Services, Done Faster With AI

Dubbing a minute of video costs $20 to $50 at an agency, $5 to $15 from a freelancer, and as little as $0.41 on an AI platform.

Sit with that spread for a second, because it contains both the opportunity and the trap and most people only see one of them.

The opportunity: a person with a $20 subscription can now deliver at freelancer quality for platform cost, and keep the difference. The trap: so can everyone else, and the freelancer rate is the thing being compressed. Whether you end up on the good side of that spread depends entirely on what you sell, the output or the judgement.

This page is about the least glamorous and most reliable AI earning route: taking a service people already pay for and doing it faster.

Why this beats most of the section#

Every other route here asks you to acquire something new, a credential the labs want, a platform to learn, an app to build. This one asks you to keep doing what you already do and pocket the time.

The demand already exists and is already budgeted. You are not persuading anyone that AI is the future; you are quoting for a job they were going to commission anyway. And unlike the routes that only work in the US, this works wherever your clients are.

The video localisation market alone is put at around $4 billion in 2026. Transcription, subtitling, translation, editing, resizing, summarising, formatting, every one of those is a real line item in somebody’s budget, and every one of them got dramatically cheaper to produce this year.

Where the margin sits#

Localisation is the clearest example because the numbers are public.

Rates vary enormously by language pair: English to Hindi runs $10–25 a minute, English to Japanese $30–50, rarer languages $50 and up.

The same shape applies across the board. Transcription that took an hour takes four minutes. A rough subtitle pass that was a paid job is now a button. Cutting a long video into vertical clips, covered in detail under selling a skill — went from six hours to two.

Pricing guides and platform pages, not measured data.

The distinction that decides whether you earn or get undercut#

Here is the part that matters more than any tool.

If what you sell is the output — “I will dub your video into Spanish”, then your competition is a $0.41-a-minute platform your client can also use, and your price will be driven towards it. You are selling a thing that is no longer scarce.

If what you sell is the judgement — “I will dub this into Spanish and catch the three places where the machine translation makes you sound like an idiot to a Mexican audience”, you are selling something the platform cannot do, and the platform’s low price becomes your cost base rather than your ceiling.

Concretely, the second version is: knowing which of the AI’s four voice options sounds trustworthy rather than uncanny for this audience. Catching the idiom that translated literally. Noticing the on-screen text nobody localised. Knowing that a technical term has an accepted translation in the industry and the model picked the dictionary one. Fixing the timing where the dub runs over the visual.

None of that is glamorous and all of it is why the client comes back. It is also the reason to specialise in something you know, a language you actually speak, an industry whose vocabulary you have, an audience you belong to.

What to sell#

The services where AI collapsed the cost and human judgement still decides the quality:

• Localisation and dubbing. Highest visible margin, and the language-pair pricing means a rarer language you actually speak is a genuine advantage.
• Subtitling and captioning. Machine output is 90% right and the 10% is names, jargon and timing. Accessibility compliance work is real, ongoing, and often has a budget attached to it by policy rather than by choice.
• Transcription of hard audio. Accents, crosstalk, poor recordings, technical vocabulary, exactly where the automatic version still fails and someone still needs it right.
• Repurposing. One long video becomes clips, a newsletter, a thread, an article. The transformation is cheap now; deciding which forty seconds matter is not.
• Document and report work. Turning messy notes into a usable document, summarising long material accurately, formatting to a house style.
• Research and briefing. Faster to produce, and worth money only if someone verifies it, which is the whole job.

Pricing it, since this is where people give the gain away#

One rule and it is the same rule as everywhere else on this site: never price by the hour once AI has made you fast.

If a job took you four hours and now takes forty minutes, hourly billing hands the entire benefit to the client and cuts your income by 85% for getting better at your job. Price the deliverable. Charge $180 for the localised video, not $30 an hour for the six times you touched it.

Two other things to hold:

Do not advertise the tool. Nobody buys “AI-assisted subtitling”, because that tells them they could do it themselves. They buy “subtitles that are actually correct, in two days”.

Do say so if asked and say what you did. “First pass by machine, then I check every name, every technical term and all the timing” is a better answer than either concealment or a shrug. It describes the value, which is a rare thing in a sales conversation.

Whether you must disclose depends on what you produce and where the client is, the EU’s transparency obligations have been in force since 2 August 2026 and reach further than people expect. Disclosing AI use has the actual rules, and the short version is that labelling costs you nothing and being caught costs everything.

What it costs to start#

One subscription, roughly $20 a month, plus whatever specialist tool your particular service needs. Many have usable free tiers, which is enough to do your first paid job and fund the paid tier from it.

The other startup cost is a small amount of your own work as a sample, which is the same as in selling a skill: take a public video, localise or subtitle or repurpose it properly, and show the before and after with a sentence on each decision you made. The sentences are the portfolio. Anybody can show output; showing judgement is the differentiator.

Who buys#

Creators with long-form content and no time, the same client base described in selling a skill and the easiest to reach because their work and their contact details are both public.

Small businesses with a video or a document problem and no in-house capacity.

Agencies and other freelancers with overflow, which is the quietest and most reliable source of continuing work in this whole field.

And for localisation specifically: anyone whose audience is growing in a market whose language they do not speak. That is a large and expanding group and they usually do not know that doing something about it is now affordable.

The risk: rates are being compressed#

Rates in this category are being compressed. The same tools that give you the margin are available to your client, and the trajectory is the one already visible in UGC rates: the undifferentiated end gets competed away and the specific end holds.

Two implications. Do not build a business on being cheap, because there is always something cheaper and it is a platform. And put the money into something that compounds — a product, an audience, a niche reputation, while the margin is good.

Getting started this week#

1. Pick the one service where you have genuine judgement, a language you speak properly, an industry whose vocabulary you know, a format you understand.
2. Learn the AI tool for it well enough to know where it fails. Its failure modes are your product.
3. Do two samples on public material. Show before and after, with the decisions written out.
4. Price per deliverable, never per hour.
5. Approach fifteen people who visibly have the problem, a creator with long videos and no subtitles, a business with an English-only site and a foreign audience.
6. Deliver early, then ask what else on their list is slow. That question is where the second job comes from.

The rate calculator will price a specific job including rush and revisions.