Key takeaways
- AI assistants repeat your old name because the public record still has far more proof behind it than the new one. Training data, retrieved pages, directories and backlink anchor text all lag behind your launch.
- Waiting it out is a bad plan. Old-name landing pages, docs and comparison pages can stay cited in ChatGPT for months, while rebrand press fades in about five weeks.
- Start with an audit of what each AI platform says today, then fix owned pages, directories, schema and third-party mentions in that order.
- Do not block AI crawlers on the old domain. They need to see your redirects and your "formerly known as" signals.
- Treat this as continuous monitoring, not a one-time cleanup. Platform retrieval behavior changes overnight.
The problem: you rebranded, the AI did not
Here is a pattern that plays out constantly. New name, new positioning, new domain. Eight months later a sales rep pings the CMO: a prospect said ChatGPT still calls the company by its old name. Everything you control is updated. Everything you do not control is not.
Microsoft fans know this one well. One Reddit thread complains that Microsoft's own main site still said "Azure Cognitive Search" for months after the rename to Azure AI Search. If the company that has the most direct line to the name change takes months to update its own pages, a smaller team has little chance of a clean switch in a week.
The cost is real but mostly invisible. Prospects get confused about whether you are the same company, whether the old product still exists, or who supports it. Sales cycles drag. And you rarely hear about it unless someone mentions it on a call.
Why AI keeps quoting the old name
The simplest way to put it: the old name still has more proof behind it. AI builds its picture of your company from the public record, and the public record does not update all at once. Several layers of inertia stack on top of each other.
Training data is frozen in time
Models answer partly from what they learned during training. If your old name appears in years of articles, reviews, docs and forum threads, and the new name appears in a few months of coverage, the weight tilts toward the old one. This is the layer you cannot edit directly. You can only dilute it over time with more and better new-name evidence.
Retrieved pages still say the old thing
Search-enabled assistants fetch live pages. If those pages are old-name landing pages, docs, comparison posts or help articles, the model repeats them with a citation. Promptwatch Data's content half-life study tracked every page ChatGPT cited on July 6, 2026 for 90 days. After 30 days ChatGPT still cited 78% of landing pages and 76% of docs. After 60 days it was still citing 69% and 67%. Comparison pages held at 72% after 30 days and 63% after 60.
News is the opposite. Half of cited news pages were gone after 36 days. So your rebrand announcement, which is news, fades quickly. The old-name pages that count as landing, docs or comparison content do not.
Entity authority lags
The old name is an established entity with links, mentions and a history. The new one starts from nearly zero. Until search engines and models connect the two as the same entity, they tend to treat them separately.
Third-party anchor text and listings
Backlinks that say the old name, directory profiles, review sites, Crunchbase-style records, app marketplaces and partner pages all keep the old identity alive. You rarely own these, which is why they linger longest.
ChatGPT now searches your domain directly
There is a newer wrinkle. Promptwatch Data found that on August 8, 2026, ChatGPT started using the site: operator at scale. site: queries jumped from 0.37% to 16.8% of fanout queries, roughly 1 in 6, and average fanout queries per response rose from about 1.08 to 1.83. That means ChatGPT often runs searches limited to specific domains. If a lookup on your own domain turns up stale old-name pages, you are feeding the problem yourself.
Can you just wait it out?
No. A hypothetical makes the arithmetic clear. If your old name sits in 10 million historical documents and the new one in 500,000, the ratio favors the old name in every training cycle for years. That ratio is illustrative, not measured data, but the direction is real. Every month of delay means more confused conversations.
One caveat on timelines. Vendor blogs promise recovery that is measurable at month 3, substantial at month 6 and stable at 12 to 18 months. Those are estimates, not measured benchmarks. Plan for a long tail and measure your own curve.
The rebrand recovery playbook
Step 1: Audit what AI says right now
Before touching anything, document the baseline. Run a prompt set across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. Start with new-name questions:
- "Who makes [new product name]?"
