An Explanation Tool, Not a Brand Film
Website marketing videos are recordings placed on your own pages to explain, demonstrate or evidence something a buyer must understand before they can proceed. That is narrower than the industry's usual definition, and the narrowness is deliberate, because three adjacent things get budgeted as if they were the same activity.
- The brand film — drone footage, staff laughing in a corridor, a strapline at the end. It is an internal morale product and answers no question a buyer has.
- Advertising creative for paid placement on YouTube, LinkedIn or connected TV. Different job, different length, different measurement; it belongs to the media plan.
- Social video, made to be watched without sound while scrolling — distribution content rather than site content.
Site video is judged on whether it makes something clearer than a paragraph could. Two minutes of a machine running, an interface used on real data, or an engineer explaining what goes wrong during installation is worth more to an evaluating committee than any production value spent on establishing shots.
What Video Is Genuinely Better At Than Text
Four things video does that prose cannot, worth naming so you stop commissioning video for the other reasons.
- Physical objects in three dimensions. Scale, finish, how two parts fit together. Photographs approximate this; video settles it.
- Software in motion. How many clicks a task takes, what the latency feels like, whether the interface is as simple as the screenshot implied.
- Sequence and process. Anything with an order — installation, changeover, a laboratory workflow, an onboarding path — is easier to follow shown in the order it happens.
- The credibility of a named human being. A person with a title saying a specific thing on camera is harder to fake than a page of claims, and it gives a committee a face to attach to a supplier they have never met.
Nothing on that list is served by a script performed to camera by a marketer. Put the practitioner in front of the lens and accept the imperfection; the imperfection reads as evidence.
Three Formats Worth the Money
Most useful B2B video falls into three types, each with a distinct job and a characteristic way of going wrong.
| Format | What it is for | How it fails |
|---|---|---|
| Product demonstration | Showing the thing working on real data or real hardware, at real speed | A scripted feature tour narrated over a sanitised demo environment nobody believes |
| Customer evidence | A named customer describing the problem, the decision, and what measurably changed | Approved adjectives with no specifics, cut to ninety seconds of praise |
| Recorded webinar or technical session | Depth for the evaluator who wants an hour of detail | Published as one long file with no chapters, no transcript and no editing |
The recorded session is the most undervalued of the three. It already exists, it was made for a real audience with real questions, and turning it into a durable page costs editing and a transcript rather than a crew. Chapter it, put the questions in the text, and it keeps earning long after the live date.
Where the File Lives, and What Each Choice Costs You
Hosting is a trade rather than a best practice; the right answer depends on whether you value reach, measurement or page performance most.
| Option | What you gain | What you give up |
|---|---|---|
| YouTube | Free delivery at scale, its own search and recommendation surface, a familiar player | Coarse analytics, recommendations that lead away from you, a heavy embed, and clicks that land on YouTube rather than your site |
| Video platform (Wistia, Vimeo, Vidyard, Brightcove) | Per-viewer engagement data, adaptive streaming, calls to action, optional gating, CRM integration | A recurring licence cost that scales with bandwidth or seats, and no discovery surface of its own |
| Self-hosted file | Total control, no third-party scripts, no data leaving your infrastructure | Bandwidth cost, no adaptive bitrate unless you build it, poor delivery on weak connections |
A defensible pattern is both: publish to YouTube for reach, embed from a video platform on site, so the page keeps its measurement while the file keeps its audience. Self-hosting is justified mainly where content cannot leave your control for legal reasons.
Why the Autoplaying Hero Video Hurts the Page
An autoplaying video behind the headline is a reliable way to damage real-world page performance, and the damage shows up in the metrics Google publishes. Largest Contentful Paint is good within 2.5 seconds; a heavy video or its poster frame often becomes the largest element and misses it. Cumulative Layout Shift should be 0.1 or less, and a player that loads then resizes after the text has rendered pushes content around. Interaction to Next Paint should be 200 milliseconds or less, and video decoding competes for the same main thread as the button your visitor is trying to press.
