What AI is really changing in customer education and enablement

For two years, customer education and enablement teams have heard the same promise: AI would handle production, and a team of two could finally ship like a team of ten. A good part of it is true today. Polished scripts come out of rough narration, voiceovers don't need a microphone, and in our study of 45,000 new videos the median video built by an AI agent reached export about 15 minutes after someone started it, proving AI-made videos are fitting the bill in most places.

Making videos faster, however, didn't leave teams with less to teach. A sales onboarding lead told us on a call about a program she planned with a colleague in another region. The two of them spent a week on eight-hour calls mapping it out, then asked for one new hire to build it. The hire wasn't approved, so they're now building it themselves with AI video tools.

AI "definitely speeds up the process, but doesn't completely fix the issue," she said. The tools can produce the videos, but what the program needs to teach is still in those two people's heads. Getting it into the videos means writing a prompt, watching the result, changing the prompt, and watching again.

I think that's an accurate account of what AI has done to this work. It cut the cost of producing a training video to minutes, while the cost of the expertise the video has to carry, and of checking that the AI got it right, is about where it was. Those were always the expensive parts of education, and now they're nearly all of it. The teams getting real value from AI have noticed, and they put their effort into getting expert knowledge into the tool instead of expecting the tool to supply it.

AI made training videos cheap to produce, not the knowledge inside them

Think about what a training video used to cost a small team. Someone recorded the workflow, wrote a script, recorded a voiceover, and edited it all together, and on most teams that was the same overloaded person on different days.

AI has absorbed most of that list. AI voiceover reads a script in whatever voice you pick, a rough narration gets cleaned up into the script, and the written step-by-step guide comes out of the same recording. Agents went from building almost no Clueso videos in May to 19% of new ones in October.

None of that touches the scarcer parts of the job, which is adding a context layer in the videos you produce. When it's routine tasks such as the settings a new admin actually needs on day one or the mistake nearly every customer makes in their second week, AI-made videos do the job really well.

But adding enough context for a differentiated audience is where you come in. An automated video might be enough for training, but "how the messaging in it comes to life for an SDR might be different than an AE versus a CSM," so managers end up adding that context by hand.

AI can draft each of those versions in minutes once someone has decided what each audience needs to hear. It can't make that decision, because the information it would need isn't written down anywhere it can read. In most companies it sits with a product manager or with the CSM who has answered the same question all quarter. Keeping those videos accurate after the next release is a separate cost, which we covered in why maintaining product videos now costs more than making them.

Better prompting leads to higher quality videos and more exports

We wanted to know whether the effort people put into directing the AI shows up in what they finish. So we pulled every video our customers started by typing a request into Clueso's AI chat between June 1 and October 19, 2026. That came to about 16,000 videos with no file attached, and we checked how many of them were ever exported.

People who wrote 100 words or more exported 12.3% of their videos, about 2.7 times the rate for prompts under 15 words. Our initial assumption was that this is mostly intent, because someone who writes a paragraph is further along than someone who types eight words to see what happens (we do plenty of that ourselves), and we still feel some of it is. A marketer cutting a webinar into a short teaser described the rest of it on a call: "The problem is I didn't really know what I wanted. I was hoping it would just generate something that I like."

The bigger gap showed up when people gave the AI something other than words. When a video request came with an uploaded recording, 17.3% of those videos were exported, against 4.3% for video requests that came with a prompt alone. A recording carries what the person knows about the task, including the real screens and the real order of clicks. Most of that never makes it into a typed request.

An expert's recording is now the most valuable input you have

If expertise is the scarce input, the hard part is getting it out of an expert's head. Asking a product manager to write up a workflow takes an afternoon and still misses the steps they do without thinking. Recording them doing it takes about as long as the task itself, and the steps they'd forget to mention are in the footage anyway.

At Clueso, the product is built around that recording. The expert records the workflow and narrates roughly, without restarting when they fumble a sentence. Clueso cleans the narration into a script, can replace their voice with an AI voiceover if they'd rather not be heard, adds zooms where the action is, and generates the written guide from the same recording. The expert's time goes into the one part only they could supply, and the education team gets a video and an article out of it.

