job. Transcription tasks inside a newsroom fall into recognisably different shapes, and each shape rewards different features.
For a recorded video interview, Riverside is the natural choice because transcription sits inside the recording studio. It lets you tell the system what you are transcribing, a podcast, an interview, a webinar or video content, and what the output is for, captions, key insights or a social post. The paid plan starts at £35 per month, or £24 per month billed annually, and adds 4K video, filler word removal and podcast hosting. If your video team already edits elsewhere, the free transcriber alone will cover you.
For a live or virtual meeting, Otter.ai earns its place. It transcribes in real time and generates automated summaries from video calls, and the paid tier adds templates, taggable speakers and integrations with customer systems. It supports English, French and Spanish, so it is a reasonable choice for a single-language desk and a poor one for a multilingual newsroom.
For source interviews where the recording itself may become evidence, Good Tape is the closest fit to a journalist’s needs. It was built by and for journalists, supports 100+ languages, and the paid plan adds in-transcript notes, bulk uploads, AI summaries and speaker labels. Its launch feature called Explore lets you query your archive and surface related passages across recordings, which is genuinely useful when you are revisiting a long-running story.
For quick turnaround on a single quote, Done Typing is the fastest route to text. It runs in a browser tab with no app download, no account and no system permissions, which matters on a locked-down office machine. The trade-off is depth: it is a plain transcriber, so expect to do the formatting yourself.
Accuracy, language coverage and data handling
Where the recording is stored matters as much as the transcript
Accuracy gets all the attention and data handling gets none, which is the wrong way round. Every tool here produces a transcript that needs sub-editing, because names, acronyms and regional accents defeat all of them at some point. What varies far more is where the audio goes and what happens to it afterwards.
Convert Express offers a useful model for this. It lets you choose a clean or a verbatim transcript style, and it accepts optional vocabulary hints so names, acronyms and domain terms come back correctly. Verbatim output matters when a quote has to stand up to a legal read, and clean output matters when you are feeding a newsletter. The engine choice also changes coverage: it swaps between three models, one supporting 61 languages, one 60 and one 57, and two of the three cover seven regional South Asian languages that the third misses entirely. A newsroom serving Punjabi, Bengali or Telugu readers should check which engine it is actually using.
Storage is the other dividing line. Good Tape states plainly that it does not train models on customer data and supplies a standard data processing agreement for individuals, which is the kind of sentence to look for in any vendor contract. If your recordings include vulnerable sources, confidential documents being read aloud, or anything covered by a court restriction, the retention policy is an editorial decision and not an IT one.
Ask three questions of any tool before the first sensitive upload: does it train on my data, how long does it keep the audio, and can I delete it on demand. A tool that cannot answer all three in writing is not ready for your investigations desk, however good the transcript looks. For wider context on how newsrooms are handling these questions, the Reuters Institute publishes regular research on AI adoption in editorial teams.
Putting transcription inside your publishing platform
The transcript is the start of the workflow, not the end
A transcript sitting in a vendor account is a stranded asset. The value arrives when it sits alongside the published piece, the notes, the archive record and the analytics for that story. That is the argument for treating transcription as a feature of your publishing platform rather than as a standalone app bought on a card by one reporter.
Consider what actually happens after a transcript lands. Someone lifts quotes into the CMS, someone else cuts a clip for social, a third person checks the legal read, and six months later a colleague searches the archive for the same interviewee. If the transcript lives in a separate tool under one journalist’s login, that last step fails and the earlier steps depend on manual copy and paste. A platform such as Publishrs.com keeps the source material with the published output, so the audit trail survives staff turnover.
There is a commercial argument too. Transcription costs are trivial next to the cost of a missed deadline, but they are invisible when they sit across a dozen personal subscriptions. Consolidating them into a platform budget makes the spend visible and lets you judge it against output. As Press Gazette has documented repeatedly, publishers who can attribute cost to specific content make better decisions about what to keep funding.
