I have a bad habit of hovering my cursor over bylines. Not because I'm suspicious of every writer on the internet — I promise I'm more fun at parties than that sentence makes me sound — but because somewhere along the way, "who actually wrote this" stopped being a trivia question and became a genuinely useful diagnostic for whether a brand still knows what it sounds like.
Which is how I ended up thinking, for the hundredth time, about CNET.
The byline you had to hover over to see
In late 2022, CNET quietly began publishing personal-finance explainers under the byline "CNET Money Staff." Nothing in the visible byline said the articles were AI-written. You had to hover over it to see a small disclosure noting the piece had been produced "using automation technology." Seventy-seven articles went out that way before anyone outside the building noticed.
Then someone did notice. A tech-industry outlet flagged factual errors in one of the pieces, and CNET ran a full internal audit. The result: more than half of the seventy-seven articles — forty-one of them — needed corrections, some described by CNET's own editor-in-chief as substantial. One explainer on compound interest got basic arithmetic wrong in a way that would have handed a reader actively incorrect financial advice, not just an awkward sentence. CNET paused the program, then, within months, said it intended to keep using the tool going forward, just with a tighter review process attached.
I want to be precise about what happened here, because it's not the story people default to. Nobody at CNET set out to mislead readers about how much interest they'd earn on a savings account. What happened is quieter and, I think, more useful to understand: a production process got automated faster than the editorial identity of the outlet could keep pace with it, and the gap between the two ran for months before anyone was looking closely enough to catch it.
That gap has a name now. I call it Brand Drift — and naming it precisely is most of the point.
Why "drift" is the right word, and "failure" isn't
Every brand has two versions of itself running at once. There's the one it intends — the values, the tone, the standards it would tell you about if you asked. Aaker and Kapferer, whose work sits underneath most serious brand theory, call this the identity: the internally held sense of what the brand is actually for. Keller's customer-based brand equity model adds a useful second half to that picture — the pyramid of salience, performance, judgments, feelings, and resonance that gets built entirely out of whatever the brand's touchpoints have actually said and done. That second half is the image: the version that actually reaches people, built out of whichever combination of human and automated output happened to produce it.
Those two versions are supposed to track each other closely. Drift is what happens when they stop. Not because someone made an active decision to let standards slip, but because nobody was explicitly translating "here's who we are" into "here's what the system that now writes half our output should and shouldn't say" — and the system, left to its own devices, produced exactly what you'd expect from something with no encoded identity: content that's fluent, plausible, and quietly generic. CNET's articles read fine on the surface. That's precisely the problem. Fluency is not the same thing as fidelity to a specific editorial standard, and AI is very good at the first and has no innate access to the second unless someone has done the deliberate work of building it in.
The reason "failure" is the wrong word is that it implies a single decisive moment you could point to and fix. Drift doesn't work that way. It compounds in increments too small to trigger anyone's attention individually — one slightly-off explainer, one unreviewed batch, one quarter where nobody was checking as closely as the last — until the gap is wide enough that an outsider walks in and asks the question nobody inside the building had gotten around to asking: does this still sound like you?
Why smart teams fall for this anyway
I think the honest, slightly uncomfortable answer is that fluency is genuinely persuasive, even to people trained to catch problems. A sentence that's grammatically correct, confidently phrased, and roughly on-topic clears most people's mental bar for "this seems fine" without ever being checked against the much narrower, much more specific bar of "this is exactly what we'd say, the way we'd say it, with the care we'd bring to it." Editors proofread for typos because typos are visible. Drift is invisible by design, because the thing that's missing — a specific, deliberate point of view — is precisely the thing that's hardest to notice the absence of. You don't clock a missing opinion the way you clock a missing comma.
This is, in miniature, the exact mechanism behind Sports Illustrated's fabricated-byline scandal the same year — a case I've written about at length elsewhere, so I won't re-walk the whole thing here, except to say the shape is identical: fluent, plausible, technically functional content, produced faster than anyone with real editorial authority was checking it against what the brand actually claimed to stand for.
What made it findable
Here's the part I keep coming back to, because it's the actionable half of the story. CNET's drift wasn't invisible in any deep, unsolvable sense. It was checkable the entire time — any editor who sat down and read a sample of the AI output against the outlet's own editorial voice and accuracy standard would have caught the compound-interest error before publication, not after. The gap existed because nobody had built that checking into the workflow as a standing practice, not because the gap was undetectable.
This is the finding that keeps showing up across every version of this story I've researched, across very different industries: drift is rarely a detection problem. It's a "nobody was assigned to look" problem. The information needed to catch it is almost always sitting right there in the output, waiting for someone whose job includes actually reading it.
What I'm actually asking, underneath the framework
My own research treats Brand Drift as a specific, structural phenomenon — not a vibe, not a synonym for "the content felt off this week," but a real gap between stated identity and actual output that opens under a specific, nameable condition: production scaling faster than oversight. That's a testable claim, which is exactly why I think it deserves more rigor than "AI ethics" conversations usually apply to it. Broad AI ethics asks whether a system is fair or biased at a societal scale. I'm asking something narrower and, for a founder reading this on a Tuesday, more immediately useful: is the gap between what your brand says it is and what it's actually outputting this month widening or holding steady — and who, specifically, is watching that gap?
If you can't answer the second half of that question with a name, not a department, not a vague "the team handles it" — that's the finding. That's drift, waiting for its own version of the hovering cursor.
The tell you can check for yourself, today
You don't need an audit team. Pull your last twenty AI-touched customer communications — emails, replies, product descriptions, whatever your tools have produced recently — and read them back to back, cold, the way a stranger would. Not for typos. For whether you'd have said it that way, in that tone, with that level of care about being right. If the honest answer is "mostly, but not that one," you've just done, in five minutes, the exact review CNET didn't do until an outsider forced it.
I think about this obsessively, in the same way I think about most strategy problems — turning it over while I'm supposed to be relaxing, the way other people replay an argument they wish they'd won. But this one earns the obsession. The businesses I talk to across this region are automating their brand touchpoints faster than almost anyone is watching them, and the gap doesn't announce itself. It just sits there, compounding, until a stranger with a cursor and some curiosity finds it first.
Wondering whether your own brand's output still matches what you'd say it stands for? The free Brand Bloom Audit is built around exactly this question. Take the audit →
References
Aaker, D. A. (1996). Building Strong Brands. Free Press.
Duffy, C. (2023, January 25). Plagued with errors: A news outlet's decision to write stories with AI backfires. CNN Business.
Guglielmo, C. (2023, January 25). CNET editorial statement on the use of AI-assisted articles. CNET.
Keller, K. L. (2013). Strategic Brand Management: Building, Measuring, and Managing Brand Equity (4th ed.). Pearson.
Kapferer, J.-N. (2008). The New Strategic Brand Management: Creating and Sustaining Brand Equity Long Term. Kogan Page.
Leffer, L. (2023, January 17). CNET is reviewing the accuracy of all its AI-written articles after multiple major corrections. Gizmodo.