A teleprompter for public sector announcements — approved wording, read exactly

A council announcement is not content; it is a record. Bin-collection changes over the holidays, a boil-water notice, a road closure, the election-registration deadline. If the officer on camera says “in the morning” where the approved text says “from 7am”, the phone lines pay for it the next day — and on a statutory notice, an inaccurate paraphrase is a legal problem, not an awkward one. The wording was approved for a reason, and the video has to carry it exactly. The approved text also has an audit trail — who signed it off, when, in response to what — and a video that drifts from it quietly breaks that chain.

Accessibility is not a preference either. Public-sector accessibility rules make captions an obligation, and the web team’s workflow wants a caption file it can attach, check and correct in the CMS — words burned into the picture are the wrong shape for compliance, because nobody can fix a misspelled street name in a flattened frame.

Then there are languages. The same notice needs to exist in the languages residents actually speak, each version read by a different colleague, and the versions must say the same thing on the same day. Community languages include right-to-left scripts, which most tools treat as an afterthought that breaks the layout.

And the comms team is two people and a phone. A studio booking for a bin-day change is absurd, an agency invoice for a road closure is unjustifiable — so video only happens for emergencies, which is precisely when there is no time to learn a tool. The realistic production unit is the phone already in the drawer, operated by whoever last recorded a video at a family birthday — and the tool has to fit that person, not a video department that does not exist.

The default is a written statement on the website and a graphic for social media. It is accurate and nobody watches it: no face, no voice, and residents scroll past the third notice-shaped image that week. And reach here is not a marketing metric: the resident who missed the boil-water notice is not a lost conversion but a phone call, a complaint, and in the worst case a safety incident.

When video does happen, an officer memorises the statement and paraphrases it — which review catches, and the whole thing is recorded again — or reads from a printed page beside the phone, with the tell-tale eye-line drop on every sentence.

Captions get left to the platform’s auto-captioning, which guesses at street names and gets them wrong in public. The other-language versions quietly stop happening, because the process was too heavy for a two-person team the first time, and nobody restarts it. Which is a loss the metrics never show: a notice read to residents in their own language is the one they remember — and act on.

The script library holds the approved text, pasted in as written. Folders keep each campaign or notice type together, favourites pin the recurring ones — the bin-day change comes back every December — and every script shows a word count and an estimated read time, so you know the announcement fits sixty seconds before anyone presses record. When a councillor wants one word changed, the change happens in the script before the take, not in a paraphrase during it — and the script is saved beside the recording, so what was actually read stays on record.

Voice follow scrolls the script as the officer reads, so the wording on screen is the wording that was signed off, spoken exactly, with pauses that stay natural. Each language version lives as its own script in the same folder, and the prompter follows the reader in each language the same way — record the English take, then the Polish, then the Urdu, on the same phone in the same afternoon. The read-time estimate earns its keep twice here: it warns you which language version runs long before a colleague is already standing in front of the lens.

The prompter’s typography carries the part most tools drop: text size, line spacing and margins adjust to whoever is reading, five faces include Lexend and OpenDyslexic, and right-to-left scripts lay out correctly without a setting. The Urdu version is not a special case that breaks the tool — it is another script in the folder.

Captions export as a separate .srt file per video — the file the web team attaches in the CMS, checks against the approved text, and can correct if a street name needs it. Each language version gets its own, so the accessible version and the translated version stop being separate projects.

A water main bursts, and the supply-interruption notice is signed off at 15:00. The comms officer pastes it into the script library; the read-time estimate says fifty seconds — it fits the format without cutting a sentence legal wanted kept.

The English take is recorded in one pass, the prompter following the officer’s voice. A colleague records the Polish and Urdu versions from the same folder — the Urdu script laying out right-to-left without anyone touching a setting — and each take gets its .srt exported alongside.

The web team attaches the caption files in the CMS, and all three versions are live before 17:00: the same signed-off wording in every language, captioned for compliance, with no studio, no contractor, and no paraphrase for tomorrow’s phone lines to pay for. The three scripts stay in their folder, favourites pinned, against the next burst main — because there will be one.