How Long Do AI Companies Keep Your Chats? (2026 Sourced)

Longer than most people assume, and most companies refuse to name a number. Google says human-reviewed Gemini conversations can be kept up to three years and are not deleted when you delete your activity. Chai's own notice caps retention at the life of your account plus five years after you delete it. A 2025 US court order forced OpenAI to preserve even deleted ChatGPT conversations for roughly four months. The uncomfortable pattern across the industry is that "delete" rarely means gone, and the services that keep the least are usually the ones that charge you money instead of monetizing the data.

Below is what each major service actually says, drawn from company documentation and court filings read in July 2026, with the gaps marked rather than filled with a guess. Retention policies change, so treat any figure that matters to your decision as something to re-check at the source.

How long each AI company keeps your chats

Two things stand out from that table. First, the companies with the longest, clearest retention windows are the free, mainstream assistants, the exact tools most people reach for with the most personal questions. Second, most of them will not give you a single number, which is itself an answer.

Why "as long as necessary" is not a retention period

The phrase you will see most often in these policies is some version of "we keep your data as long as reasonably necessary." That is not a time. It is a reservation of the right to decide later, and it can mean thirty days or the rest of the company's existence depending on what turns out to be convenient. A company that intends to hold data briefly tells you how briefly, because a short number is a selling point. A company that writes only "as long as necessary" has chosen not to commit, and you should read that choice for what it is.

Contrast that with the two services above that do state a number: Google says up to three years for human-reviewed Gemini chats, and Chai says account life plus five years. Those are unflattering figures, but they are honest, and an honest bad number is more useful to you than a comfortable vague phrase. When you are comparing services, weight the ones that quantify their retention over the ones that dodge it, even when the quantified number is worse.

Does deleting a chat actually delete it?

Not always, and the details matter. Deleting a conversation usually removes it from your visible history right away, but three things can keep a copy alive behind the scenes: backups that roll off on their own schedule, chats already pulled for human review or safety systems, and legal holds. Google states plainly that Gemini chats selected for review are not removed when you delete your activity. And the ChatGPT case is the clearest warning of all, because a court, not the company, kept deleted conversations alive for months.

Then there is training, which no delete button reaches at all. Once a conversation has contributed to a model's weights, deleting the original chat does nothing to unwind it, and no company retrains a production model because one user changed their mind. This is why training is the retention question that matters most, and we walk through the full nuance in a separate piece on whether deleting AI chats actually deletes them.

Which AI keeps your chats the longest?

Of the services with a published figure, Chai's five-years-past-deletion is the longest hard number, and Google's three-year window for human-reviewed Gemini chats is the longest for a mainstream assistant. But "longest published" is not the same as "longest kept," because the companies that publish no number could be keeping data indefinitely under an "as long as necessary" clause. The safe assumption for any free service that will not commit to a period is that your conversations persist until it decides otherwise.

The pattern that actually predicts short retention is the business model. A subscription product has no advertising engine to feed and no data-licensing upside, so keeping your old conversations is pure cost and liability rather than an asset. That is the structural reason paid, privacy-focused services tend to retain less and delete for real, and it is worth more than any single promise, because it aligns the incentive instead of asking you to trust one. Our comparison of the best private AI chatbot options ranks the major services on exactly this, using each company's own policy language.

How to find any service's real retention in five minutes

Open the privacy policy and search it for four words: retain, delete, train, and improve. Look for a number next to retention; if there is none, treat "as long as necessary" as indefinite. Check whether training is on by default and whether the opt-out is available in your country, because several of the biggest opt-outs are regional. Then ask who pays for the compute, because a free product is funding those tokens somewhere, and your retained conversation is often part of the answer.

If a policy will not load, or a company publishes no retention period and no clear pricing, that absence is information rather than a neutral blank. It means you are relying on marketing instead of a commitment you could hold them to. For conversations you would not want read back to you, that is the wrong thing to rely on, and it is why a genuinely private AI chat states its data terms in specific words instead of vibes.

Where your data goes after the chat

Retention is only half of the exposure. The other half is what the company does with the copy it keeps: whether it trains on it, whether it shares it with vendors, and whether any of it flows into advertising profiles or data-broker databases that outlive the chat app entirely. A conversation that a chatbot deletes in thirty days can still have seeded an ad profile or a broker record that persists for years. If that broader footprint is what worries you, having your details pulled from data-broker listings is a separate cleanup worth running alongside tightening up which chatbot you use, because chat retention and data-broker exposure are two different leaks.

The short version

Most AI companies keep your chats longer than you would guess and refuse to name a period. Google keeps human-reviewed Gemini chats up to three years past deletion, Chai keeps data up to five years past account deletion, and a court kept deleted ChatGPT chats alive for four months in 2025. The services that keep the least are the paid ones with nothing to gain from your conversation. When you compare, trust the number a company will commit to over the vague phrase it hides behind, and remember that training is the one form of retention no delete button ever reverses.