We Need to Get the Facts Straight. Make Your AI Show Its Receipts

Happy Wednesday! This week:

Let’s get started!

Not long ago, I realized I'd skimmed right past a number the AI had pulled into a deck I was building. I went back and checked it — it was right, thankfully — but I couldn't shake the question: what if it hadn't been? The number was tucked into a sentence that read beautifully, and that's exactly why I'd sailed past it.

For a sense of what "what if" looks like, ask Deloitte. Its 237-page report for the Australian government — a roughly $290,000 engagement — included an inaccurate quote attributed to a federal court judge and cited expert reports that don't exist. A university researcher caught it, the story made headlines around the world, and Deloitte refunded part of the fee. That's a major professional services firm handing a client a deliverable with invented citations in it.

Here's the problem: AI is a fluent writer, and fluency is camouflage. A wrong number reads exactly as smoothly as a right one. When you proof, you're reading for flow, tone, and whether the argument lands — and AI copy almost always passes that test now. The facts sit inside sentences that sound right, which is what makes them so easy to miss.

I want to be careful here, because I think expecting AI to be perfect is one of the biggest things holding people back. They see one hallucination and decide the whole thing can't be trusted. But let's face it — we make mistakes too. Every communicator has a typo, a wrong title, or a stale stat somewhere in their past, and the answer was never to stop writing. It was to build a check into the process. LLMs suffer from similar problems, and when it comes to them, the old “nobody’s perfect” adage still applies. We just need to account for that.

For a while, my safeguard was to ask the chatbot to do a sweep of everything it cobbled together — validate all the facts it or I asserted, check all the links to make sure the primary sources support those assertions. It didn’t matter if I found the detail or if it did — I wanted it double-checked. It's a good habit, and I don’t intend to get rid of it. But when it comes to the facts that the LLM discovered (or maybe, “discovered”), it’s me asking the AI to grade its own homework.

The better fix came from a fellow traveler on a standing AI-in-communications call I'm on (if you'd like in, just reply to this email and I'll see what I can do). I was discussing this on the call, and he shared what he does: one last step where he asks the AI for a bulleted list of every assertion of fact in the piece. The stats and claims come out on their own, away from the narrative. His comes back with something like "here are the nine stats I cite," and those are what he checks. It was a great idea — and I took it a bit further.

I made it a standing rule in my AI chief of staff. (No code involved — I just told Claude this was a standing rule going forward.) Now every draft it hands me, whether it's an email, a LinkedIn post, a deck, or this newsletter, ends with what I call a fact receipt. Every factual claim gets its own line, the source it came from (this is the critical addition!), and one of three tags: ✅ something I previously told it I checked, 🔁 something I told it myself, or ⚠️ unverified. The ⚠️ items go at the top so I can't miss them.

Think of it like a grocery receipt. You don't replay the whole trip through the store to see if you got charged twice for the avocados — you scan the line items. Each check is a click and a read, a few seconds apiece. I'm hoping it helps with something else too: I've caught myself asking for "one more check" over and over before I hit send, and a receipt with every line checked is a pretty clear finish line.

Why this matters for communicators: Fact-checking was always our job, and AI hasn't changed that. What it has changed is where the errors hide. The editing question used to be "does this read well?" AI copy will almost always read well now, so the question becomes "what is this claiming, and who says so?" That's a judgment call, and judgment is still ours.

Your next steps:

Bob's Your Uncle has spent twenty years helping challenger brands win their categories. But when buyers asked ChatGPT about their own category, the agency didn't show up.

With ChatFeatured, their AI visibility went from 7% to 22% in 90 days, and they went from invisible to first in ChatGPT.

They're not the only ones:

Vessel: AI visibility up 2,200%+

SCOJO: AI visibility up 160%

WikiSleep: brand mentions up 113%

ChatFeatured finds the questions where AI leaves you out, writes the content that gets you cited, and reports the lift every week. Your team decides what ships.

Most customers see their first citations inside 7 days.

Get recommended by AI

Up top, I wrote about catching AI's mistakes before they leave your desk. This week, Sarah Evans turned the question around: what is your personal LLM doing to the information you see about yourself?

In her latest PR@ctical, Sarah — the CEO of Zen Media — shares a prompt that asks an AI to put everything it believes about a company on one page. Each line gets marked as something it remembers or something it's guessing, and it ends by naming the fact it would fix first. The clever part is that she ran it on her own agency twice. As herself, her ChatGPT knew Zen Media inside and out. Then she added one line telling it to "answer as if a stranger sent this prompt," and it described a generic digital marketing agency, got the category wrong, and couldn't name the current CEO. "Nothing else changed between the two runs except who was prompting," she writes.

She also points to where a lot of these errors start. Per data from Profound, which sells AI visibility tools, 54% of brands checked had at least one inaccurate AI claim that cited the brand's own content — and pricing was 12% of claims but 24% of the errors. Muck Rack is now treating this as something to measure, too: at its Generative Pulse Summit last week, Matt Dzugen made "error rate" one of four parts of what Muck Rack calls AI portrayal, as Gab Ferree recapped.

Your AI’s memory makes your brand look better known than it is — to everyone else, it’s a stranger.

The AI you use makes mistakes you can catch. The AI talking about your brand makes mistakes you'll never see — unless you go look for them the way a stranger would. It's the other half of the two paths I keep coming back to: the AI you use, and the AI that acts on your brand whether you use it or not.

Why it matters: Your own AI knows your organization because you taught it. Your buyers, reporters and job candidates are asking an AI that never learned any of that.

Your next step: Run Sarah's prompt on your organization with her stranger line at the top: "Ignore your saved memory, custom instructions and anything you know about me." When something comes back wrong, check your own website first — the About page, old PDFs, legacy pricing pages.

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.

Get $5 off Lovable

_{^{Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.}}

*Sponsored: I may earn a commission if you sign up through this link.

There’s plenty of talk about AI changing comms. What do the people doing the work actually think? Luke Sinclair and Gemma Saint from Signal Intent are gathering perspectives for their second Intelligence report: what’s promising, what’s proving difficult, and how ready teams feel for the next three years. If you lead or help shape employee communications, take the 3–4 minute survey. You can request the findings when they’re ready.

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