Q16 Blog

OpenAI Adds Advertising Data Collection to ChatGPT's Free Tier

A new "Ads data" category appeared in OpenAI's consumer privacy policy in June. Our monitor flagged it, we rated it critical, and OpenAI's privacy score dropped from 5.1 to 4.8.

July 13, 2026 · Q16 PBC

Correction, October 3, 2026. This post compared two different documents. The May 18 version is OpenAI's US privacy policy, and the June 4 version is its general policy for other regions, so the "new" Ads data sentence below did not replace anything in the US policy. It appears in the general and Europe policies and not in the US one. We have restored OpenAI's rating to 5.1 and the index to 3.56. Details are in the Q3 2026 Policy Watch Report. The original post follows unchanged.

On July 13, the Frontier Watch policy monitor flagged a change to OpenAI's consumer privacy policy. The document's effective date had moved from May 18 to June 4, 2026, and roughly a third of the text was new.

The substance is a new category of data collection called "Ads data." The policy now reads: "For Free and Go users, subject to your choices and controls, we collect information about your ads history and interests, such as the types of ad content that you view or engage with."

No advertising category existed in the previous version. The phrase "Ads data" appears four times in the new text and zero times in the old. Advertising has arrived in ChatGPT, and the data collection that supports it is now in the policy.

How we rated it

Every change our monitor detects gets a severity rating before it reaches you. We rated this one critical.

The reasoning is simple. Plenty of policy edits enlarge an existing practice, a longer retention window here, a new vendor there. This one creates a practice. Ad history and inferred interests are a new class of personal data for OpenAI to hold, and interest profiles built for advertising describe you in ways a chat log does not. A category that did not exist last month now does.

What it did to the score

OpenAI's consumer privacy rating fell from 5.1 to 4.8 on our ten-point scale. That moves ChatGPT from second to third among the ten model makers we rate, below Google's Gemini at 4.9 and further behind Anthropic's Claude at 5.3.

The entire drop came from one dimension, the one that scores third-party sharing and advertising use. We left the training dimension untouched, and the reason matters: the policy says nothing about whether ads history feeds model training. We score what the text says, and silence is not a disclosure. If OpenAI later states that ad data enters training, that dimension moves too.

The boundaries

Two limits are in the text itself. The new category applies to Free and Go users, so the policy does not extend advertising data collection to paid ChatGPT plans. And it applies to the consumer product, not the API. OpenAI's business terms still prohibit training on customer content by default, and its API scores 8.3 on our rubric. The distance between OpenAI the model provider and ChatGPT the consumer product was already wide. It just widened again.

That gap is the recurring finding of this project. The models are governed by strict contracts. The products wrapped around them are governed by consumer policies that keep getting thinner, and the free tiers are thinnest of all.

See for yourself

The full ratings, the six-dimension rubric, and every score's basis are public at watch.q16pbc.com. Rated companies can dispute any score with evidence at info@q16pbc.com, and evidence-backed corrections are re-scored within ten business days.

← All posts

See where the frontier stands.

Frontier Watch tracks how every major AI lab handles your data — and flags it in plain language the moment a policy changes. The baseline reading is free.

Open the Dashboard