The workflow question also shapes which tool you choose. If the transcript has to feed a subscription product, a newsletter and a podcast, then integration and export format matter more than a marginally better word error rate. If it is a one-off quote for a news story, speed wins. Decide the destination first and the tool choice becomes obvious.
A practical way to test before you commit
Run the same recording through three tools
Every vendor demo uses clean audio, a single clear voice and a sympathetic topic. Your recordings are not like that. They have a train announcement in the middle, two people talking over each other and a source who mumbles.
The fastest way to choose is to take one real, messy, awkward recording from your own archive, around 20 minutes long, and run it through three shortlisted tools. Score each on four things: how many names and job titles came back correct, whether the speaker labels were usable, how long the file took to process, and whether the export dropped cleanly into your CMS.
Then check the commercial terms against your actual volume. If you publish 40 interviews a month at 45 minutes each, that is 1,800 minutes, which sits above Otter’s 1,200-minute Pro ceiling and above Good Tape’s 20-hour Pro allowance of 1,200 minutes. At that volume metered pricing at £0.02 a minute comes to roughly £36 a month, which reframes the whole comparison. Do the arithmetic before the trial ends, because the free tier will not tell you what your real bill looks like.
Finally, agree who owns the transcript. If the answer is “whoever recorded it”, you have a retention and compliance problem waiting. A publishing platform built for editorial teams, such as the one from Publishrs.com, gives you a single place to keep that material under a policy you control rather than one set by a consumer app.
Frequently asked questions
Is a free AI transcription tool good enough for daily newsroom use?
For evaluation, yes. For production, usually no. Free tiers cap minutes, file length or lifetime uploads in ways that break on a busy week, and most withhold speaker separation, which is the feature you notice first on a two-person interview.
Which tool has the most generous free tier?
Riverside gives unlimited free transcription with no sign-up across 100+ languages, but without speaker separation, storage or editing. Good Tape allows three files a month at up to 30 minutes each, while Done Typing offers 5 hours a month once you sign in, again at no cost.
How much does paid transcription actually cost?
Good Tape Pro is £16.75 a month, or £13.50 billed annually, for 20 hours. Otter Pro is £12.75 per user per month, or £6.25 annually, for 1,200 minutes. Riverside Pro starts at £35 a month, or £24 annually. Metered tools charge about £0.02 a minute.
Do these tools support languages other than English?
Most do. Riverside and Good Tape both cover 100+ languages, Done Typing covers 50+, and Otter.ai covers English, French and Spanish only. On metered tools the language list depends on which engine you select, so check before a foreign-language interview.
Will the transcript be accurate enough to quote directly?
Not without checking. Every tool in this category mishears names, acronyms and thick accents, and runs of dialogue between overlapping speakers are the weakest point. Treat the transcript as a fast first draft for sub-editing rather than as a quotable record.
What should I ask about data protection before uploading interviews?
Three questions: does the vendor train models on your audio, how long is the recording retained, and can you delete it on demand. Good Tape states it does not train on customer data and offers a standard processing agreement, which is a reasonable benchmark to hold others to.
Should transcription sit in the CMS or in a separate app?
In the platform where possible. Transcripts kept in individual accounts disappear when a reporter leaves and cannot be searched in the archive. Keeping them with the published piece, as a platform such as Publishrs.com allows, preserves the audit trail and the reporting context.
Start with one recording, not one contract
Transcription is the cheapest part of a reporting workflow and the easiest to get wrong, because the wrong choice hides inside a free tier until the week you cannot afford it. Take one awkward recording from your archive, run it through three tools, and judge them on names, speakers, speed and export rather than on marketing copy.
Then decide where the transcript should live. If the answer is anything other than “with the story”, that is the real problem to solve. The team at Publishrs.com builds publishing platforms for editorial teams who want their source material, their output and their audience data in one place, and it is worth a conversation before you sign another per-seat licence.
Source: Journalism.co.uk, 30 October 2026. Prices converted to GBP by the source and accurate as of 29 October 2026.
This article provides general information about publishing industry trends and best practices. For specific advice about implementing new systems or processes at your publication, we recommend consulting with your technical and editorial teams.








