ENGR SOLUTIONS
On the desk and off it
When library staff should, and should not, use AI
Earlier this year a large public library system in British Columbia asked us for something specific. Not another AI policy; they had the principles. They wanted the practical line their staff could carry in their heads: when does AI belong in the work of serving the public, and when does it not?
Our answer starts with where you are standing. The same tool that saves an hour behind the scenes can do real harm across the desk. So the guideline splits in two: on the desk, and off it.
Why the desk is different
At a service desk there is a person waiting. The question is real, the answer will be acted on, and there are about ninety seconds to give it.
Generative AI tools write fluent, confident answers that are sometimes wrong. The Government of Canada’s own guide on generative AI lists inaccurate content first among the risks. Off the desk you have time to check every claim. At the desk you do not, and an unchecked answer from a library carries the library’s credibility with it.
That credibility is the whole point. Your library pays for databases, catalogues and licensed resources so that the answer a patron gets is one you can stand behind. AI is not a substitute for them.
On the desk: three rules
1. Don’t use AI to answer a reference question. Use the catalogue, the licensed databases and the official websites, as you always have. If you would not have time to verify what the tool says, and at a busy desk you will not, it does not belong in the answer.
2. Help the patron who brings AI with them. More and more people arrive having already asked a chatbot. Treat it like any other source they bring in. Look at what it said together, check it against something credible, and show them where you looked. “The AI told me this, is it true?” is an information literacy lesson waiting to happen.
3. On high-stakes questions, go straight to the official source. Immigration, legal, health and benefits questions are where a confident wrong answer does the most damage. Point newcomers to the Government of Canada’s immigration and citizenship pages rather than a chatbot’s summary of them. A translation tool can help someone understand a letter. It should not write what they submit. Nothing personal about a patron goes into a public AI tool, ever; the federal guide says the same about personal information and sensitive data.
Off the desk: where AI earns its place
Behind the scenes is where these tools help most, because there is time to check the work. Staff use them to:
- turn a programme idea into a session plan;
- rewrite a notice in plain language, or at a shorter length;
- summarise a long report before a meeting, then read the parts it flags;
- draft the FAQ for a new service before patrons start asking;
- produce a first-pass translation that a fluent colleague then checks.
Use the tool your organisation has approved. For many public libraries that is a tool licensed through the municipality, such as Microsoft Copilot. It is the one where your data protections have been reviewed, so it is the one for hands-on work. Staff should still know the others by name and roughly what they do, because patrons will ask about them. Awareness is not the same as use.
A one-page guideline you can adapt
Here is the version we would put on the staff room wall.
On the desk
- Reference answers come from the catalogue, the databases and official websites. Not from AI.
- When a patron brings an AI answer, check it with them against a credible source.
- Immigration, legal, health and benefits: send people to the official source.
- No patron’s personal information goes into an AI tool.
Off the desk
- Use the organisation’s approved tool for drafting, summarising, planning and translation drafts.
- You are the author. Check every name, date, number and title before anything leaves your hands.
- Say so when it matters. If a public notice was translated with a tool, note that a human check is welcome.
- Learn the other tools well enough to answer questions about them.
Putting it in place
Getting there takes a few weeks, not a policy rewrite.
- Name the approved tool. If staff do not know which tool they are allowed to use, they will use whichever one is open in their browser.
- Adapt the one page. Change the examples to your library’s own databases, services and community. Keep it to one page.
- Walk through three real desk moments at a staff meeting. The chatbot answer, the immigration letter, the busy Saturday reference question. Ask what each person would do.
- Review it in six months. The tools will change. The line between the desk and the back office probably will not.
In the Mount AI model we use with library leaders, this is the Train stage: a shared language and clear guardrails, in place before a pilot ever touches patron services. Canada’s new AI strategy, AI for All, names public libraries as partners in bringing AI literacy to every community. Libraries can only take that role once their own staff are confident first.
Partner with ENGR Solutions
We are more than consultants, we are your AI transformation partners.
- Strategy and policy guidance
- AI literacy and workforce training
- Pilot design and technical implementation
- Change management and digital leadership
Whether your staff need the line between the desk and the back office drawn for them, or your leadership team is planning its first pilot, we will help you climb Mount AI while staying true to your mission and the communities you serve.
Let’s put AI to work. Visit engrsolutions.ca or write to info@engr.solutions.
Sources and further reading
- Government of Canada, Guide on the use of generative artificial intelligence
- Government of Canada, Immigration and citizenship
- Innovation, Science and Economic Development Canada, Canada’s National Artificial Intelligence Strategy: AI for All
- ENGR Solutions, AI for All: Canada’s National AI Strategy, and our place in it
- ENGR Solutions, Mount AI
Photos: Declan Sun and Ilya Semenov on Unsplash.