- "What are the features of [new product name]?"
- "How is [new company name] different from [competitors]?"
Then the old-name set:
- "Is [old company name] still operating?"
- "What happened to [old company name]?"
- "Who owns [old product name]?"
Record every answer. Note whether the old name comes from the model's memory (no source) or from a live citation with a URL. That distinction decides your next move: memory problems need time and dilution, citation problems can be fixed this week.
Doing this by hand across five platforms and 30 prompts gets tedious fast, and answers vary run to run. Tools like Promptwatch track prompts daily from the real interfaces of the major AI platforms and show which of your pages get cited, so you can see old-name sources instead of guessing.
Other trackers in the same space include [tool:profound], [tool:otterly-ai] and [tool:peec-ai]. For a wider look, the directory at bestgeosoftware.com lists more options.
Step 2: Clean up every property you own
Search your own domain for the old name. Check:
- Homepage, about page, pricing, product pages and footers
- Docs, help center, changelogs and API references
- Blog posts that are still useful (update them, do not delete them)
- Email signatures, support templates, PDF downloads and metadata
- Old-name subdomains and microsites
Good tools for finding stragglers: [tool:screaming-frog-seo-spider] for a full crawl with custom search for the old name, and [tool:google-search-console] to see which old-name pages still earn impressions.
Do not scrub history dishonestly. Where the old name is useful context, say it once, clearly, with a date: "Acme Analytics, formerly known as DataLoop, since March 2026." Be accurate about the event, too. A rename, an acquisition and a product retirement are different things. Do not write "formerly known as" if it implies two separate companies were always one.
One addition most teams skip is a dedicated "formerly known as" page at a stable URL. It catches people typing the old name, gives AI a clean disambiguation page to cite, and should stay short: one paragraph, a link to the new homepage, and schema listing both names.
Step 3: Handle redirects and the old domain properly
A 301 redirect is necessary but it does not finish the job. Redirects move users and link equity. They do not rewrite what a model already learned or what third-party pages say.
The key point: do not block AI crawlers on the old domain. Promptwatch Data's AI crawler traffic report shows OpenAI at 79.8% of verified AI crawler requests in the week of Aug 31 to Sep 6, 2026, down from 94.8% in mid June, with Anthropic, Google, Perplexity and Mistral making up the rest. If a robots.txt or WAF rule blocks those bots on the old domain, they never see the redirect and never learn the connection. Check your rules for every provider.
A page-to-page redirect map beats a blanket redirect to the homepage. Send each old URL to its closest new equivalent.
Step 4: Update every major directory and profile
These are the sources AI leans on when it checks basic facts:
- LinkedIn company page, Crunchbase, Wikipedia or Wikidata if you have entries
- G2, Capterra, Trustpilot and other review platforms
- App stores, marketplaces and partner directories
- Google Business Profile and local listings for multi-location brands
- Social bios and press kits
For review sites, ask for a name change instead of a new profile so reviews and history carry over. For local listings, platforms like [tool:yext] and [tool:synup] can push a name change across many directories at once. [tool:trustpilot] and similar sites usually have a manual rename process.
Step 5: Publish a clear announcement with machine-readable context
Your press release is a news page, so it decays fast, but it still helps seed the connection. Make the relationship explicit in the first sentence: "[New Name], formerly known as [Old Name], today announced..." Include the effective date, what changed for customers and where to get support. Syndicate it where it will be indexed, and add an FAQ on your site that answers the obvious questions.
Monitoring tools such as [tool:meltwater] or [tool:brand24] help you find where the old name still gets used in fresh coverage.
Step 6: Add schema that links both names
On your homepage, about page and formerly-known-as page, use Organization (or Product) markup with the new name as name and the old name in alternateName. Add sameAs links to your official profiles. Use dateModified honestly.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Acme Analytics",
"alternateName": "DataLoop",
"url": "https://acmeanalytics.com",
"sameAs": [
"https://www.linkedin.com/company/acme-analytics",
"https://www.crunchbase.com/organization/acme-analytics"
]
}
Schema does not force a model to change its mind, but it gives retrieval systems and knowledge graphs an explicit statement of identity. [tool:schema-app] and [tool:yoast-seo] can help roll this out at scale.
