There's a number I keep landing on, informally, across the drift cases I've spent the most time with: roughly a year and a half. That's about how long it tends to take, in the slower-moving cases, between the moment a gap opens up between what a brand says it stands for and what it's actually producing, and the moment someone outside the building notices and says so publicly, loudly enough that the company can no longer treat it as an internal matter. CNET's compound-interest errors ran for months before an outside outlet flagged them. Sports Illustrated's fabricated bylines quietly reappeared under new names for a stretch before a journalist connected the dots. Eighteen months is a rough average, not a law, but it's consistent enough across the cases I've studied to be worth taking seriously as a pattern rather than a coincidence I happened to notice twice and got attached to.
Then there's the other kind of case — the one where the gap gets caught in days, not months, because the audience itself does the job nobody internal was assigned to do, and does it for free, in public, without waiting to be asked.
The diversity initiative that lasted six days
In March 2023, Levi Strauss announced a partnership with Lalaland.ai, an Amsterdam-based studio that generates AI-rendered fashion models, with the stated goal of using them to "supplement human models, increasing the number and diversity" of Levi's product imagery. On paper, that reads like a company acting on its own stated commitment to inclusivity. Within days, the backlash made a much simpler point than any AI ethics paper would: you don't increase diversity by replacing real models of different body types and skin tones with computer-generated approximations of them, while the actual people you'd be increasing representation for stay unhired. Critics called it lazy, and worse, across social platforms in numbers Levi's clearly hadn't anticipated when the announcement went out. Levi's issued a revised statement within the week clarifying that the initiative was never meant to replace real photoshoots or substitute for its actual diversity, equity, and inclusion commitments — a walk-back that, notably, conceded the framing had been wrong from the start, not just misunderstood by an audience that had somehow missed the intended meaning.
The gap was checkable before the press release, not after
Here's the detail I think matters most, and it's the same detail that shows up in every slower-moving case too: the tension was available to catch before the announcement went out, not just in hindsight. Someone, at some point, greenlit language that claimed an efficiency initiative — cheaper, faster product photography — was also a diversity initiative, without anyone apparently running the fairly obvious check of whether "generate synthetic diversity" and "increase representation for real people" were actually the same commitment or two different ones wearing the same sentence. That check didn't require external backlash to perform. It required someone internal to read the press release against the company's own stated DEI language and ask whether the two were actually consistent, the same read-it-back-cold exercise I keep recommending for AI-drafted customer communications, just applied one step earlier, to strategic initiatives before they ship.
Levi's got its answer in six days because the internet is an extremely fast, extremely public oversight mechanism when no internal one exists. Most brands don't get that speed of feedback. Most gaps run the full eighteen months, quietly, because nobody outside is looking as closely as Levi's critics happened to be, and by the time someone does look, the drift has had a year and a half to compound instead of six days.
Why the fast case and the slow case are the same mechanism
It's tempting to read the Levi's story as a completely different animal from CNET's slow-burning correction cycle, because one played out in days and the other in months. I don't think that's the right lens. Both are the identical structural gap — a stated identity claim and an actual initiative that didn't match it — just discovered at different speeds because the discovery mechanism was different. CNET's gap needed a specialist outlet to notice a factual error buried in financial arithmetic most readers would never double-check. Levi's gap needed no specialist knowledge at all — anyone scrolling past the announcement could see the mismatch immediately, which is exactly why the discovery clock ran in days instead of months. The lesson isn't "hope your drift is the visible kind so it gets caught fast." It's that the speed of external discovery tells you nothing about whether the internal check should have happened anyway, before either clock started running.
What actually shows up before the public notices, if you know to look
Across both the fast cases and the slow ones, the same handful of tells keep appearing well before any outside discovery:
Language that borrows the weight of a stated value without actually delivering it. Levi's press release used the word "diversity" to describe a technical efficiency, and the two aren't automatically the same thing just because one sentence claims they are. Watch for your own brand doing this with words like "personal," "trusted," or "human" attached to something that's actually just faster.
An efficiency initiative touching an identity-sensitive area with no explicit check run against your stated values before launch. Not every efficiency gain needs this level of scrutiny. The ones that touch representation, trust, safety, or anything your brand has made a public promise about specifically do, every time, before the announcement goes out — not after.
A growing volume of AI-touched output next to a flat or shrinking review capacity. This is the CNET and Sports Illustrated pattern specifically — production scaling past the point where anyone's actually reading a representative sample of what's going out. If your output has grown and your review process hasn't grown with it, you're already inside the eighteen-month window; you just don't know which month you're in yet.
Turning the timeline into an actual habit
If eighteen months is roughly the runway before an unwatched gap gets noticed from outside, the practical move is to put a check somewhere well inside that window, rather than waiting for it to expire and hoping you get lucky the way Chevrolet and DPD eventually did with their much smaller, much faster-resolving incidents. A quarterly read of your last month's AI-touched output, run four times before the eighteen-month mark would otherwise arrive, gives you three separate chances to catch a widening gap while it's still small enough to fix quietly, internally, without anyone outside ever needing to notice it existed at all. Waiting for the annual review, by contrast, gives you roughly one shot to catch something that's already had most of a year to compound, which is a much harder position to fix from and a much more visible one once you're forced to.
I'd also add a specific check for anything shaped like the Levi's case — an announcement, campaign, or initiative that explicitly claims to serve a stated value, not just routine customer-facing content. Those deserve a dedicated read before launch, not folded into the general monthly sample, because the stakes of getting the language wrong in a public announcement are structurally higher than the stakes of one slightly-off customer email, and the discovery clock on a bad announcement, as Levi's found out, can run in days rather than months.
The actual point of naming a timeline at all
I don't share the eighteen-month figure to give anyone false comfort that they've got over a year of runway before anything's at stake, and I'd actively discourage reading it that way. I share it because the length of the gap is itself the finding: drift has time to compound precisely because almost nobody is checking during that window, and every week inside it is a week where the fix would have been cheaper and quieter than it will be once someone outside finally notices. The tells are visible the entire time. The only real question is whether anyone's assigned to look for them before month eighteen arrives, or before six days of public backlash does the looking for you instead, at a much higher volume and a much lower level of privacy than an internal check would ever have cost you.
If you suspect your own brand might already be somewhere inside that window, the free Brand Bloom Audit is built to help you find out before day six or month eighteen decides it for you. Take the audit →
References
Levi Strauss & Co. (2023, March 28). Statement on the Lalaland.ai pilot partnership.
NBC News. (2023, April). AI models Levi's controversy backlash.
Rangefinder. (2023, April 6). After backlash, Levi's says AI-generated models "to increase diversity" won't replace real shoots.
Stanford, K. (2023). Levi's faces backlash after announcing it will use AI to "increase the number and diversity" of models. In The Know (Yahoo).