OpenAI updated its U.S. privacy policy on February 9, 2026, and rolled out an advertising pilot inside ChatGPT alongside it. The two moves are connected: the policy update exists specifically to spell out how the new ad ecosystem works, what data feeds it, and what it doesn’t touch.
For most people, the relevant question isn’t abstract. It’s: I’ve used ChatGPT to summarize a document, describe an old photo, or help organize something personal — does any of that now feed an ad business? The honest answer requires separating what OpenAI has stated from what’s still genuinely unclear.
What Actually Changed
Marketing cookies are now on by default for free-tier users. Under the revised policy, ChatGPT’s free version enables marketing cookies automatically, rather than requiring an opt-in. These cookies track behavior for targeted advertising and marketing purposes — a standard mechanism across the ad-supported web, but a new one for ChatGPT specifically.
Ad targeting is described as contextual, not built from a persistent behavioral profile. OpenAI’s stated model relies on real-time contextual signals — things like ad interactions or the immediate context of a conversation — rather than building a long-term profile of you across sessions. That’s a more limited approach than the profile-based targeting common on social platforms, at least as OpenAI describes it.
Free and Go-tier users see ads; the policy explicitly ties ad personalization to your account settings. The policy states OpenAI may personalize ads “subject to your settings,” and separately confirms OpenAI may now receive purchase information from advertisers and data partners — a new disclosure that wasn’t part of the privacy policy before this update.
How OpenAI Got Here
The shift toward advertising wasn’t a sudden pivot — it followed mounting, well-reported financial pressure. Running large-scale AI models at the usage volume ChatGPT has reached is expensive, and subscription revenue from paying users alone hadn’t closed the gap reporters and analysts had been pointing to for much of the prior year. An ads pilot, layered onto a product that already has hundreds of millions of free users, is a fairly conventional answer to that kind of revenue gap — it’s the same playbook nearly every major consumer tech platform has followed before it.
What makes the timing notable is less the decision itself and more how quickly it moved from “pilot” to a privacy-policy rewrite spelling out the mechanics in detail. That pace suggests the ad program is intended to scale, not stay a small experiment — which is exactly why the details of how it’s bounded matter more now than they would for a genuinely minor test.
How This Compares to Google and Meta’s Ad Models
Context helps calibrate how big a shift this actually is. Google’s advertising business has, for two decades, been built on deep behavioral profiling across search history, email content (historically), location, and browsing activity — the most extensive profile-based targeting system in the consumer internet. Meta’s model is similarly profile-heavy, drawing on social graph data, engagement history, and now, following the Instagram encryption change, potentially message content as well.
Measured against that baseline, OpenAI’s stated contextual-only approach is meaningfully more conservative — closer to how search ads have traditionally worked (matching an ad to what you’re doing right now) than to the persistent-profile model Google and Meta have run for years. That comparison doesn’t make ChatGPT’s ad rollout a non-event, but it’s a useful anchor for how much should change in your behavior versus how much is simply a new business line layered onto an existing product.
What OpenAI Says Advertisers Don’t Get
OpenAI has been specific on one point: the company states that advertisers do not get access to your conversations, your chat history, your personal details, or anything stored in ChatGPT’s memory feature. Ad personalization, per this claim, draws only on signals that stay within the ad system itself — interactions with ads, not the substance of what you’ve discussed or uploaded.
That’s a meaningful line to draw, and worth taking seriously rather than dismissing outright — it’s a structurally different approach from a platform that mines conversation content directly for ad targeting. It’s also, necessarily, a claim that depends on trusting OpenAI’s internal data boundaries, since there’s no independent, ongoing audit of the separation between ChatGPT’s conversational data and its ad infrastructure that an outside user can verify directly.
Where the Nuance Actually Lives
“Contextual, not behavioral” is a real distinction, but it’s worth being precise about what it does and doesn’t rule out.
Ad interaction history is itself a form of profile, even if it’s not built from your conversation content. Which ads you’ve seen, clicked, or dismissed, combined with purchase data now disclosed as something OpenAI may receive from advertising partners, is still data accumulating about you over time — just through a narrower channel than full conversation mining.
