After a rebrand, AI keeps using your old name because the old name lives in the model's trained memory, where it sounds confident and settled, while the new name exists only in the live retrieval layer, where it sounds tentative. You cannot wait this out. The fix is to make the new name the best-connected entity AI can retrieve, and to bridge the old name to it everywhere a model looks: Wikidata, schema, redirects, the knowledge panel, and earned media.
A rebrand is the one moment when your hard-won AI visibility works against you. Every citation, every Wikipedia sentence, every "best tools for X" listicle that named the old company is still out there, still being retrieved, still teaching models that the old name is the real one. The new name has none of that history yet. So the model does the rational thing: it answers with the entity it knows best, which is the one you are trying to retire.
Why does AI keep using my old name?
Because AI models carry two kinds of memory, and your rebrand only touches one of them. The first is parametric memory: everything baked into the model's weights during training, up to its knowledge cutoff. The second is the retrieval layer: the live pages a model fetches at answer time. Your old name sits in both. Your new name, immediately after launch, sits only in retrieval.
This split is not cosmetic. As Duane Forrester, a former Bing and Microsoft search lead, puts it: "Your brand's foundational narrative, if it exists clearly in parametric memory, presents with the confidence of internalized knowledge. Your recent product news, if it only exists in the retrieval layer, arrives with the hedging language of external evidence." A model that learned your old name during training states it plainly. The new name shows up wrapped in "according to recent sources." Parse tracks AI visibility across ChatGPT, Google AI Overviews, and Perplexity, covering 4.6 million analyzed AI responses and more than 56 million citation observations across 581,000+ brands, and renamed brands are a recurring pattern: the answer layer keeps splitting their authority between two entities for months.
How long does the lag actually last?
Longer than a domain redirect, and unevenly across platforms. The reason is the knowledge cutoff. Flagship models in 2026 carry cutoffs that trail their release by months: GPT-5-class models cut off around August 2025, the still-widely-deployed GPT-4o at October 2023, and Gemini 3 at a January 2025 parametric cutoff (Otterly, 2026). A name you changed after those dates simply does not exist in the model's trained memory, and historically the gap between a model's cutoff and its public release has run 6 to 18 months. So a brand that rebranded in early 2026 may be current in one model, unknown in another, and confidently wrong in a third.
Typical historical gap between a model's training cutoff and its public release.
Knowledge cutoff for GPT-5-class models; names changed after it are absent from parametric memory.
Time for ChatGPT's base training to absorb a name change, even with strong retrieval signals.
How long Google says to keep 301 redirects live after a domain move.
The practical reading: the retrieval layer is the only part you can move quickly, so the entire playbook is about making the new name win at retrieval while the training data catches up on its own schedule.
Start with a rebrand visibility audit
Before changing anything, measure how deep the old name runs. Guessing wastes the limited leverage you have.
- Run 25 to 30 prompts across ChatGPT, Perplexity, and Google AI Overviews that ask about your category, your product, and your brand by name, using both the old and new names.
- Score each answer on three things: which name the model leads with, whether it knows the two names refer to the same company, and which sources it cites.
- Note the split. A model that says "X, formerly Twitter" has already bridged the entities. A model that treats them as two separate companies, or never mentions the new name, has not.
- Pull the cited URLs into a list. These are the exact pages teaching models the old name, and the ones you will need to update or get updated.
If most answers still lead with the old name or fail to connect the two, you do not have a content problem. You have an entity continuity problem, which is what the rest of this playbook fixes. Our entity disambiguation guide covers the adjacent case where two unrelated brands share a name.
Anchor the new name in Wikidata and Wikipedia
The single highest-leverage move is to make the open knowledge graph treat the rename as a continuation, not a new entity. Wikidata is where to do it, because it feeds Wikipedia infoboxes, Google's Knowledge Graph, and the entity context several AI products pull at answer time.
Do not create a second item for the new name. Update the existing item: change the primary label to the new name, demote the old name to an alias, and keep both discoverable. This is exactly how the platform formerly called Twitter is modeled. Its Wikidata item now carries the label "X" with aliases including "Twitter," "Twitter/X," and "X (Twitter)," so a system retrieving either string resolves to one entity. If you qualify for a Wikipedia article, make sure the rename is stated in the first sentence with the "formerly known as" construction and that a redirect points the old title to the new one. See our Wikidata playbook for the statement set that makes an item machine-legible.
If you want to see which sources shape AI answers about your brand, run a free brand check — it takes a minute.
Make your schema say "same company, new name"
Structured data is where you assert continuity in your own voice, and most teams ship the wrong version of it after a rebrand by simply swapping the name and deleting every trace of the old one. Keep both, with the right roles.
In your Organization schema, set name to the new brand name and add the old name as an alternateName. schema.org defines alternateName as "an alias for the item" and legalName as "the official name of the organization, e.g. the registered company name," so use legalName if your registered entity changed too. Most important is sameAs, which schema.org defines as a "URL of a reference Web page that unambiguously indicates the item's identity." Point it at your updated Wikidata item, your Wikipedia article, your LinkedIn page, and your verified profiles.
Google's Organization structured data guidance reinforces this: it asks for the same name and alternateName you use for your site name, describes alternateName as "another common name that your organization goes by," and treats sameAs as the link between your site and the profiles that confirm who you are. That reciprocal web of links is what lets a retrieval system collapse old and new into one brand.
Keep the old name as a bridge, not a rival
The instinct after a rebrand is to scrub the old name everywhere. For AI visibility, that is the wrong move, because the old name is the bridge that carries your accumulated authority to the new one.