The same holds when the AI doing the work is an assistant your team already uses. With the Clueso MCP, available on every plan, a team working in Claude or ChatGPT can build videos without leaving the chat. What we saw in the chat data applies there too: hand the agent the recording or the document, not a one-line description of it.

Reviewing what the AI made is now the slowest step

That loop of prompting, reviewing, changing, and reviewing again is where most of the time goes now. Production that used to take a week takes minutes, but every version still needs someone who knows the product to watch it end to end and catch the step that's in the wrong order. When each fix means regenerating the whole video, each fix also means watching the whole video again.

The calls we sit in suggest two ways teams shorten that loop. The first is settling style decisions once, in a template, so a reviewer checks whether the content is right rather than whether the intro and the fonts match last month's video. The second is editing the part that's wrong instead of regenerating everything. A team evaluating Clueso at a research platform asked us exactly that: whether they could rework one section so "you don't have to re-render the whole video and re-prompt."

In Clueso, a Clueprint holds your layout, branding, and editing style, so every video built from it starts consistent. Community Clueprints are on every plan, and custom Clueprints are on Enterprise. Clueso Agents edit an existing video from an instruction, changing one line of script along with its voiceover and callout without regenerating the rest (Enterprise).

The headcount case for education teams got harder to make, and it shouldn't have

Education teams were short of people before any of this. In Thought Industries' 2024 State of Customer Education report, a survey of more than 200 customer education leaders, 57% named insufficient headcount as a barrier and 53% said their team was overloaded. Then AI arrived and made output look abundant, which is a hard backdrop for asking for another hire. A team that suddenly ships far more videos is easy to point to in a budget conversation that ends with the suggestion that AI can cover the rest.

The sales onboarding lead whose request was turned down wasn't short of production capacity, though. They were short of hours from the people who know what the program should teach, and of someone with time to review what the tools produce. A request built around those two things is a different ask from "a person to make videos," and AI tools don't reduce either of them.

A customer at a fast-growing healthcare software company described the same pressure from the other side: "with how much growth our product is having, our people growth has not been able to keep pace." When the product grows faster than the team, the hours that decide whether training is any good are the ones experts spend on camera and reviewers spend checking.

How we measured this

The prompt-length and upload figures come from Clueso's product analytics. We included every video a customer started from a request in Clueso's AI chat between June 1 and October 19, 2026, and counted only the first request for each video. A video counts as exported if it finished at least one video export by November 2, 2026.

Prompt length is the number of words in that first request. Clueso employees, our internal workspaces, and test environments are excluded. These are aggregates only, and no prompt text was read. The 45,000-video figures come from our earlier study, which explains its own method.

Frequently asked questions

Will AI replace instructional designers and customer educators?

Not on the evidence we see. AI has taken over most of the production work, including the script, the voiceover, the edit, and the written guide. Deciding what a course should teach and whether a video is accurate still takes someone who knows the learner and the product. In Clueso, the screen recorder turns a rough recording into a scripted video with an AI voiceover and a matching help article, which leaves the educator's time for the design and the review.

How is AI changing customer education?

It has made producing content fast and cheap, which moves the bottleneck to expertise and review. Teams now spend more of their time getting knowledge out of product experts and checking AI output than building videos. Clueso's AI chat and the Clueso MCP build videos from a prompt or from files you upload, on every plan. In our data, the videos started from an uploaded recording were exported about four times as often as prompt-only ones. For specific tasks to hand to AI first, see seven AI moves for customer education teams.

What's the best input for an AI-generated training video?

A recording of an expert doing the real task, narrated roughly, gives the AI the screens and the order of steps that a written prompt leaves out. If you only have a prompt, a longer and more specific one does better in our data than a short one. In Clueso, the recording becomes the script, the AI voiceover, and a step-by-step written guide, and a Clueprint keeps the style consistent across every video built from it.

Check what your last ten AI videos started from

Pull the last ten videos your team made with AI's help and note what each one started from: a one-line prompt, a document, or a recording of someone who knows the task. Then mark which of them actually shipped. If your numbers look anything like ours, the recordings will be overrepresented in the shipped list, and that's a fair guide to where next quarter's expert hours should go.