Evidence: where does your conversation go?

The matrix, cell by cell

Claude (Anthropic): no, no, no. CONFIRMED at primary. Anthropic committed publicly on 4 February 2026 (“Claude is a space to think,” anthropic.com/news/claude-is-a-space-to-think): “We want Claude to act unambiguously in our users’ interests. So we’ve made a choice: Claude will remain ad-free.” No sponsored links adjacent to conversations, no advertiser influence on responses, no unrequested placements. The stated reason is incentives, not privacy: even visually separate ads “would introduce an incentive to optimize for engagement,” and “the most useful AI interaction might be a short one.”

ChatGPT (OpenAI): yes, yes, no. CONFIRMED; third cell is the vendor’s position. Ads live since 9 February 2026 on Free and Go tiers. Market sequence: US (9 Feb) with Canada, Australia and New Zealand in the launch cohort; UK from 6 Jun; Japan, South Korea, Mexico and Brazil confirmed live in OpenAI’s own update of 11 Aug 2026 (openai.com/index/testing-ads-in-chatgpt). Nine markets for users seeing ads; the advertiser-side buying list is shorter and separate. On 15 August 2026 OpenAI Ireland notified Free and Go users in the EEA and Switzerland that ads begin “later this month,” non-personalised at launch; Plus, Pro, Business, Enterprise and Education stay ad-free. Third cell, stated precisely: OpenAI’s documentation describes ad selection using conversation context inside the product; no mechanism sending conversations to other surfaces is disclosed. The “no” records the vendor’s stated design, not an independent audit. Sources: OpenAI, “Testing ads in ChatGPT” (openai.com/index/testing-ads-in-chatgpt, update 11 Aug 2026); OpenAI Ireland user notice 15 Aug 2026; press: ppc.land 15 Aug, trendingtopics.eu 17 Aug.

Gemini (Google): disputed, disputed, disputed. The app carries no ads as of mid-August 2026. The position moved in public: 8 December 2025, VP Global Ads Dan Taylor: “There are no ads in the Gemini app and there are no current plans to change that.” 29 April 2026, CBO Philipp Schindler on the Q1 earnings call, as quoted in press coverage of the call: “open-minded,” and a format that works in AI Mode “would transfer successfully to Gemini app” (attribution via relay; transcript check open). Google already sells ads in its AI search: live in AI Overviews, in testing in AI Mode per Google’s own ad liaison, with four formats unveiled at Google Marketing Live, 20 May 2026. Disputed marker held until either a commitment or a launch. Sources: Search Engine Land 8 Dec 2025; Alphabet Q1 2026 earnings call via press; blog.google 20 May 2026.

Grok (xAI): disputed, disputed, disputed. Musk announced ads inside Grok’s answers to advertisers in August 2025. As of 17 August 2026 no primary source confirms paid placements live inside answers. Note to avoid a common conflation: X has integrated Grok into its Ads Manager as an advisory tool for advertisers (2026); that is a different product from ads inside Grok’s answers and is not evidence the placement shipped. Sources: Search Engine Land 8 Aug 2025; Engadget 7 Aug 2025.

Meta AI (Meta): no, yes, yes. CONFIRMED. This is the red cell. Since 16 December 2025, conversations with Meta AI feed ad and content personalization across Facebook and Instagram. No opt-out beyond not using Meta AI. Sensitive categories excluded by platform filter. Only post-16-December chats are used. The EU, UK and South Korea are excluded on data-protection grounds, not product design. Sources: Reuters, Oct 2025 announcement; Forbes 6 Oct 2025; Meta policy notice, live 16 Dec 2025.

Method note

Cells answer the question as of the check date, from vendor primary statements where they exist and dated press where they do not. Disputed means the vendor’s own statements conflict or a launch is announced but unverified. Scope: this matrix covers advertising and cross-surface data flows only. Whether conversations train models is a different question with a different map; that one comes later. This page updates when cells move; the matrix panel is dated and stays as shipped.

Addendum 18 Aug 2026, on a reader challenge (Arkady Pildes): column 2 has a resolution limit. It marks whether chats influence ads, not how. Two mechanisms share one square: in-session contextual selection (OpenAI's stated model, non-personalised at EU launch) and persistent profile building (Meta's model: conversations feed the advertising profile that follows the user across surfaces). The distinction matters and the matrix does not show it; this page now does.

Addendum 18 Aug 2026: the five vendors' full terms and privacy policy texts were not audited line by line for this page. Cells record public commitments, launches and dated statements, at the vendors' word. A policy-text-level read is the standard for the training map, and that is the next page.

Source exclusions: one AI-generated answer page carrying a conflicting market list was found and excluded from the source base (machine-compiled, no editorial provenance).