You got the news article de-indexed from Google. Then someone asked ChatGPT about you and the arrest came up anyway. This is the AI resurrection problem, and it is becoming one of the most common complaints in online reputation management. Here is how it happens, why each AI platform behaves differently, and how to systematically remove an arrest record from AI search results.
AI tools have longer memories than Google. ChatGPT, Perplexity, and Bing Copilot draw from training data that may predate your suppression efforts by years. A de-indexed article can still live inside an AI model.
Source removal must come first. If the original article still exists at its URL, every AI platform will continue to reference it. Removal or de-indexing is the foundation of any AI cleanup strategy.
Each AI platform has a different privacy request process. OpenAI, Google, Microsoft, and Meta all have distinct privacy portals with different requirements, timelines, and success rates.
State expungement laws do not automatically bind AI companies. California AB 1985 and similar statutes apply to government agencies and certain data brokers, but AI platforms are not currently required to honor expungement orders automatically.
Full clearance takes time. A complete arrest record AI search removal across all platforms typically requires three to six months when pursued systematically.
The Core Problem
For years, the standard playbook for managing an old arrest record online focused on two things: getting the news article removed or de-indexed, and pushing it down in Google rankings with positive content. That strategy still works, but it no longer covers the full landscape.
AI search tools, including ChatGPT, Perplexity, Google AI Overviews, and Bing Copilot, do not operate like traditional search engines. Traditional search engines crawl the current web and serve links to content that exists right now. AI language models are trained on snapshots of the internet, often collected over years, then deployed and used for months or years afterward. This means the model's knowledge of you is frozen in time, at a point when your arrest record articles were still live, widely indexed, and prominently ranking.
The practical result: a person might successfully suppress an arrest article in Google's search results in 2024, only to find that ChatGPT, when asked about them by a hiring manager or business partner, produces a confident summary of the arrest from 2019. The article is no longer on page one of Google. But it lives on inside the model.
This phenomenon has been called the "AI resurrection problem" within the reputation management industry. Content that was effectively buried can reappear through AI-generated summaries, and because users often perceive AI responses as more authoritative than a list of links, the reputational damage can be significant.
The scale of the training data problem is often underestimated. Major AI language models were trained on web crawls that go back to 2008 or earlier. That means an arrest article from 2010, 2014, or 2018 that has since been removed from the publisher's site or de-indexed from Google may still exist in the training corpus. The article's deletion from the live web does not retroactively remove it from a model that was trained before the deletion occurred. This is not a bug in these systems, but it is a serious privacy consequence that platforms are only beginning to address.
The risk is compounded by how people use AI tools. A hiring manager who types "tell me about [your name]" into ChatGPT is getting a synthesized answer with no source links to evaluate. There is no opportunity to assess whether the source is a decade-old local news brief or a major investigative piece. The AI presents information with equal confidence regardless of the origin or the current accuracy of the data.
Platform Breakdown
Not all AI platforms work the same way, and understanding the difference matters for your removal strategy. There are two fundamentally different architectures: models that primarily rely on static training data, and retrieval-augmented generation (RAG) systems that query the live web before generating a response.
ChatGPT's standard models are primarily trained on static datasets. When you query ChatGPT about a person, the response draws from the model's training data rather than conducting a live web search (unless you are using the Browse with Bing feature or GPT-4's web access tools). This means that even a fully removed article may inform the model's response if the article was part of the training corpus.
OpenAI provides a privacy request portal for individuals who want personal information removed from responses. The process involves submitting a request at privacy.openai.com, documenting the specific information that appears, and explaining why it should be removed. OpenAI reviews requests under its privacy policy, and where feasible, can retrain or fine-tune models to reduce the likelihood of surfacing specific information.
Perplexity operates primarily as a retrieval-augmented system, meaning it searches the live web, retrieves sources, and then synthesizes a response with citations. This is both better and worse news for people with old arrest records. Better, because if the original article is removed or de-indexed, Perplexity is less likely to find it. Worse, because Perplexity is very effective at finding content that Google has de-ranked but not de-indexed, including archived versions and cached pages.
Google AI Overviews use Google's own search index as their data source, combined with Google's language models. This means that content de-indexed from Google is less likely to appear in AI Overviews, making Google de-indexing requests more effective for AI purposes than for other platforms. Google provides a de-indexing request process through its Search Console and its content removal request form, and a successful de-indexing typically cascades to AI Overviews relatively quickly.
Bing Copilot (formerly Bing Chat) uses a combination of Microsoft's language models and live Bing search results. Like Perplexity, it conducts real-time web retrieval before generating responses. Content de-indexed from Bing has a meaningful chance of not appearing in Copilot responses. Microsoft's content removal request process runs through Bing Webmaster Tools and its content removal request form.
Even after a source article is de-indexed from both Google and Bing, it may still appear in AI responses for weeks or months because search engine index updates and AI model updates operate on different schedules. Do not assume that de-indexing equals immediate removal from all AI platforms. Submit direct privacy requests to each AI platform separately.
Have an arrest article surfacing in AI search results? Our team has developed a platform-by-platform removal strategy that addresses both the source content and the AI layer simultaneously.
