How to write job postings and screen applicants with AI, fairly

You posted one opening for a shop assistant on Monday. By Thursday there are 140 applications, half of them clearly written by a chatbot, and you have a business to run. The temptation is obvious: let a different chatbot write the next ad, and let it pick the five people worth calling.

The first half of that is a sensible use of AI. The second half is where small employers get into trouble, with the law and with the people they never interviewed. This guide walks through doing both carefully.

Who this is for and what you need

This is for owners and managers who hire a few people a year and do it themselves, without an HR department. You need a chat assistant (ChatGPT, Gemini, Copilot or similar) on a business account, your existing job description or notes on the role, and about an hour the first time.

The legal details below use Ontario's rules, because they are the most specific in Canada on AI in hiring. If you hire in another province, the fairness steps still apply; check your own province's employment standards for the posting rules.

Decide what the AI is allowed to do

Before you open any tool, split hiring into two jobs. Writing is drafting the ad, the interview questions and the rejection email. Judging is ranking, scoring, filtering or rejecting people.

AI is good at writing and needs little oversight there, because you read every word before it goes out. Judging is different. The tool's choices are invisible unless you check them, and each one affects a real applicant.

The law draws a similar line. Since January 1, 2026, Ontario employers with 25 or more employees must state in a publicly advertised job posting whether they use artificial intelligence to "screen, assess or select applicants", according to the province's Employment Standards Act guide. A simple statement is enough; you do not have to describe the tool.

The usual mistake is assuming you don't use AI for judging because you never set it up. Many applicant tracking systems (the software that collects applications, including the one built into your job board) now rank or filter candidates by default.

Draft the posting, then fix what it gets wrong

Give the assistant the facts, not a blank page: the duties, the hours, the location, the pay range, and what someone must be able to do on day one. Ask it for a plain-language posting of about 300 words. Then edit.

Three things to check in every draft. First, the requirements. AI tends to pad a list with "nice to have" items dressed up as musts, and every unnecessary requirement shrinks your pool. Second, the language. Cut phrases that describe a type of person rather than a skill, such as "young, energetic team" or "native English speaker". Third, the legal lines. Under Ontario's rules, a covered posting must:

• include the expected pay or a range, and the range can be no wider than $50,000 a year (unless the pay is over $200,000);
• not require Canadian work experience;
• say whether the job is an existing vacancy;
• include the AI statement, if AI screens, assesses or selects applicants.

A chat assistant will not add those by itself, and it may add the Canadian-experience line you are no longer allowed to use. Those details come from the Ontario guide, which also covers the exceptions.

Write down what you are hiring for before you screen

Before any application is read, by you or by software, write five to eight criteria that actually predict success in the role. "Can lift 20 kg", "has handled cash", "available Saturdays". Each should be something you can see in an application or ask about in an interview.

This is the step that makes AI screening checkable. Without written criteria, a tool ranks people by how closely they resemble something, and you cannot tell what that something is.

That is not a hypothetical risk. In 2018, Reuters reported that Amazon had scrapped an experimental tool that scored applicants from one to five stars. It had learned from ten years of résumés, mostly from men, and it marked down résumés containing the word "women's" and graduates of two all-women's colleges. Amazon had engineers to find that. A five-person business will not, unless it looks.

If you cannot say what the tool is ranking people on, you cannot say it is fair.

Screen with AI, but never reject with it

If you do use AI on applications, use it to summarize and sort, not to decide. Paste in your written criteria and ask it to say, for each application, which criteria are met and where in the text it found them. That turns a black box into something you can check in thirty seconds.

Then read every application the AI would put in the "no" pile, at least for the first few hires. You are looking for patterns: people with career gaps, foreign credentials, non-English names or unusual formats being marked down for reasons that are not on your list.

This matters legally as well as ethically. The Human Rights AI Impact Assessment from the Ontario Human Rights Commission and the Law Commission of Ontario says plainly that human rights laws apply to AI systems, and names hiring and job screening among the areas at risk. It adds that an organization "is not shielded from human rights liability" because the system's makers did not know it was discriminatory.

The usual mistake here is trusting the ranking because it looks tidy. A neat top five tells you nothing about the people below the cut.

Protect the applicants' information

A résumé is personal information: names, addresses, work history, sometimes age or immigration status. Canada's privacy regulators, in their guidance on AI, privacy and your business, ask organizations to limit the personal information they share with generative AI and to label when it has been used in making decisions.

In practice:

1. Use a business or team plan whose terms say your data is not used to train the vendor's models. Free personal accounts often do not promise that.
2. Remove phone numbers, email addresses and home addresses before pasting anything in. The AI does not need them to check your criteria.
3. Tell applicants. In Ontario, if you are covered and AI screens applications, the posting has to say so anyway. Elsewhere, one line in the ad is still good practice under PIPEDA, the federal privacy law that covers most private businesses.

Disclose, notify and keep records

If you are unsure whether your setup counts as AI screening, you are in good company. A February analysis by the law firm Osler notes that Ontario has not defined "screen", "assess" or "select", and suggests they likely cover tools that "filter, rank, score, recommend, or otherwise evaluate candidates". It says many employers now disclose AI use on every posting rather than decide case by case.

For a small business that is the simple answer. If your job board or applicant system does any ranking, add a line such as: "We use AI tools to help review applications. A person reviews every application before any decision."

Two more Ontario rules come with the same change. You must tell anyone you interviewed whether they got the job within 45 days of their last interview, and keep postings, applications and those notices for three years. A chat assistant can draft a polite, specific decline in a minute. Send it.

Common mistakes

• Letting the AI invent requirements. A draft that asks for a degree the job does not need loses you good applicants and may screen out people unfairly.
• Forgetting the job board has AI in it. Check the settings of any platform that sorts applications, not only the tools you chose yourself.
• Asking the AI "who is the best candidate?" Ask which of your written criteria each person meets. The first question hides its reasoning; the second shows it.
• Pasting everything into a free account. Strip contact details and use a plan that does not train on your data.
• Using AI to answer interview questions for you. Interview notes and your decision should be yours, and written down.

Where to start

1. For your next opening, write the five to eight criteria first, then have AI draft the ad from them, and check it against the posting rules above.
2. Open the settings of whatever job board or applicant system you use and find out whether it ranks or filters candidates. If it does, add the AI line to your postings.
3. On your next hire, read every application the AI would have rejected. If that pile looks wrong, stop using it for screening.

If you would like a second pair of eyes on your hiring setup before the next opening, that is a reasonable use of a 30-minute call.