Short answer, as of September 2026: a rebrand breaks AI visibility in a way redirects cannot fix, because assistants identify you by entity and by third-party mention, not by URL. Expect the old name to persist in AI answers for six to twelve months after the switch, and plan the work that shortens that rather than assuming a 301 handles it.
Teams generally arrive at this problem a few weeks after launch. The site moved cleanly, rankings recovered, and then someone asks an assistant about the company and gets a confident answer using the old name — or worse, an answer that treats the old and new names as two separate businesses.
That is not a bug in the redirect. It is a different system failing, and it needs different work.
Why a rebrand is not a migration
A site migration is a URL problem, and the web has a mature solution for it. Redirects pass signals, search engines reprocess, and within weeks the new addresses inherit the old ones' standing. The site migration playbook covers that case and it mostly works.
A rebrand is an identity problem, and there is no 301 for identity. An assistant answering a question about your company draws on three things, and a redirect only touches the first.
Your own site, which you control and can change on launch day.
Third-party mentions — directories, review platforms, news coverage, partner sites, forum threads, conference listings — which still carry the old name and which you can only change by asking, one at a time.
Training data, which is a snapshot of the web from before your rebrand and cannot be edited at all. It ages out. That is the entire remedy.
This is why the lag is measured in months. Engines using live retrieval will pick up a well-executed rebrand relatively quickly, though they differ widely in how fast they refresh. Engines leaning on parametric knowledge will keep producing the old name until the next training cycle, and you have no lever on that schedule.
The four states an assistant can be in
Diagnosing where you are matters, because the fixes differ.
| State | What the assistant says | Root cause | What fixes it |
|---|---|---|---|
| Old name only | Answers entirely under the former name | Nothing has propagated yet | Time, plus third-party updates |
| Split entity | Treats old and new as two companies | No published link between the names | An explicit "formerly known as" statement |
| Merged but stale | New name, old facts about services or ownership | Retrieval found the new site, cached facts remain | Correct the stale facts at source |
| Clean | New name, current facts, old name as history | Working as intended | Maintain |
The dangerous one is the second. A split entity halves your authority: some citations accrue to a company that no longer exists and some to one that looks brand new with no history. Left alone it can persist much longer than the simple lag, because nothing in the data tells the model the two are the same organisation.
Before you launch
Three things are much cheaper to do before the change than after.
Baseline your current AI visibility. Run your prompt set through a tracker and record who is named, what they are called and which sources are quoted. Without this you cannot tell, three months later, whether a drop is the rebrand or something else entirely — and you will be asked. If you have never built one, how to build a prompt set worth tracking covers the mechanics.
Inventory every place your name appears off-site. Directories, review platforms, industry bodies, partner pages, software marketplaces, podcast and conference listings, your own social profiles. This list is always longer than expected and it is the actual work of a rebrand.
Write the transition statement before launch day. One paragraph, on a stable URL, stating plainly that the old name is now the new name, when it changed, and that services and entity continue unchanged. This single paragraph is the strongest available tool against the split-entity problem, because it is the fact an assistant needs in order to merge the two.
On launch, and for the month after
Publish the "formerly known as" statement and leave it up. Put it on the about or homepage, keep the old name in the prose, and do not remove it after a quarter because it feels like clutter. It is load-bearing for two or three years. Phrase it in complete, self-contained sentences — "Acme Field Services, formerly known as Bob's Plumbing Supply, changed its name in September 2026 and continues to operate the same business under the same ownership" — because a passage like that can be quoted whole and still make sense.
Update your structured data. Organization schema supports alternateName, and it is the cleanest machine-readable way to state the relationship. Keep sameAs pointing at the profiles you have actually updated.
Update the high-authority third-party sources first. Rank the inventory by how often each source appears as a citation in your tracked answers, not by how easy it is to edit. Your Google Business Profile, your largest review platform and the two directories assistants keep quoting are worth more than forty small listings — and getting the high-value ones consistent is the same entity-consistency work described in entity consistency: the unglamorous work that decides whether AI engines trust you.