Get a Free ConsultationComparison Data
| AI Platform | Primary Data Source | Removal Option | Likelihood of Success | Typical Timeline |
|---|---|---|---|---|
| ChatGPT (OpenAI) | Static training data + optional web browsing | OpenAI Privacy Portal (privacy.openai.com) | Moderate | 8 to 16 weeks |
| Perplexity AI | Live web retrieval (RAG) | Source de-indexing is primary lever; Perplexity content request form | Higher if source removed | 4 to 10 weeks post source removal |
| Google AI Overviews | Google Search index | Google de-indexing request; Google Privacy Removal Request | Higher | 4 to 8 weeks |
| Bing Copilot | Live Bing search results + Microsoft LLM | Bing content removal request; Microsoft Privacy Dashboard | Moderate | 6 to 12 weeks |
| Meta AI | Static training data; Meta platform content | Meta Privacy Center; Right to erasure request (for EU/UK users) | Lower (limited non-EU process) | 8 to 20 weeks |
The Removal Process
Effective arrest record AI search removal requires working in a specific sequence. Skipping the source removal step and going directly to AI platform requests is a common and costly mistake. Here is the correct order of operations.
We have processed hundreds of arrest record AI search removal requests since AI overview tools became mainstream. The single biggest factor in success is timing between the source removal and the AI platform request. Submitting to OpenAI or Microsoft before the source article is fully removed typically results in rejection, because reviewers find the article still accessible. Wait until you have written confirmation that the original article is down or de-indexed, then submit your AI platform requests within the same week.
Legal Landscape
Many people assume that receiving a court-ordered expungement means their arrest record automatically disappears from the internet, including from AI tools. This assumption is incorrect in nearly every U.S. jurisdiction, and understanding why matters for setting realistic expectations.
Expungement orders are directed at government agencies, not private publishers or technology companies. When a court grants expungement, it orders law enforcement agencies, courts, and certain government databases to seal or destroy the arrest record. It does not create a legal obligation for newspapers, AI platforms, or websites to remove their reporting about the arrest.
However, expungement documentation is extremely useful as leverage in removal requests. When you approach a publisher or an AI platform with documented proof that a court has ordered your arrest record sealed, you significantly strengthen your argument that the continued publication of the information is harmful, outdated, and no longer serves a legitimate public interest.
California's AB 1985, which took effect in 2023, expanded the state's existing data broker law to require data brokers to honor deletion requests for certain types of personal information. While the law does not directly cover news publishers or AI companies as currently defined, it reflects a broader legislative trend toward recognizing individuals' rights to control personal criminal history data in digital form. Advocates have argued that AI platforms that train on and reproduce arrest record data from California residents may eventually fall under this regulatory umbrella as the law evolves.
Colorado, Illinois, and several other states have enacted or are considering legislation that expands expungement protections or creates data deletion rights applicable to private entities. The Federal Clean Slate Act, introduced in Congress in multiple sessions, would automate federal expungement for certain offenses and could eventually create stronger grounds for online removal requests. For the most current state-by-state analysis, the Collateral Consequences Resource Center maintains updated research on expungement law developments.
Even without a specific statute compelling removal, expungement documentation combined with a well-reasoned editorial argument remains one of the most effective tools for negotiating article removal. Our guide on removing an old arrest article from Google covers how to frame this argument for different types of publishers.
Platform Policies
Understanding what each AI company's privacy team actually reviews and acts on is essential for writing an effective removal request. Submitting a vague request citing "my privacy" is far less effective than a specific, documented request tied to each platform's stated policies.
OpenAI's privacy policy acknowledges the right to request deletion of personal information in jurisdictions where applicable, including California under CCPA and the EU under GDPR. Even if you are not in an explicitly covered jurisdiction, OpenAI's privacy team reviews requests on a case-by-case basis. The strongest requests are those that (a) identify specific outputs containing your arrest information, (b) demonstrate that the underlying source article has been removed or corrected, and (c) show that the information is sensitive personal data whose continued surfacing causes documented harm, such as employment consequences.
Google provides a specific removal request pathway for content that involves personal information like criminal records. Under Google's policies, you can request de-indexing of content that includes your name in combination with arrest or booking record information, particularly where the arrest did not result in a conviction or where the content is outdated. Google evaluates these requests by balancing the individual's privacy interest against the public interest in the information. Arrest records that are old, resulted in dismissal, or involved minor offenses generally fare better in this analysis. See our companion guide on removing content from ChatGPT and AI search for more on the platform-specific process.
Microsoft's content removal process for Bing and Copilot focuses on content that violates its content policies or involves sensitive personal information. Arrest record data, particularly where the arrest did not result in conviction, can qualify. Microsoft's Privacy Dashboard at account.microsoft.com/privacy provides the starting point for personal data requests.
Ready to start the removal process? Our team handles the full sequence: source article, Google de-indexing, and AI platform requests across every major platform.
Start with a Free ReviewFrequently Asked Questions
We have helped 1,000+ people clear old arrest records from search results and AI tools. The process is methodical, confidential, and results-based.
Confidential consultation. No obligation. Pay only for results.