Like you, we've been using AI across our entire business, and our dependence on it to scale our work is ever-increasing. However, what we're now seeing is that the use of it to gloss over gaps in information is accentuating an underlying structural problem that I see in many organisations.

Let me explain. What we call 'confident garbage' is when an AI tool provides an output that is highly polished and sounds convincing, but is actually grounded in very little of substance.

We all can recognise the tells. It started with em-dashes. They are grammatically correct things, but no human ever uses them. I can't actually find the em-dash key on the keyboard as I write this. There are other almost harmless ones that you'll notice. AI loves the word compounds. Everything compounds over time, if you keep adding more value one step at a time. I can't recall a time when a human actually told me that their value compounds, yet I read it everywhere in their copy.

But more seriously, when our commercial engine is trying to convince potential buyers and we don't have a sound basis for them to believe us, AI will do an excellent job of polishing over that gap. Some examples.

In a recent proposal I read, there were 6 sentences with the word "most" in them, none of which were substantiated. Most buyers... Most companies in this space... Most customers... etc. There was no data, so AI just made up a false assertion, and the human supervising this let it through. It passed the human gate as it seemed harmless, but convincing.

Why Is This a Problem?

The reason we should stop now and force ourselves to retrench from the reliance on AI to help us craft 'convincing' messaging is that the research now tells us that it is having the exact opposite effect.

When we put out polished messaging without substance, our audience now trusts us less, perceives us as less competent and even less likeable, and is less likely to engage with us in future.

A team of researchers at the University of Tokyo carried out a study with 261 people, collecting 990 responses, where they measured perceptions of trust, goodwill, competence, likability and likelihood to engage in future, against the ratio of AI disclosed in writing that had different purposes.

The outcome was clear. The more AI people believed a writer had used, the less they trusted them, the less goodwill they felt, and the less competent and likeable they found them. The drop was sharpest in writing meant to build a relationship, where people were also less keen to engage with that writer again. In short, if you want people to trust you less, have less goodwill towards you, believe you to be less competent, be less liked or be less keen to work with you in future, then you should make your use of AI more obvious to them.

Their conclusion, naturally, was that we now need to create "transparent, context-sensitive writing systems that better preserve trust and authenticity".

The time has come for us to move beyond the hype and excitement of pushing out more and more content with AI, and to return to outputs that are grounded in evidence and shaped by human judgement.

The Problem We See Beneath Confident Garbage

When we see a lot of confident garbage coming from inside companies, there's a bit of a pattern taking place. Leadership is often insisting on leveraging AI to create greater efficiencies. There are fewer humans involved and we're asking for more output in the hope that volume overcomes lack of depth. And of course, there's a clear and accurate perception that AI is able to gather much more data than any number of humans could, so it's 'better'.

However, the nub of the issue is that AI systems are better at gathering more external data, but they are generally not being used to gather more internal data. If we're just pointing at external data sources, then we are simply recycling everything that is already in the public domain. It's a zero-sum game.

What sets us apart. What makes us unique. This can only come from inside. Inside your organisation. Inside your own head.

What I've seen happen recently is that with fewer people, especially in marketing functions, there is less and less coming from within the organisation. Marketing has probably suffered most from the "AI will easily replace humans" conversation that takes place across leadership teams. The perception is that AI can now do most of the things that the marketing team did in the past. But the point is missed. Marketing's only function is to convince the world why it should buy from us... what is uniquely compelling about our offer... and that only comes from inside the business.

In an organisation where we have fewer Marketing folks talking to Sales, Customer, Product or Operational teams, we end up with less of what actually helps us win and grow more customers bubbling to the surface. Without those human ears listening to what's actually working for our business today and without the human judgement anchored in years of experience, we're straight back to the confident garbage world.

Scratching the Itch

At Liffey, we help organisations with this problem. We step in and listen, with our human ears. And we apply human judgement to the patterns we spot. Very often, the keys to growth and success are in those patterns. It can be as simple as 5 different customers mentioning the same opportunity over 3 weeks. On its own, that's a signal. With human judgement applied to that and based on the full internal and market context, that could be a massive unlock.

AI can help massively with this. We love AI, but we're stepping right back from trusting its outputs that are not also anchored in the full internal context of a business. This bit is hard, which is why we've been building our own commercial brain, which connects with all of those invaluable internal sources: everything from Slack conversations to commercial team meetings and calls with prospects and customers. It's all gold, and we capture it.

We're also building out our own 'house rules', a set of checks and guidelines for what we will stand over. We check for everything that AI and humans have a tendency to do that undermines the credibility of our work. We force ourselves to never make a claim or assertion without it being grounded in human judgement backed by evidence that we can put our hands on.

We're happy to share the output of this, and you can run your own website or copy through it by checking out our helpfully named Confident Garbage Checker.