Step 7: Chase the highest-value third-party mentions
Rank the sites still using your old name by authority and by how often AI cites them. Then reach out to the top ones, politely, with the exact text you want changed and a link to your formerly-known-as page. Prioritize:
- Articles and listicles that AI already cites
- Partner and integration pages
- Analyst, directory and comparison sites
- Reddit threads and YouTube videos that mention the old name
A short manual list beats a mass email blast. Ten corrections on sources AI actually cites matter more than a hundred on pages nobody reads.
Platform-specific notes
| Platform | What drives the old name | What to check first |
|---|---|---|
| ChatGPT | Training memory plus live search with site: lookups | Old-name pages on your domain, directories, citations in answers |
| Perplexity | Almost entirely live retrieval, so cited pages matter most | Which URLs it cites for old-name questions |
| Gemini and AI Overviews | Google index, knowledge graph and entity data | Search Console, Business Profile, schema, Wikidata |
| Claude | Training memory, with search where enabled | Whether the answer cites anything or recites from memory |
Perplexity is usually the fastest to fix, because if it cites a stale page you can usually update that page and see a change within days. You can run the same old-name prompts directly in [tool:perplexity] and [tool:chatgpt] to eyeball results between audits.
What to monitor, and how often
Platform behavior moves. The August 8 site: change happened overnight, which is why a single audit is not enough. Promptwatch's own note on that report is that keyword and fanout mappings built before that date are stale.
A workable cadence:
- Weekly for the first 90 days: the old-name and new-name prompt sets, plus citations that contain the old name
- Monthly after that: broader brand prompts, sentiment and the share of answers using the new name
- Always: crawler access and redirect health
It helps to know whether AI bots are actually reading your redirected and updated pages. Promptwatch's AI crawler logs show IP-verified bot visits and crawl errors, and its Brand Narrative feature extracts the claims AI makes about you, which is handy for spotting "formerly X" statements that are wrong. A free option for a quick snapshot is the free AI visibility report. Competitors like [tool:scrunch] and [tool:athena-hq] cover monitoring but are less focused on crawler-level evidence.
Common mistakes
- Assuming a 301 fixes everything. It fixes navigation, not memory.
- Deleting all old-name content. You lose the context that links the two names.
- Blocking the old domain from AI crawlers.
- Relying on the announcement alone. News decays in about five weeks.
- Leaving old-name subdomains, help docs and microsites live.
- Skipping platform-specific checks. Each assistant fails for a different reason.
- Declaring victory after one good answer. Answers vary run to run.
Frequently asked questions
How long until ChatGPT stops using our old name?
There is no guaranteed date. Retrieval-based errors can clear within weeks once the cited pages change. Memory-based errors depend on model updates and how much new evidence exists. Expect months and measure your own trend.
What about old product names that were renamed inside the company?
Same playbook at smaller scale. Update docs, pricing and comparison pages first, since those stay cited longest, and keep a changelog entry that states the rename with a date.
What if our old name was a generic term?
Then disambiguation matters more. Use full phrases ("DataLoop analytics platform"), schema with alternateName, and consistent descriptors in every profile.
How do we know the work is paying off?
Track the share of audit prompts where the new name appears correctly, the number of citations that still carry the old name, and the old-name queries that reach your formerly-known-as page. If you want help running a recurring audit and fix plan, an SEO and GEO agency such as 1001 SEO Media handles this kind of cleanup alongside content and PR.
A final word
The rebrand is done on your side. It finishes when the pages AI reads, the directories it trusts and the sources it cites all say the same thing. Run the audit this week, fix the long-lived pages first, and keep checking.