Marketing cookies being on by default for free users is a meaningful shift in posture, even if the targeting mechanism itself is contextual. Defaults matter: most users never change a default setting, so “you can turn this off” and “this is off unless you turn it on” produce very different real-world outcomes for the same underlying feature.
Paid tiers are explicitly treated differently in OpenAI’s messaging, which is itself informative — it suggests the ad-supported free tier is the one where this tradeoff is most concentrated, and that paying removes you from at least part of that equation.
If You’ve Ever Uploaded a Personal File to ChatGPT
A lot of ChatGPT use isn’t conversational at all — it’s document and photo processing. People upload scanned IDs to extract text, paste in handwritten notes to transcribe, describe old family photos to help organize captions, or upload a PDF to summarize. None of that activity is new because of the ads rollout, but it’s worth re-examining now that the platform’s business model has visibly shifted to include advertising.
OpenAI’s claim that advertisers can’t see chat content or uploaded files is the operative protection here, and it’s a real one if it holds. But it’s also worth noting that this claim covers the relationship between ChatGPT and advertisers specifically — it says nothing about OpenAI’s own use of uploaded content for model training or product improvement, which is governed by separate settings and separate parts of the privacy policy, not the ad-specific provisions.
The Setting Worth Checking Today
Check your ad personalization controls. OpenAI states users can dismiss ads, give feedback on why an ad was shown, delete their ad data, and manage personalization settings directly. If you haven’t looked at these since the rollout, they’re worth a five-minute review — particularly the option to delete accumulated ad data, which resets at least that portion of the profile being built.
Separately, check your model training and memory settings. These predate the ad rollout and aren’t new, but the ads update is a reasonable prompt to confirm they’re still set the way you intend — particularly if you’ve ever uploaded a personal document or photo and want confidence about whether that content can be used to improve OpenAI’s models.
What This Doesn’t Change — and What It Signals
Uploading a personal file to any AI tool for one-off processing has always carried its own separate exposure question, independent of whether that tool shows ads — it’s about retention, training use, and who can access stored conversation history, not about advertising specifically. The ads rollout doesn’t make that exposure worse on its own.
What it does signal is a broader pattern worth tracking: ChatGPT’s monetization is diversifying beyond subscriptions, and a platform funded partly by advertising has a different long-term incentive structure than one funded entirely by people paying directly for the product. That’s not a reason to stop using ChatGPT for document and photo tasks — it’s a reason to be deliberate about which tasks you route through an ad-supported free tier versus a paid one, and which files you’d rather not put through either.
A Closer Look at What “Contextual” Actually Tracks
It’s worth pressing on the word “contextual” one more time, because it’s doing a lot of work in OpenAI’s framing. Contextual targeting, in the advertising industry generally, means an ad is matched to the immediate situation — what page you’re on, what you just searched, what you’re currently doing — rather than to a long-term behavioral dossier built from months or years of activity.
Applied to ChatGPT, that means an ad shown during or after a conversation about, say, travel planning might be travel-related, based on that conversation’s general topic, without that topic being logged into a persistent profile that follows you into unrelated conversations weeks later. That’s the theory, and it’s a real, meaningfully narrower approach than persistent profiling — but it still requires the platform to read enough of the conversation’s topic to know an ad is relevant, which is a different claim from “advertisers never have any relationship to what you discussed.” The boundary OpenAI describes is between advertisers and your content, not between ChatGPT’s own ad-serving logic and your content in the moment an ad is selected.
That distinction is subtle, but it’s the difference between a real privacy protection and a marketing reassurance that sounds like one. Whether OpenAI’s internal systems hold that line as cleanly as the external messaging suggests isn’t something a user can verify directly — it’s a claim to track over time, especially as the ad program scales past its pilot phase.
A Cleaner Default for Files You Just Want Stored, Not Analyzed
Not every photo or document needs an AI tool to process it — sometimes you just need somewhere to keep it. For that category, daftei is a simpler proposition than any AI platform’s privacy policy: it doesn’t run ads, doesn’t train AI on your content — its own or any third party’s — and doesn’t sell your data, so there’s no business-model question to track as it evolves.
Files are encrypted in transit with TLS 1.3 and at rest with AES-256, daftei is available on iOS, Android, and the web, with 5 GB free and unlimited storage on Pro. For the documents and photos you want kept rather than analyzed, that’s the simpler default.