The 33 Checks We Run

Right now there are 33 things we fight with our own AI systems almost every day. We've locked them in as what we call our "house rules". Nothing gets past us unless we check it and flag any exceptions for human judgement. We share them here, nicely formatted with the help of AI, so maybe you can start to see where your own tooling might be letting you down.

The Five Questions

  1. How Do You Know? Is there any evidence behind the claim? A grounded claim points to where it came from. An ungrounded one just sounds more certain.
  2. Could This Be Any Business? Swap your company name for a competitor's. If the copy still reads true, it says nothing about your business.
  3. Who Says So? Quoting our clients or industry experts borrows their authority without naming anyone. A source can stay anonymous, as long as a reader can tell who they are, like a logistics client, Q2 2026.
  4. Could a Reader Check It? A standard, award, partner or piece of research counts for nothing until it is named clearly enough for someone to look it up.
  5. As at When? A count, a price or a status was true on a particular date. Without that date, a reader can't tell whether it still is.

What You Can Charge

  1. Define Yourself by What You Are. Saying we're not another X puts that very objection in the buyer's head. You can always say what you are, and it tells the buyer the same thing.
  2. Show What Makes You Different. Leading and best-in-class ask a buyer to believe you are slightly better, which nobody can check. Say exactly what you do that others don't, and who it is for. That is a claim only you can make.
  3. Lead With What the Buyer Gets. The buyer is buying the result. They only start to care how it works once they want it.
  4. Never Sound Small. Language that makes you sound small, one person named as the whole service, or a time of day attached to what you do all give a buyer a reason to value you less.
  5. Never Reduce the Offer to an Hourly Rate. Dividing a fee by the hours makes expert judgement look like time for sale, and every extra hour you put in makes the number look worse.
  6. Speak Ill of Nobody. If another approach has a weakness, describe it as a strength of yours. In a customer story, blame the process, and leave the client's people out of it.

Whether You Are Believed

  1. Make Claims About Yourself. Most businesses struggle with this is a guess about everyone, with nothing counted. Put a claim about your own business there, one a buyer can check.
  2. Every Claim Can Answer How Do You Know? One big claim you can't back up in a meeting makes a buyer doubt the claims around it that you can.
  3. Use Numbers That Stay True. A number that changes every week, like a client count, soon makes the page look out of date. A made-up exact figure looks measured until someone checks it.
  4. Give Every Action a Person. One number starts everything has no one doing anything. Write the person and what they do, because a reader can check that.
  5. Name Who Does What. A buyer can judge a named person who reviews your position every month. They can't judge a capability, a layer or a dynamic.
  6. Say What You Mean. Words like quietly and tellingly hint at a criticism without saying it. A reader can't check a hint or answer it.
  7. Never Write a Seesaw. Not X, it is Y spends half the sentence on something you are not claiming, and the reader remembers the wrong half.

Whether You Are Read

  1. Plain English, Even on a Technical Page. Your specialist will check the detail, but the person signing off the purchase has to follow it too.
  2. Strip the AI Tells. These habits mark writing as machine-made: em-dashes, strings of very short sentences, lists that always come in threes, and neat one-line endings.
  3. Break the Rhythm Before It Becomes a Chant. Three sentences in a row with the same rhythm sound like a machine wrote them.
  4. Choose the Plain Word. Leverage, seamless and delve sound like business writing, and the plainer word is almost always the more accurate one.
  5. Use the Verb That Says What Happens. Words like holds, carries and lands sound thoughtful but hide what actually happens, so a reader can't check it.
  6. Start at the Claim. It is worth noting weakens the sentence that follows and adds nothing.
  7. Never Grade Your Own Argument. State the fact, and let the reader decide what it is worth.
  8. Keep Internal Names off the Page. The names of your internal systems and processes mean nothing to a buyer. They are buying what those systems produce.
  9. Colons Introduce Lists, Nothing Else. Using a colon to build suspense before the end of a sentence is a writing trick that draws attention to itself.
  10. Don't Talk About the Page, or Announce What's Coming. This page explains and here is the key point narrate the writing. A person would just say the thing.
  11. Don't Point to Other Parts of the Page. Above, below and the third one make no sense once a line is quoted on its own. Use the name of the thing.

How the Copy Travels

  1. Every Question and Heading Stands Alone. Search engines and AI assistants lift headings off the page, so each one needs to say what it is about.
  2. Say the Same Thing the Same Way Every Time. A guarantee or a form notice worded identically on every page looks like a policy. Wording that changes from page to page looks made up on the spot.
  3. Write to the Reader, in the Present Tense. Use the present tense for what the reader has, contractions where people would use them in speech, hedging only where things really are uncertain, and you for the reader.
  4. Structure the Page for the Reader. If you are listing things, use bullet points. Give each page one call to action. Break up long blocks of text so people keep reading.