On July 8, 2026, a damning screenshot began circulating on social media. Microsoft’s Copilot Search had answered a question about the energy giant Shell plc — but it called the company “Royal Dutch Shell,” a name that hasn’t been official since January 2022. The error was posted by an X account named RoyalDutchShell, an ironic reminder that even when a company permanently retires a moniker, AI tools can struggle to keep up.

This wasn’t a beta hiccup. Copilot Search, built on a combination of GPT‑4 and Microsoft’s Bing index, is a mature product at this point, used daily by millions. Yet it presented a defunct corporate identity as if it were current. For anyone who relies on AI‑generated summaries for business or financial decisions, the mistake is a stark warning: the underlying data can be stale, and the algorithms don’t always know the difference.

What Exactly Did Copilot Get Wrong?

The screenshot shared by the RoyalDutchShell account shows Copilot’s response to an unspecified query about the company. Instead of using “Shell plc” — the London‑based multinational’s legal name since a sweeping simplification in early 2022 — the answer repeatedly referenced “Royal Dutch Shell.” The error isn’t a minor typo; it’s a present‑tense use of a brand that no longer exists. For context, Shell’s board approved the name change in November 2021, and it took effect on January 21, 2022. That means Copilot was lagging by more than four and a half years.

Microsoft has not commented officially on the specific incident, but the pattern is familiar. Retrieval‑augmented generation (RAG) systems like Copilot Search fetch information from a live web index and then use a large language model to craft an answer. In theory, that pairing should keep the output current. In practice, the retrieval step can surface legacy web pages that still refer to “Royal Dutch Shell,” especially if those pages are authoritative in other respects or have robust backlink profiles. The model then repeats the outdated name, often without any contextual warning that it might be obsolete.

The RoyalDutchShell X account appears to be a watchdog of sorts, satirically named after the former company identity. It regularly monitors AI tools for name‑related blunders, making this episode part of a broader critique of how language models handle entity updates.

Why This Matters for Users and Businesses

For the everyday Windows user who might ask Copilot “What does Shell do?” or “How big is Shell?” this kind of error is a nuisance. It might not cause direct harm, but it erodes the trust that is essential for AI assistants to be useful. If a tool can’t get a company’s name right four years after a widely publicized rebranding, what confidence should you have in its numbers, dates, or legal summaries?

The stakes rise quickly for business users. Imagine an analyst preparing a market overview, a journalist fact‑checking a story, or a procurement officer vetting a supplier. Citing “Royal Dutch Shell” as a current entity could lead to embarrassing corrections, or worse, suggest the author is working with outdated information. In regulated industries, referencing a defunct corporate name in contracts or compliance documents could introduce legal ambiguity.

Businesses that have undergone rebranding themselves should take note. Even if you update every page on your own site, old mentions persist across the web. A 2015 press release, a third‑party database, or a Wikipedia entry that took weeks to reflect the change can all feed the AI’s index. That means your shiny new name might be drowned out by the old one simply because search engines rank legacy content highly. Brand management now extends to how AI tools represent your company.

For IT administrators and developers who embed Copilot capabilities into internal tools, the lesson is about grounding. If your organization relies on AI‑generated summaries about partners or competitors, consider adding a verification layer—perhaps a fact‑check against a known‑good knowledge base—or at least training users to treat AI output as a starting point, not gospel.

How an AI Can Miss a Major Rebrand

The root of the problem lies in the tension between static training data and dynamic retrieval. Large language models are pre‑trained on a snapshot of the internet that ends at a certain date. Even when that date is recent, the corpus contains countless references to “Royal Dutch Shell” because that was the official name for decades. The model learns that “Royal Dutch Shell” and “Shell” refer to the same company but might not reliably discern which is current.

RAG is supposed to overcome this by pulling fresh data from a search index at query time. However, search engines like Bing optimize for relevance, authority, and user engagement, not for recency per se. A 2018 financial report or a well‑linked corporate profile might still dominate result pages, and the retrieval component may not be tuned to prioritize the most recent official name unless it’s explicitly triggered to do so. Additionally, entity disambiguation—mapping mentions in documents to a canonical record—often relies on knowledge graphs that can themselves be out of sync. If Copilot’s knowledge graph or the Bing entity repository hasn’t been updated to mark “Royal Dutch Shell” as a historical name with a “valid until” date, the system might treat it as a legitimate synonym.

Microsoft has invested heavily in grounding Copilot against proprietary and licensed data to improve accuracy, but this incident shows that the gap between “mostly right” and “consistently current” remains a challenge. Similar errors have plagued other AI tools: Google’s Gemini occasionally resurrects old product names, and Anthropic’s Claude has been caught using deprecated branding. The scale of the web makes perfect freshness a moving target.

Steps You Can Take to Protect Against AI Inaccuracies

If you’re a user:
- Cross‑check critical details. Before acting on an AI‑generated fact about a company, visit the company’s official website or a trusted source like Wikipedia. Look for dates—press releases, annual reports, and the copyright footer often reveal if the information is current.
- Use the feedback mechanism. Copilot Search includes a “thumbs down” button and often a text feedback form. Reporting specific errors helps Microsoft fine‑tune its models and retrieval algorithms. The more concrete the report, the better—include the query and the incorrect name.
- Adjust your expectations. Think of AI search as an advanced autocomplete with a confidence problem. It’s great for summarization and ideation but not yet a substitute for due diligence.

If you run a business that has rebranded or may do so:
- Audit your digital footprint aggressively. Search for your old name across the web, and contact high‑ranking sites that still use it. Submit corrections to Wikipedia, Wikidata, Crunchbase, and industry databases—these are often the canonical sources that knowledge graphs pull from.
- Consider a “name change” SEO campaign. Publish a dedicated press release or blog post that explicitly links the old name to the new one and explains the transition. Encourage news outlets and partners to update their references. The more signals search engines receive about the new name being definitive, the more likely it will surface in AI retrieval.
- Monitor AI tools directly. Regularly query Copilot, Google Gemini, and other assistants with prompts like “What is [old company name]?” If they still respond with outdated information, reach out to the platform’s support or developer relations teams. Some offer direct avenues for requesting entity updates.

For developers and IT pros integrating Copilot:
- Add a fact‑checking layer. If your application pulls Copilot responses about known entities, run a secondary validation against an internal database or a frequently updated source like a Microsoft Graph connector pointed at your CRM.
- Educate end users. Provide clear disclaimers that AI‑generated content may contain inaccuracies, and require human review for any output used in official documents.

What’s Next for AI Search Accuracy

Expect Microsoft to tighten the freshness signals in Copilot Search. The company has been expanding its use of near‑real‑time data via the Bing index and exploring ways to attach “best‑by” dates to entity facts. An incident as visible as the Shell name slip could accelerate those efforts, much like CEO errors forced improvements in financial data handling earlier in 2025.

For the industry, this is another reminder that retrieval‑augmented generation requires an out‑of‑the‑box entity resolution strategy. We’re likely to see dedicated “fact‑checking” modules that compare AI output against structured knowledge graphs updated on daily or hourly cycles. Microsoft, Google, and others are also under pressure from regulators who increasingly expect AI answers to be verifiable. In the European Union, the AI Act’s transparency obligations may soon mandate that AI tools cite their sources and indicate the age of the information they rely on, making stale answers like this one harder to justify.

For now, the best defense is a skeptical user. Copilot Search is a powerful tool, but as the RoyalDutchShell account hilariously proved, it can still be stuck in the past.