What problem does it solve?
Most public information is written, in the national language, and published as leaflets, web pages and PDFs. The municipality of Amsterdam says nearly all information from the city and its public health service (GGD) is written and in Dutch, so important information reaches people who do not speak Dutch or read with difficulty too little. Walsall Council, which serves more than 294,000 residents in a diverse borough, found inaccessible formats left behind residents with limited English, visual impairments or low literacy, and that traditional video was slow and costly.
The same gap appears in places that are not a city hall. Serco's HMP Dovegate found that written messages did not reach prisoners whose average reading age is around seven or who do not speak English, and that PDF messages got little attention.
How does it work?
- Start from approved text. Communications staff pick the leaflet, web page or notice that matters most for the audience, such as the vaccination information Amsterdam's GGD started with, and write it for speaking.
- Translate with AI. The text is translated into the languages of the local community. Walsall translates each video into 52 languages with a few clicks.
- Check the translation. In Amsterdam a communication adviser or translator checks the translated texts, and checks again the text the avatar speaks, so two human checks happen before publication.
- Generate the video. An avatar speaks the text. Walsall also built avatars of real local staff to make the content more relatable, with sign off and moderation steps built into the tool.
- Publish where the audience is. The videos go on the service's web page, as in Amsterdam, or on the intranet and in cell screens, as at Dovegate.
- Track what changes. Synthesia's case study says Walsall is measuring whether the videos reduce email and phone queries. Synthesia's Serco case study reports engagement for its violence reduction campaign.
- Audience
- Customer facing
- Autonomy
- Copilot
- Adoption
- Emerging
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Cycle time | Not pooled | 2 days | 1 | 1 vendor |
Value drivers: Inclusion and access, Speed and cycle time, Lower cost to serve, Customer experience.
Indicative value
A local authority that publishes 100 resident information videos a year
USD 24,000 to USD 280,000
Annual translation and production spend avoided per year
How this is calculated
Formula: videosPerYear * languagesPerVideo * costPerLanguageVersion * savedShare. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Videos published per year videosPerYear, videos per year | 100 | 100 | The reference authority. Replace with your own volume. |
| Languages added per video languagesPerVideo, languages | 4 | 8 | Editorial assumption, well below the 52 languages Walsall Council reports for each video. |
| Cost of one translated and voiced version today costPerLanguageVersion, USD per version | 150 | 500 | Editorial assumption, replace with your own translation and voice recording quotes. |
| Share of that cost avoided with AI translation and an avatar savedShare, fraction of cost | 0.4 | 0.7 | Editorial assumption. Synthesia's Walsall case study reports production time falling from at least a week to 2 days, a time reduction and not a cost reduction, so this range is not taken from it. The low value is more cautious because translation review stays. |
What it leaves out: Counts translation and voice production only. It leaves out license costs, review time by translators and communications staff, and any effect on how many residents understand the information or contact the authority afterward, which none of the three deployments on this page has published. The Walsall time figure comes from the vendor, Synthesia.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Gemeente Amsterdam
Netherlands · Government and public sector · 2024
The municipality of Amsterdam and its public health service (GGD) piloted AI avatar videos in the EU project CommuniCity to reach residents who do not speak Dutch or read with difficulty. Texts about vaccinations were translated with AI and spoken by a human like avatar in three languages for the GGD vaccination page. Two human checks applied: a communication adviser or translator reviewed the translated texts and the texts the avatar speaks. The register entry says early results were promising and the videos were watched a lot, without giving figures. The entry runs from May 2024 and is marked as out of use since February 2026.
No outcome disclosed.
Walsall Council
United Kingdom · Government and public sector · 2026
Walsall Council in the West Midlands delivers more than 400 services to about 294,000 residents, many with limited English, visual impairments or low literacy, and found that leaflets, PDFs and outsourced video were slow and costly. After a trial it made video its default channel, using Synthesia avatars, including custom avatars of real staff, built in moderation and sign off workflows, and translation of each video into 52 languages. Synthesia reports that production fell from at least a week to 2 days. The case study says Walsall is measuring whether the videos reduce email and phone queries, and gives no result for that.
- Cycle time: 2 days
"3.5x Faster Production: What once took at least a week can now be done in 2 days."
