What to look for in an AI tool for PR

PRSA, the CIPR and the Global Alliance for Public Relations and Communication Management have each published guidance on using AI in public relations. The guidance describes what a practitioner should check when choosing a tool. The questions below come from it, so you can use them when you compare tools.

1. Where does your client's information go?

PRSA's "Promise & Pitfalls" guidance, released in November 2023, is organised around five provisions of the PRSA Code of Ethics, and one of them is Safeguarding Confidences.

A September 2026 article in PRSA's Strategies & Tactics by Dr. Gleb Tsipursky puts that into practice. He lists what counts as confidential: unreleased financials, personnel matters, litigation, customer data, product plans and crisis details. He says teams should use templates that prompt staff to strip personal data, client secrets and embargoed details before they prompt an AI system, and that "prohibited use should include uploading protected information to an unapproved system."

Before you test a tool, find out what it sends to which provider, whether it stores that content, and whether it uses it for training. These questions are my suggestion, and they follow from the guidance above.

2. Is it a system your team has approved?

Tsipursky's rule is that employees use only approved systems with approved categories of information. He sorts AI work into three tiers: low-risk assistance, governed use and prohibited use. Decide which tier a tool belongs in before anyone puts client material into it.

Ask a vendor the data questions from the first section in writing, and keep the answers. Vendors that answer clearly make the approval decision simpler.

3. Does it keep a person in charge of the result?

The Global Alliance's Responsible AI principles say that AI use "must be governed by human oversight and aligned with public interest," and that practitioners "are responsible for the AI-generated and AI-assisted work products they produce, publish, and share."

Tsipursky adds what that looks like in a workflow: "Review should name the person, the source of truth and the questions that must be answered before release." For a media pitch, his examples are checks on the reporter, the claims and the relevance of the sources.

A tool that returns a finished-looking result makes that review easy to skip. Look for one that gives you a draft to work on.

4. Can you check where a claim came from?

The Global Alliance principles expect professionals to research thoroughly, rely on trusted sources, fact-check and think critically. Review goes faster when the tool shows where a statistic or a quote came from. If a tool names a figure with no link, you have to find the source yourself. That point is my own addition, based on the guidance.

5. Does it fit your disclosure rules?

The Global Alliance's fourth principle says "open disclosure and transparency in the use of AI-generated content, data, and interactions are essential." Tsipursky recommends that teams define disclosure triggers in advance, and he suggests asking four questions before release: what did the system create, who verified it, what could change the audience's interpretation, and who approved release.

PR Daily's 2023 report on the PR Council's guidelines says they advise telling clients when AI has played a "substantial role" in creating content.

Decide your own rule first, then check that a tool's output and workflow let you follow it.

6. Does it match a checklist you can reuse?

The CIPR's guide, "The Responsible Use of AI in PR" (June 2026), takes each of the Global Alliance's seven principles and works through the dilemmas practitioners meet in practice. It includes a checklist for assessing AI tools before use, handling data, reviewing outputs and the questions to ask before publishing AI-assisted content. It is open to non-members as well as members, and the CIPR positions it as a reference for evaluating tools.

How The PR Assistant answers these

These answers come from our privacy policy, our AI Governance page and the product. Your request and the client's saved context are sent to our AI provider, currently Anthropic, to generate a response. Some tools also use live web search, which sends your query terms to that search function. We don't use account or client content to train AI models, ours or our provider's. Each client's data is scoped to your own account or organisation. Research steps in the workflows are set to link the pages they used, and every result is a draft for you to review.