Questions to Ask Your PR Agency About AI Before You Sign

What questions should I ask a PR agency about AI before signing?

Before you sign with any PR agency, ask 12 specific questions covering data handling, named AI sub-processors, training opt-outs, human-in-the-loop review, regulated-client carve-outs, and a written correction process. The goal is not to confirm the agency uses AI — every competent agency now does — but to confirm they use it responsibly, with named accountability and a public policy you can hold them to. An agency that fumbles these answers is either careless or hiding something. Either way, it is not the partner you want representing your brand to journalists, investors, and AI answer engines.

The PR industry quietly absorbed generative AI in under eighteen months. Research that used to take a junior associate three days now takes ninety minutes. Outline drafts, media list enrichment, byline ghostwriting, press release first cuts — all of it routinely passes through ChatGPT, Claude, or Gemini before a human ever sees it. That is not inherently a problem. The problem is the gap between what agencies actually do and what they are willing to disclose.

This is your buyer's checklist. Use it on the sales call. Ask the questions in order. Note the answers. The agencies worth hiring will give you direct, specific responses with documents to back them up. The rest will deflect.

Why does an agency's AI policy matter for your brand?

Three reasons, in order of how often they bite clients.

1. Confidentiality leakage. If the agency pastes your unreleased product roadmap, deal terms, or executive bios into a consumer ChatGPT account, that data may be retained, reviewed by humans on the vendor side, or — depending on the tier — used to train future models. API-tier access with a no-training agreement looks identical from the outside but is fundamentally different on the back end.

2. Fabricated facts in your name. Generative models hallucinate quotes, citations, statistics, and source URLs with complete confidence. A press release or thought-leadership byline that goes out under your name with a fabricated stat is your reputational liability, not the agency's. The fact-check protocol matters more than the writing protocol.

3. Regulatory exposure. If you are a Reg A issuer, a public company, or anyone subject to FINRA, SEC, or FTC rules, material non-public information cannot touch a general-purpose AI tool. A carve-out workflow is not optional — it is the difference between a clean filing and a securities lawyer's nightmare.

The 12 questions, by category

Data handling (questions 1–3)

1. What client data goes into AI tools, and what never does? You want a clear line. Acceptable inputs: public information about your company, draft copy you have already approved, research queries about your industry. Unacceptable: non-public financials, unreleased deal terms, executive PII beyond what is already public, anything you have marked confidential.

2. Where is my data stored when AI is involved? The right answer is "in our backend, not pasted into consumer AI interfaces." Agencies still copying client documents into the free ChatGPT web app are a hard pass.

3. Do you use the API tier with a no-training agreement? Consumer tiers (free ChatGPT, Claude.ai, Gemini consumer) may use inputs for training. Enterprise API tiers do not. The agency should know the difference and be on the right side of it.

Sub-processors and vendors (questions 4–5)

4. Can I see your list of named AI sub-processors? Expect a short list — typically OpenAI, Anthropic, Google AI, Perplexity, and one platform-level provider (in our case, Lovable AI Gateway). Each one should have a confirmed no-training-on-inputs commitment in writing.

5. What is your process if a vendor changes its data policy? Vendor terms change. The agency should commit to reviewing sub-processor policies on a recurring schedule and notifying clients if a material change occurs.

Human review and fact-checking (questions 6–8)

6. Who reviews AI-assisted output before it leaves the agency? Get a named role, not a vague "our team." Editor, account director, partner — someone with accountability.

7. What can never be AI-generated, period? The right answer includes: quotes attributed to real people, statistics presented as fact, source citations, journalist names, publication names, and any claim about a regulated client. If the agency cannot list these, they are not enforcing them.

8. What happens when AI output contains an error that gets published? Look for a written correction protocol: a public corrections page, defined response time, and a process for notifying you before any public correction is made.

Regulated and sensitive clients (questions 9–10)

9. Do you have a carve-out for securities-regulated clients? If you are a Reg A issuer or anyone subject to securities rules, this is non-negotiable. The agency should describe an isolated workflow where material non-public information never touches a general AI tool, with a documented chain of custody for drafts.

10. Can I request a zero-AI workflow on my account? The answer should be yes, with a clear addendum to the contract. Expect a longer turnaround and a higher cost — human-only is more labor — but the option must exist.

Transparency and accountability (questions 11–12)

11. Is your AI use policy public? A public policy is a forcing function. It commits the agency to specific practices in writing, in a place clients and journalists can hold them to. If the only AI policy is "trust us," there is no policy.

12. How do I report a concern about AI use on my account? A named contact, an email address, and a response-time commitment. Anything less is theater.

People also ask

Do PR agencies disclose when they use AI?

Most do not — yet. Industry surveys put real AI usage in PR shops above 90%, but published AI use policies are still rare. That gap is exactly the leverage you have as a buyer: ask the questions on this checklist, and the agency that answers cleanly is the one that has actually thought about it.

What are AI sub-processors in PR?

AI sub-processors are the third-party vendors (OpenAI, Anthropic, Google AI, Perplexity, and platform-level providers) whose models may process client data routed through the agency's workflow. A responsible agency names them, confirms each is on an API tier with no training on inputs, and updates the list when vendors change.

Can PR agencies use ChatGPT for client work?

They can, but the version matters enormously. Consumer ChatGPT (free or Plus) may retain inputs and use them for evaluation. The OpenAI API and ChatGPT Enterprise tiers do not train on inputs by default. If the agency is using the consumer web app for client work, your data is in a different risk bucket than if they are routing through the API.

Related searches

AI ethics in public relations

The AI ethics conversation in PR centers on three things: disclosure (does the audience know AI was involved?), accuracy (is the content fact-checked by a human before publication?), and consent (do clients know what tools touch their data?). The Public Relations Society of America (PRSA) has published guidance, but enforcement is voluntary and uneven.

How to vet a PR firm

The classic vetting checklist — case studies, client references, media placements, billing transparency — still applies. AI use is now a fifth pillar. Add it to your RFP and treat vague answers as a disqualifier, the same way you would treat vague answers about pricing or scope.

PR agency contract red flags

Watch for: no disclosure of AI tools, no carve-out language for confidential or regulated material, no defined correction protocol, no named human reviewer, and no written commitment that consumer-tier AI tools will not be used for your account. Any one of these in isolation is a yellow flag; two or more is a no.

AI in content marketing disclosure

Disclosure norms vary by channel. Press releases and earned editorial generally require no AI disclosure if a human reviewed and approved the final copy. Sponsored content and brand-owned blog posts increasingly do, especially in regulated industries. Your agency should have a position on this and apply it consistently.

How to use this checklist

Print the 12 questions. Bring them to the sales call. Ask in order. Take notes. After the call, request the agency's public AI use policy in writing — if they have one, it should land in your inbox within an hour. If they need a week to "put something together," you have your answer.

For full transparency on how we handle these questions ourselves, our AI use policy covers data handling, our five named AI sub-processors, the human review chain, the Reg A carve-out, and the correction protocol — in plain English, on a public page.