Claimed by: vendor
Serco
United Kingdom · Government and public sector · 2025
Serco runs HMP Dovegate, a Category B prison for up to 1,160 prisoners, where many have low literacy and some do not speak English, and where an Ofsted inspection found education needed improvement. The prison now sends weekly and monthly video updates made with Synthesia's AI video platform, covering news, incentive announcements, director messages and behavior campaigns, in seven languages and shown on in cell devices. Synthesia reports nearly nine times the engagement for a violence reduction campaign and that 55% of prisoners said video education was more effective than paper. The page offers these as results without a stated method, and the record maps them to no KPI.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Approved source texts written for speaking, with a named content owner
- A translator or native speaker reviewer for each language offered
- Consent from any staff member whose likeness is used for a custom avatar
- A list of the languages and formats the community needs, based on local data
Systems to integrate
- The authority's website or service portal that hosts the videos
- Content management and sign off workflow
- Where relevant, intranet or screen systems inside a facility
Complexity: Low
The video and translation tool is a ready made product, as Walsall's and Amsterdam's use of Synthesia and similar tools shows. What the authority supplies is a source text that has been checked by hand, a translator for each language, an accessibility review of the finished video, and a decision on where the videos are hosted and who may sign them off.
- 1
Pick information with a language or literacy gap
Start with a topic where residents are known to miss written information, as Amsterdam's GGD did with vaccination texts, and where wrong information would be easy to correct.
- 2
Set two human checks
Have a translator check the translated text and a communications adviser check what the avatar finally says. Amsterdam's register entry names two checks, each done by a communications adviser or a translator.
- 3
Decide the avatar policy
Choose between stock avatars and avatars of real staff, get written consent for the latter, and label the videos as AI generated.
- 4
Build accessibility in
Add captions and transcripts and use accessible templates, as Walsall describes with WCAG compliant templates, and test with residents who read little.
- 5
Measure outcomes for residents
Track views, queries and calls on the topic before and after, per language, so the program can show whether people understood. Amsterdam reports only that the videos were watched a lot.
Guardrails
- Every translated text and every spoken script checked by a person before publication
- Avatars labeled as AI generated and no real person's likeness without written consent
- No personal data about residents used to generate the videos
- Moderation and sign off workflow so one team cannot publish for another unseen
- Captions and transcripts published with each video
KPIs to instrument
- Days from approved text to published video, per language
- Views and completion per language version
- Emails and calls on the topic before and after publication
- Translator corrections per video
Human in the loop
A communications adviser signs off each script and a translator signs off each language, and the service owner approves any video that tells residents what they must or must not do.
Common failure modes
- Wrong or stilted translation in a language nobody on staff reads
- A mistranslated instruction reaches residents who trust the video. Use a named translator per language and sample after publication.
- Videos that residents cannot use
- An avatar speaks quickly or the video lacks captions, so people who read little still miss the point. Test with the audience and add captions and transcripts.
- A pilot that ends without a stated reason
- Amsterdam's register entry shows its GGD avatar pilot ended in February 2026 without a reason. Record the decision criteria and results at the start, so an ending teaches something.
- Trust damaged by hidden AI
- Residents learn that a friendly local face was synthetic. Label the videos and say when an avatar is modeled on a real staff member.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
Producing information videos is not listed in Annex III. Providers of tools that generate synthetic video must mark the output as AI generated, and deployers must disclose deep fakes, such as an avatar modeled on a real staff member (Article 50(2) and (4)). A system that also decided who gets a benefit or service would need a separate assessment under Annex III point 5.
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Marking of synthetic video and audio by providers, and disclosure of deep fakes by deployers.
- Code of Practice on Transparency of AI generated Content (European Commission, Europe). Supports compliance with the marking and labeling duties for AI generated content.
Controls to put in place
- Public register entry or equivalent transparency record for the avatar program
- Translator sign off log per language
- Register of avatars and consent records for any real person's likeness
- Accessibility check before each publication
Frequently asked questions
- Who has deployed AI avatar videos for residents?
Public deployments are few. Walsall Council in the UK translates each video into 52 languages with Synthesia, Serco's HMP Dovegate sends Synthesia AI video updates in seven languages to prisoners, and the municipality of Amsterdam piloted GGD vaccination videos in three languages from May 2024. Amsterdam lists the pilot as out of use since February 2026.
- Does it save money for a public body?
Synthesia's case study reports that a Walsall video that took at least a week now takes 2 days, which is a vendor figure about time, not cost. The case study also says Walsall is measuring whether videos reduce email and phone queries and gives no result. Amsterdam's register entry says only that early results were promising.
- Do residents have to be told the presenter is an AI avatar?
Under the EU AI Act, deployers must disclose deep fakes, such as an avatar modeled on a real staff member, and providers must mark synthetic video as AI generated. Labeling every avatar video is the simple policy, and Amsterdam's register entry shows how a public body can publish the checks it uses.
How to cite this page
Blits.ai AI Use Case Library, "AI avatar videos for public information in many languages", last verified 14 October 2026, https://www.blits.ai/ai-use-cases/public-information-avatar-video. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 14 October 2026: First published