These metrics are assessed at the 75th percentile of real page loads, segmented separately for mobile and desktop, so a page can pass on desktop and fail on phones. And Interaction to Next Paint cannot be measured in a lab, so a perfect Lighthouse score is not evidence a page passes.
The fix is undramatic: a lightweight poster image, a play button, and the player loaded only when someone asks for it.
Captions and Transcripts Are Not Optional Extras
Captions are an accessibility requirement first. WCAG treats captions for prerecorded video as a Level A criterion and audio description as Level AA, which puts them inside most corporate accessibility commitments and the procurement questionnaires that ask about conformance.
They are also how a good deal of the viewing happens, in open-plan offices with the sound off. Auto-generated captions are a starting point and need correction, because the words a model transcribes worst are the ones that matter — product names, part numbers, industry acronyms.
A full transcript on the page does separate work. Search engines read text and do not watch video, so the transcript is the only genuinely indexable prose in the asset. It lets the reader who will not spend twelve minutes scan for the passage they need, makes the content findable through your own site search, and is raw material for written pages you would otherwise never have commissioned.
Video in Search Results: What Is Actually Verifiable
Google's structured data gallery still lists Video as a documented rich result type, and Video markup remains reasonable to add: it states the title, description, thumbnail, upload date and duration in a form Google can read without inference.
Google does not publish how a video result is selected for a given query, and nobody outside Google knows. What can be observed is the direction of travel, and it is not expansionary: Google removed its video carousel guidance in April 2024, and learning video markup was among the types deprecated in the June 2025 effort to simplify the results page, with the documentation removed that September.
One trade deserves stating plainly. A video hosted on YouTube can rank as a YouTube result, and that click goes to YouTube, where the next recommendation may be a competitor. A video embedded on your own page needs the page to rank. Decide which outcome you are optimising for before choosing where the file lives.
Measuring Engaged Views, and When Not to Make the Video
A view is not a metric. Every platform defines it differently and each definition is set to look generous. Define an engaged view yourself, in advance — a fixed share of the duration, or a fixed number of seconds — and report that instead.
Four numbers repay a monthly look: play rate, plays as a proportion of views of the page carrying the video; the distribution of watch-through rather than its average, since averages hide the pattern where most viewers leave in ten seconds and a minority watch everything; the exact point people leave, usually the moment the video stops explaining and starts selling; and whether accounts that watched behave differently. If you wire video events into GA4, note that a standard property allows 50 event-scoped custom dimensions and 30 key events — do not spend them on playback milestones.
Skip video when there is nothing to show, when no practitioner will appear on camera, or when the budget would otherwise buy the documentation buyers are asking for.
Frequently Asked Questions
How long should a B2B product video be?
As long as the explanation takes and no longer, which means different lengths for different jobs. A demonstration of one task is usually a couple of minutes. A customer account of a decision runs longer. A recorded technical session can justify an hour if it is chaptered so nobody has to watch all of it.
Be sceptical of prescriptive length rules; those figures come from individual platform datasets and are not comparable.
Does adding video help our search visibility?
Indirectly, and not in the way it is usually sold. The video file is not readable prose, so what helps the page rank is the text around it: transcript, headings, explanation. Video structured data tells Google the title, thumbnail, duration and upload date, but Google does not publish how video results are selected.
The defensible reason to add video is that it makes the page better for a buyer.
Do we need a professional production company?
For customer evidence filmed on a client site, usually yes: lighting, sound and making a nervous interviewee comfortable are real skills. For a screen recording of your own software, or an engineer explaining a failure mode at a bench, no.
The common error is inverting this: budget spent on polish for content with nothing to say, leaving nothing for the explanatory recordings buyers would have watched.
Are captions necessary if all our buyers speak English?
Yes, for three reasons unrelated to language. Accessibility conformance expects them, and WCAG treats captions for prerecorded video as a Level A criterion. A large share of workplace viewing happens with the sound muted. And captions plus a transcript give you indexable text where the video gives you none.
Auto-generated captions are acceptable only after correction: what a transcription model gets wrong is product names, part numbers and acronyms.