Before You Hand It to AI, Ask What Happens If It's Wrong

Happy Wednesday! This week:

Let’s get started!

A lot of the AI conversation in comms is about speed — how fast we can adopt, how much faster we can produce. Less of it is about what happens when the output is wrong, and who's on the hook when it is.

Catharine Montgomery and I got connected a few weeks ago and ended up swapping notes as fellow practitioners in the digital channel space. A lot of what she said matched my own thinking, so I asked if I could share it here. She's the founder and CEO of Better Together Agency, with more than 15 years in PR and crisis communications, and she leads the agency's annual Biases in Generative AI Survey, now in its third year.

Her focus is the part most rollouts skip: accountability, human review, and the people on the receiving end.

Let's jump in.

🤔 Everyone in comms is being told to adopt AI. Before a team hands a task over to it, what should a leader be asking?

💬 Catharine's take: Start with the task and the people affected. What problem are we trying to solve? Will AI help us do this work more accurately or make it more accessible, or will it mainly produce more content faster? Then ask what happens if the output is wrong. Brainstorming a headline is different from giving someone information about a job, a health service or a public benefit. If the team cannot check the answer, explain its use or correct a mistake, it may not be the right place to use AI.

🔎 My take: "What happens if the output is wrong" is the question I wish more teams asked first. If you can't tell whether the answer is right, you haven't saved time — you've just handed the work to whoever finds the mistake later. And that, of course, may not be someone internal, making matters even worse. Having an “oops” plan is key, and I don’t think anyone has prepared for that fully.

⚖️ Say AI turns out a polished message that's biased or just plain wrong. Who owns that mistake — and what should happen when someone it affects calls it out?

💬 Catharine's take: The organization using the tool owns the message it puts in front of people. A vendor may also need to fix its product, but pointing to the vendor does not repair the harm. Someone should be able to reach a person, explain what went wrong and get a meaningful response. Internally, the organization needs to find out how the error passed review, correct it where it appeared and test for the same problem elsewhere. "A human was in the loop" means little if that person lacked the time or authority to challenge the output.

🔎 My take: Longtime readers know I shy away from the phrase "human in the loop," as it centers our work around AI with us relegated to a safety net. Catharine's last line makes a great point as well: a reviewer who can't push back isn't a check — they're a rubber stamp.

👥 Who tends to be missing from the room when organizations decide how AI gets tested and used?

💬 Catharine's take: The people most fluent in AI tend to dominate the conversation. Yet someone can be affected by an AI-generated message or an AI-assisted decision without ever opening a chatbot. That is one reason our 2026 Biases in Generative AI Survey includes adults who do not use generative AI. We are still collecting responses, so I will not claim to know what this year's findings show. The larger point is that organizations should listen to the people represented in their content and affected by their systems, then use what they hear to shape testing and correction.

🔎 My take: Guilty as charged — most of the AI conversations I'm in are with people who use it every day. The people reading what we produce often aren't, and they're the ones who have to live with it.

📣 Comms teams often get looped in after the AI decision is made, mostly to explain it. What should they have a real say in before a tool launches?

💬 Catharine's take: Communicators know how a message can land when context is missing, a source is weak or a community is misrepresented. They should help define where AI is appropriate, examine the material a tool draws from and test outputs with the audiences and situations the organization actually serves. They should also help set the rules for disclosure, human review and corrections. That role requires authority to raise a problem before launch and pause a use case when the evidence calls for it. Writing the rollout announcement after every decision has been made is too late.

🔎 My take: This is the one that matched my thinking most. It echoes what Ellen Griley told me a few weeks ago — "no one stopped to listen." Comms is often the last to be asked, when it should be one of the first.

🛠 Say the research does surface real concerns about bias and trust. How does an organization turn that into changes it can actually test — and be held accountable for?