Do not delete the old content. Pages under the old name that earned citations are the bridge. Redirect them, keep the historical references in the text, and let the old name appear in context rather than scrubbing it. Teams that purge every mention of the previous brand consistently make the split-entity problem worse.
What to do about answers that are simply wrong
Some assistants will produce confidently wrong statements during the transition — that the old company was acquired, that it closed, that the new name is a different business.
There is no correction form and no support queue for this. What works is the slower route: make the correct fact abundant and easy to find at the sources the engine is actually reading. If the wrong answer cites a specific directory, fix that directory. If it cites nothing, the fact is likely coming from training data and will persist until the next cycle, however loudly you publish.
Resist the temptation to publish a dozen pages restating the correction. One clear, well-placed statement plus corrected third-party sources outperforms volume, and a site that repeats the same paragraph everywhere looks like the kind of content engines discount. The general approach to correcting a wrong AI answer about your business is covered in when AI gets your business wrong.
What the recovery curve looks like
Rough expectations, assuming the work above is done properly.
Weeks one to four: your own site is consistent, retrieval-heavy engines begin producing the new name, parametric answers still say the old one. Split-entity answers are common and normal in this window.
Months two and three: high-authority third-party sources are updated and citations start attaching to the new name. This is the point where a baseline earns its keep, because the numbers usually look worse before they look better and someone will want to panic.
Months four to six: most live-retrieval answers are clean. Long-tail mentions on small sites still carry the old name; this is cosmetic and not worth chasing exhaustively.
Months six to twelve: training refreshes begin removing the last of it. Some references never disappear, which is fine — a company having a former name is a normal fact, not a defect.
Anyone promising a faster curve is describing rankings rather than AI answers.
If you are choosing whether to rebrand at all
Two things worth weighing that are easy to miss.
A name that is a common word or phrase is materially harder to establish as an entity than a distinctive one. If the new name collides with an existing company, a product or a general term, expect assistants to conflate you with it more or less permanently, and expect every prompt about you to need disambiguating context.
And a rebrand that also changes what you sell is two changes at once. You will not be able to tell which one moved the numbers. Where it is possible, separate them by a quarter or two — it makes the measurement legible and the recovery diagnosable.
Frequently asked questions
How long do AI assistants keep using our old company name? Typically six to twelve months. Engines that retrieve live sources pick up a well-executed rebrand within weeks, while answers drawn from training data persist until the next refresh, which you cannot influence. Long-tail mentions on small third-party sites may never fully disappear.
Will 301 redirects fix our AI visibility after a rebrand? No. Redirects solve the URL problem and are necessary, but assistants identify you as an entity assembled from your site, third-party mentions and training data. None of the last two are affected by a redirect, which is why the old name survives a technically clean migration.
What is the single most important thing to publish during a rebrand? A plain "formerly known as" statement on a stable URL, written in self-contained sentences that name both the old and new names, the date of the change and the fact that the business continues unchanged. It is the fact an assistant needs to merge two entities into one, and without it you risk being treated as two separate companies.
Should we delete old content that mentions our previous name? No. Pages under the old name that earned citations are what connects the old identity to the new one. Redirect them, keep historical references in the prose, and let the old name appear in context. Purging every mention makes the split-entity problem worse.
Which third-party listings should we update first? Rank them by how often each source appears as a citation in your tracked AI answers, not by how easy they are to edit. Your Google Business Profile, your main review platform and the two or three directories assistants keep quoting matter far more than a long tail of minor listings.
AI is saying our old company was acquired or shut down. How do we correct that? There is no correction form. Make the accurate fact abundant at the sources the engine is actually citing — if a specific directory is quoted, fix that directory. If nothing is cited, the claim is likely coming from training data and will persist until the next refresh regardless of what you publish.
Should we rebrand and change our services at the same time? Where you can avoid it, no. Two simultaneous changes make it impossible to tell which one moved your visibility, and if the numbers fall you will not know what to correct. Separating them by a quarter makes the result diagnosable.