Every page that says "NewName (formerly OldName)" is a training and retrieval signal that fuses the two entities. Delete the old name and you strand the citations, reviews, and listicle mentions that still point at it, leaving the model with a well-documented old entity and a thinly documented new one. Retire the old name from your marketing voice, but preserve it as a documented alias in schema, Wikidata, your About page, and your press materials until the answer layer has fully migrated.
The same logic applies to your domain. If the rebrand moved your URL, follow Google's site move guidance: use permanent 301 or 308 redirects, which Google says "don't cause a loss in PageRank," submit the new sitemap in Search Console, and keep the redirects live "for as long as possible, generally at least 1 year." A medium site takes a few weeks to reindex; the AI retrieval layer trails that, so the redirects are load-bearing well beyond the SEO migration.
Update the knowledge panel and platform records
The Google Knowledge Panel is one of the upstream entity sources that flows into AI answers, so a stale panel keeps feeding the old name downstream. If you have not already, claim and verify your panel, then use the correction flow. Google's help center explains that once you "verify your identity," you can search your entity, locate the panel, and click "Suggest edits," and that verified feedback is prioritized and reviewed "within a few days," though it "can sometimes take more time." Note the limit: Google "doesn't manually create or delete Knowledge Panels," so you are nudging an existing record, not forcing a new one.
Do the same housekeeping on every profile a model might retrieve: LinkedIn, Crunchbase, G2, Capterra, your app-store listings, and your social handles. Each one should show the new name primary with the old name acknowledged. Inconsistent records are how a brand ends up split across two entities for a year, which is the same failure mode behind a sudden drop in AI recommendations.
Earn third-party coverage that ties old to new
Your own pages assert the rename; third-party coverage is what makes models believe it. Because roughly four in five brand mentions in AI answers come from sources other than the brand's own site, the rename has to appear in retrievable, independent text to move the answer layer.
Prioritize coverage that explicitly connects the two names in one sentence: "NewName, the company formerly known as OldName." Trade press, analyst notes, podcast show notes, and conference listings all work, and the "formerly" phrasing is the exact bridge a retrieval system needs to co-locate the entities. This is precisely how the largest rebrands stayed legible: when Facebook, Inc. became Meta Platforms on October 28, 2021, and when Twitter announced its shift to X on July 22, 2023, every article that paired the names taught the graph the mapping. You will not get that volume, but you need the same shape. For the mechanics of how retrieved passages become recommendations, see how AI retrieval works.
What to expect by platform, and when
Set expectations by retrieval model, because the platforms migrate on very different clocks. Perplexity reflects a clean rename within days, because it retrieves live and weights recent sources heavily. Google AI Overviews typically follows within two to six weeks, once the Knowledge Graph absorbs the updated signals. ChatGPT is the slowest: its base training can take three to six months to internalize a name change, and until then it leans on browsing and indexed retrieval, so your retrieval-time signals do the work.
The honest framing for leadership: a rebrand is a known, temporary visibility tax, not a failure. If your audit at week one shows the old name dominating, that is expected. The right metric is the trend across monthly re-runs of the same prompt set, not any single answer. A brand that is methodical about Wikidata, schema, redirects, the knowledge panel, and "formerly" coverage usually sees Perplexity and AI Overviews flip first, with ChatGPT confirming last. If the old name is also attached to outright wrong facts, pair this with the brand correction playbook.
Who needs this playbook, and who can wait
This is urgent if your rebrand changed the name buyers type into an AI model, if it moved your domain, or if the old name still carries the bulk of your category citations. In those cases the split entity is actively costing you recommendations, and every week of inconsistent records deepens it. Run the audit now and work the entity, schema, and earned-media steps in parallel rather than in sequence.
You can move more slowly if the change was cosmetic: a new logo, a tagline, a visual refresh, or a legal-entity tweak that does not change the name customers say out loud. Those rarely confuse retrieval. The test is simple. Ask three AI models about your category and watch which name they lead with. If it is the old one and that name no longer routes to you, you are paying the rebrand tax, and this playbook is how you stop.
Frequently asked questions
How long until AI uses my new brand name?
It depends on the platform. Perplexity usually reflects a clean rename within days because it retrieves live. Google AI Overviews tends to follow within two to six weeks as the Knowledge Graph updates. ChatGPT's base training can take three to six months, so until then your retrieval-time signals, Wikidata, schema, redirects, and "formerly" coverage, are what carry the new name into answers.
Should I delete every mention of my old brand name?
No. For AI visibility the old name is the bridge that transfers your accumulated authority to the new one. Keep it as a documented alias in Wikidata, schema alternateName, your About page, and press materials so models fuse the two entities. Retire the old name from your active marketing voice, but do not erase it from the record until the answer layer has migrated.
Does updating Wikidata really change AI answers?
It is the highest-leverage single move, because Wikidata feeds Wikipedia infoboxes, Google's Knowledge Graph, and the entity context several AI products retrieve. Update the existing item rather than creating a new one: set the new name as the label and demote the old name to an alias, the way the platform formerly called Twitter is now modeled under one item with both names.
My domain changed in the rebrand. What matters most for AI?
Permanent 301 or 308 redirects, kept live for at least a year per Google's site-move guidance, so the citations pointing at your old URLs still resolve. The AI retrieval layer trails search reindexing, so redirects matter longer than a standard SEO migration. Pair them with a sitemap submission and consistent name records across every third-party profile.