💬 Catharine's take: Research should lead to specific questions about a specific use of AI. Where do people encounter it? What could go wrong for them? Who checks the output and who can correct it? An organization can test representative examples, document recurring errors and ask affected people whether a change actually helped. It does not need a large governance department to name an owner and create a way to report problems. Beyond the survey, I want to help communications and leadership teams turn public experience into practical testing, training and decisions they can revisit as the technology changes.

🔎 My take: Don't skip past "it does not need a large governance department." For a small team, naming an owner and giving people a way to flag problems is something you could do this week.

Thanks to Catharine for the insight. You can find her at Better Together Agency, and her team's 2026 Biases in Generative AI Survey is still open. It deliberately includes people who don't use AI tools, so if that's you or someone you know, it only takes a few minutes.

A fast AI answer isn't always a coordinated one. When AI resolves B2B issues without looping in the teams responsible for them, confidence outpaces alignment. A new Harvard Business Review Analytic Services briefing paper, sponsored by Front, looks at closing that gap.

Read the Briefing Paper

Meta's new AI agent, Muse, has been out for three weeks and is already sitting at the top of the charts. It hit No. 1 on the U.S. App Store on Sept. 18 and passed 3.4 million downloads, according to Sensor Tower data reported by TechCrunch. Unlike a chatbot, Muse is built to do things. Meta says it can send email, fill out forms and book travel, and it "keeps working after people close the app." At Meta's Connect event last week, the company added a big one: Muse is getting its own email address, so people can forward messages to it and let it handle them.

Maureen Kerr, writing for Forbes, zeroes in on a different question: "who decides whether it may stand between a business and its customers." Shopify opened its stores to Muse. Amazon blocked it, saying agents that buy on customers' behalf "should operate openly and respect service provider decisions about whether or not to participate." The line from her piece that stuck with me is written for publishers but applies to all of us: an agent "can read a publisher's reporting without generating a page view."

For communicators, that's the part to sit with. We've spent the past year learning how to show up in AI search — making sure ChatGPT tells our story correctly. Agents are the next step. If people start handing their inboxes and errands to an agent, your email, your announcement or your customer note may get read, summarized and acted on by software before a person ever sees it — if a person sees it at all. And you may not be able to tell the difference: Kerr notes that Meta's own documentation says when Muse browses the web, it appears as the user's activity.

Why it matters: An app that does things on people's behalf picked up millions of users in three weeks. There's a new middleman between you and the people you're trying to reach, and it's not waiting for anyone's permission to grow.

Your next step: Take your last important email or announcement, paste it into an AI assistant, and ask it to summarize the message and say what it would do next. If the summary misses your point, someone's agent will too.

I spent 15+ years running digital and AI-adoption strategy inside global brands — the kind of enterprise-scale channel work most "AI consultants" have never actually done from inside a company that size. I bring that same playbook to comms teams directly now: workflow audits, executive AI systems, the intranet-and-channel rebuilds I used to run in-house.

A handful of clients at a time. Reply and tell me what you're stuck on.

If identity theft happens, Coveron doesn't leave you alone. Get dedicated case managers, legal support, and up to $1M in recovery assistance. We handle the mess so you don't have to.

One scam can cost you everything - protect yourself now. The first 100 users get 20% off with code beehiivenewsletter.

Get protected today

_{30-day money-back guarantee. Terms and conditions apply.}

Looking for a role at the intersection of communications and AI? Here are some opportunities to check out:

Was this forwarded to you by a friend or colleague? Want to get this in your inbox next week? Click here to subscribe!

Quick question: how can I help?

What workflows are you struggling with? Where does AI still feel mysterious or overwhelming? What has worked that you’d like to share with others?

I’m a builder, and I’d love to help you and the rest of The Comms Stack community find great new ways to use AI.

Reply and tell me.
I read every response.

Even a one-sentence reply helps. For example:
“I wish AI could help me with ______.”

Until next Wednesday,

Dan

Advertise | Subscribe