# Avoiding AI Pitfalls In Council Communications: Hallucinations, Bias And Tone

URL: https://www.starien.com/blog/avoiding-ai-pitfalls-in-council-communications
Published: 2026-10-26
Author: Geoffrey Moffett
Topics: AI risks, hallucinations, council communications, councils, responsible AI

> The AI pitfalls that catch council communications teams out, from invented facts and quotes to bias, bland tone and fake images, with a practical checking routine for each.

## Quick answer

The main AI pitfalls in council communications are hallucinations (invented facts, figures and quotes), bias in language and assumptions, tone that's bland or wrong for the moment, outdated information, generic content and AI images mistaken for real photos. A short checking routine, clear prompts and named human sign-off prevent most problems.

## Key takeaways

- AI can invent facts and quotes confidently; supply facts and check everything.
- Bias can creep into wording and assumptions, so review with fresh eyes.
- Tone is a judgement call AI often gets wrong in sensitive moments.
- A consistent checklist catches most problems before publication.

AI can make a council comms team faster. It can also embarrass them in public. Most AI pitfalls in council communications are predictable, which means they're preventable. Here are the main ones, with what to do about each.

## Why AI Pitfalls In Council Communications Hit Harder

A business that publishes a wrong fact corrects it and moves on. A council that publishes a wrong date for a consultation, an invented quote from a councillor or an AI image passed off as a real local scene can face complaints, press coverage and lost trust. Public bodies are held to a higher standard, rightly.

## Pitfall 1: Hallucinated Facts

AI tools can produce confident, plausible, false information: a statistic that doesn't exist, a wrong opening time, a policy that was never adopted.

**Prevent it:**

- give AI the facts in the prompt;
- tell it: "Don't add any facts, figures or claims that aren't provided";
- ask it to mark gaps with [CHECK];
- check every number, date, name and link against the source.

## Pitfall 2: Invented Quotes

Ask AI for a press release and it'll often add a quote. It sounds great. Nobody said it.

**Prevent it:** quotes come from the named person or are approved by them. Never publish an AI-suggested quote without that approval.

## Pitfall 3: Fake Sources

AI may cite reports, guidance or legislation that doesn't exist, or misdescribe real ones.

**Prevent it:** never publish a reference you haven't opened and read yourself.

## Pitfall 4: Outdated Information

AI tools may rely on older information, describing a service, structure or rule that's changed.

**Prevent it:** use your own current sources as the basis for content, not the AI's general knowledge.

## Pitfall 5: Bias

AI can reflect biases in what it's learned: assumptions about who uses services, stereotyped descriptions, imbalanced examples, or language that's less respectful to some groups.

**Prevent it:**

- review content for assumptions about people;
- ask a colleague with a different perspective to read sensitive content;
- consider your Section 75 equality duties when content relates to particular groups;
- be especially careful with images and descriptions of people.

## Pitfall 6: Wrong Tone

AI defaults to cheerful and upbeat. That's wrong for a bereavement, a serious incident, a contentious decision or community tension. In Northern Ireland, topics touching identity, culture, language or the past need particular care, and AI doesn't reliably understand those nuances.

**Prevent it:** for sensitive topics, write with senior colleagues first and use AI only for light editing, if at all. Pause scheduled cheerful content during serious incidents.

## Pitfall 7: Bland, Generic Content

"We are delighted to announce this exciting initiative, which demonstrates our commitment to…" AI can make every council sound the same.

**Prevent it:**

- include local names, places and specifics in your brief;
- ban phrases in your prompt ("Don't use 'exciting', 'delighted' or 'commitment to'");
- edit for a human voice.

## Pitfall 8: AI Images Presented As Real

An AI-generated image of a "new park" or "local event" that looks like a photo can mislead residents and undermine trust in everything else you publish.

**Prevent it:** use real photos of real places and people. If you use AI images for illustrative graphics, make sure they can't be mistaken for photographs and label where appropriate.

## Pitfall 9: Data Leaks

Pasting complaints, casework or confidential papers into unapproved tools.

**Prevent it:** follow your AI policy, remove personal data and only use approved tools. See [UK GDPR and AI for councils](/blog/uk-gdpr-and-ai-for-councils-practical-considerations/).

## Pitfall 10: Over-Volume

Because AI makes content easy, teams post more. Residents tune out.

**Prevent it:** use AI to make content better, not just more.

## A Pre-Publication Checklist

- [ ] Every fact, figure, date, name and link checked against a source
- [ ] Quotes approved by the person quoted
- [ ] No references you haven't verified
- [ ] Tone right for the moment and audience
- [ ] Checked for assumptions or biased language
- [ ] No AI images that could pass as real photos
- [ ] Local specifics, not generic phrases
- [ ] No [CHECK] markers left
- [ ] Named person signs off

## A Made-Up Example: The Plausible Statistic

Imagine an officer asks AI to "write a short article about why recycling matters in Northern Ireland". The draft includes a precise-sounding statistic about recycling rates, attributed vaguely to "recent figures". It looks authoritative. It may be wrong, out of date, or invented entirely.

The fix is simple: either remove the figure, or find the official source, check it and cite it properly. Better still, rewrite the prompt: "Write a short article about why recycling matters, using only the facts below. Don't include any statistics unless I've provided them." Then supply the verified figures yourself.

This kind of slip is common, and it's exactly why "check every number" sits at the top of every checklist.

## Building A Culture That Catches Mistakes

Checklists only work if people feel able to use them. Encourage a team culture where:

- questioning a draft is normal, whoever wrote it;
- near-misses are shared openly, without blame, so everyone learns;
- new starters are shown real examples of AI errors early on;
- time for checking is built into deadlines, not squeezed out by them.

## When Something Goes Wrong

Correct quickly and openly. Update the content, note the correction where appropriate, and log what happened so your process improves. Don't blame "the AI"; the council published it.

## How Starien Can Help Train Council Teams

Our [AI Coaching](/ai-coaching/) shows teams these pitfalls with live examples, then builds the prompts and checklists that prevent them. Our [Public Relations](/public-relations/) background means we take reputation seriously. We've worked with public sector teams and spent six years leading Coleraine BID's digital delivery. [Get in touch](/contact/) to book a session.

## Frequently asked questions

### What is an AI hallucination?

When an AI tool produces information that sounds plausible but is false, such as an invented statistic, date, quote, policy or source.

### How can councils reduce AI hallucinations?

Give AI the facts in the prompt, tell it not to add any, ask it to mark gaps, and check every fact, figure, name and link against the source before publishing.

### Can AI content be biased?

Yes. AI can reflect biases in its training data, for example in assumptions about who uses a service or in descriptions of people. Human review and diverse perspectives help catch this.
