Entity Drift Over Time
Entity drift is the slow widening of the gap between what your business actually is and what AI systems believe it is. No rebrand caused it. No relocation, no pivot, no crisis. The business went about its ordinary work at an ordinary pace, and the description ChatGPT, Claude, and Gemini hand a prospect got a little further from the truth each year.
So the direct answer to the question underneath the phrase: yes, this is happening to your business, and it is happening whether or not anything went wrong. Accuracy inside an AI system is not a state you reach and then hold. It decays, because the evidence these systems read is in constant motion, and almost none of that motion is yours.
That separates drift from every other AI understanding problem an owner hears about. The others come with a cause attached: you changed your name, you moved, you added a service. Drift has no such moment. Its cause is elapsed time, which is exactly why nobody checks for it.
Drift Does Not Require Anything to Go Wrong
Start with your side of the gap, because it is the part owners underestimate.
Businesses change in increments too small to announce. The minimum job size crept up over three years. Two of your five services quietly became most of the revenue. A second location closed without ceremony. The founder stopped doing the work and started running the company.
None of that is a press release. Each one is a small, unstated edit to the truth about your company, paired with zero edits to the information AI reads. Do that fifteen times over five years and the business a prospect meets inside an AI answer is not the business you run.
The same lag shows up sharply after a pivot, which I covered in Can AI Tell When a Business Has Changed Direction?. Here there is simply no direction change to blame it on.
Your Entity Record Is Being Edited by People Who Do Not Work for You
Here is the half of drift that rarely occurs to a business owner. The gap does not widen only because you changed. It widens because the record changes too, and other people are the ones changing it.
AI builds its picture of your business from information across the web, and most of that information lives on properties you do not own. Those properties are not frozen. A partner redesigns their site and your name comes off the client list. A directory gets acquired and rebuilds its listings from a cheaper data source with your old category attached. A reviewer describes your work in their own vocabulary rather than yours, and that vocabulary starts accumulating.
Your entity file is a living document with dozens of authors, no editor, and no version control. You are one contributor among many, and your contributions stop at your own domain, the boundary set out in Schema Stops at the Edge of Your Website. Everything past that edge keeps getting rewritten by strangers, at a pace nobody is tracking.
The Models Move Too
A third clock is running, and it belongs to the AI systems themselves.
Models get retrained, and the picture a model holds of your business jumps to whatever the web said at a new point in time, which may be more current than before or, if the useful sources went offline meanwhile, less. Systems that check the live web pull a different set of pages this month than last, because search results churn constantly.
The consequence is worth stating plainly: the answer to the same question about your business can change without a single fact about your business changing. It can improve on its own. It can also degrade on its own. And because the three major systems train on different material on different schedules, they drift at different speeds and in different directions, which is the underlying reason one model’s answer tells you little about the others.
That is also what makes a check you ran a year ago close to worthless today. You tested a system that no longer exists, using a picture that has since been rebuilt. I address the broader version of that in What If AI Changes Tomorrow?
Drift Has a Direction, and It Points at Generic
Drift is not random noise. If it were, it would sometimes wander toward accuracy and you could ignore it. It pulls consistently toward two places.
The first is your past. AI weighs evidence by volume and repetition rather than by date, so old information holds its position long after you have moved on. Every year you operate without refreshing the record, the older version of you keeps its majority.
The second is the average of your category. When a specific attribute weakens, because the pages that stated it went stale or vanished, the space does not stay empty. The model fills it with what is typical of companies like yours. Your distinctive pricing structure becomes standard pricing. Your unusual specialty rounds off to the nearest familiar category. That is the slow road to the problem in Why AI May Recognize Your Business but Misunderstand Its Specialty.
Losing your specifics costs more than losing a fact. Specifics are what get you recommended when a prospect asks for someone who does exactly what you do. A generic description is not an insult. It is a disqualification, delivered politely.
Why Nothing Ever Triggers a Check
Events prompt checks. An ordinary Tuesday does not.
When a business renames itself, somebody thinks to look. When it moves, the address gets verified. Drift never crosses a threshold and never produces a day that feels different from the one before it. The answer was slightly less accurate this month than last, and no step in that sequence is large enough to notice.
The delivery never degrades either. A description built from five year old evidence arrives in the same assured tone as one built from last week’s, because AI sounds certain whether or not its understanding is current. There is no wobble in the voice to tip anyone off, least of all the prospect reading it.
What Drift Costs While It Runs
A sharp failure produces a complaint. Drift produces a slightly lower conversion rate on inquiries that never happen. Your prospect may ask AI about you before contacting you, and as the description gets vaguer and more dated, a slightly larger share of those prospects decide you are not the fit.
None of that registers as a problem. It registers as a soft market, a slow quarter, a competitor getting more aggressive. Businesses reorganize their entire sales process over losses caused by a description they have never read. That is the real expense of drift: not the wrong answer, but the years of wrong diagnosis it invites.
What Actually Slows It Down
Drift cannot be stopped. It can be kept small, and the difference between a business with small drift and one with large drift is not effort. It is cadence.
The foundation is a written record of the few facts that identify you, which I laid out in Building Entity Consistency: your exact name, your category, your location, your people, and the plain names of your main services. What that post does not address is the calendar. A record built once and never revisited starts drifting the day it is finished.
So the working version is a review on a schedule. Once or twice a year, check the record against the business as it exists now, then check the sources against the record. Use the same terminology every time, since new language introduced casually is one of the quieter drift engines. Treat it like insurance renewals or equipment maintenance: unglamorous, scheduled, and far cheaper than the alternative.
The part you cannot do blind is knowing where the drift landed. Interpretations can be corrected, but not by guessing at what needs correcting, and a business that rewrites its homepage when the problem sits in a directory listing has spent real effort on nothing.
Finding Out How Far Yours Has Drifted
Drift is only visible as a difference from a known point, and most businesses have never established one. You cannot tell whether AI’s picture of you has slipped if you have never read it.
That is what the AI Business Understanding Report gives you. It documents what ChatGPT, Claude, and Gemini say about your business right now: which version of your services they lead with, how current their sense of your scale and specialty is, where the three disagree, and which details have rounded off toward the generic. It is a dated measurement, deliberately, because that is the only useful thing to hold against a system that keeps moving. The first one tells you where you stand, and becomes the baseline that makes every later reading mean something.
Your business has been changing in small ways for years, and so has everything on the web that describes it. Whether those two sets of changes stayed in step is a question with a real answer. Ordering a report is how you find out how far apart they drifted while nobody was looking.