Getty's Image and the Sponsored Answer

Google said in July 2026 that it would automatically disclose ads made with its generative-AI advertising tools. It had already begun putting sponsored placements in AI Mode that resemble citations in its answers. One label identifies how an ad was made; the other marks a paid placement. Getty Images is licensing its catalog into ad-making tools and ChatGPT search.

A search-results page could give an advertisement its own box and leave organic links in another. An answer engine speaks in one synthesized voice, even when it draws on retrieved material and presents a paid recommendation. In that setting, provenance cannot remain a certificate checked after publication. It must become a live interface rule that separately identifies sponsorship, generated or altered material, image-use rights and the sources supporting factual claims as the operator assembles the answer.

Search’s boundary has shrunk to a label inside the reply

Google’s disclosure policy began with a narrower question. In 2023, it planned prominent disclosures for election ads containing generative-AI images, video or audio. Its July 2026 expansion covers ads made with Google’s AI advertising tools beyond elections. It does not, by itself, establish that an AI Mode recommendation is well sourced, or that every AI-made ad on the internet receives a label.

Google’s own sequence shows why policy and presentation must be kept distinct. In October 2025, the company released an AI-generated advertisement for AI Mode without a label identifying its use of Veo 3 and other generative tools. The later policy marks a change in what Google says it will disclose, not proof that users can already identify every kind of AI involvement in every setting. Meanwhile, AI Mode’s “sponsored” label carries a different burden: separating a paid placement from the citations it visually resembles.

Snap moved the commercial encounter further into conversation when it launched Sponsored Snaps in April 2026. A user can speak with a brand-specific AI agent in Snapchat’s Chat tab and receive product recommendations. That format makes the messaging platform’s role as an AI-agent gatekeeper concrete: the platform sets the terms on which a brand enters a conversation, while the user encounters the recommendation as a reply. Fluent dialogue can make guidance feel more authoritative than its commercial source warrants.

A signed image cannot identify a paid recommendation

Adobe, Twitter and The New York Times Company announced the Content Authenticity Initiative in 2019 to develop digital-content attribution. C2PA later supplied a cryptographic way to encode information about a media file’s origin and history. Both efforts addressed a real question: where did this piece of content come from, and what happened to it? An answer interface adds questions that a media file cannot settle alone.

Consider four claims an operator might need to make about one result. “Sponsored” identifies a commercial relationship. “Generated” or “altered” describes how the visual was produced. “Licensed” concerns permission for a particular use. A citation points to material that may support a factual statement. A photograph can have a valid origin record and still appear beside an unsupported claim; a licensed image can sit inside a paid placement without the license explaining who bought that placement.

Amazon’s requirement that third-party sellers label product ads containing AI-generated people illustrates the distinction. The label responds to a New York law concerning synthetic performers. It says something about the person depicted, not whether the seller’s product claims are true or whether the ad is sponsored. Combining those disclosures into one generic “AI” badge would discard the distinctions the rules are meant to preserve.

Getty is moving permission into the tools that use its images

Getty Images once had a relatively legible transaction to manage: a customer sought an image and permission to use it. Generative tools separate the moments of selection, alteration and display. Getty’s deals place its catalog closer to each of those moments, though the published announcements do not establish a single rights-checking system across them.

Getty and Nvidia agreed to develop AI tools trained on Getty stock images. Getty-backed BRIA built its generation business around more than one billion images licensed from 18 stock providers. In November 2024, TikTok added Getty content to its advertiser-facing Symphony Creative Studio. In June 2026, Getty licensed its library to OpenAI for ChatGPT search and discovery features. Training supply, ad creation and search display are different uses; a permission for one should not be assumed to cover the others.

Getty’s strategic opening is therefore more precise than “images for AI.” An advertiser selecting an image inside a creative tool needs to know what the license permits in that channel. A person viewing an image in ChatGPT search needs to be able to distinguish the image’s provenance from the evidence for the accompanying text. The agreements put licensed supply inside both workflows. They do not show that Getty images appear in sponsored ChatGPT answers, or that either operator automatically verifies every downstream use at display.

Metadata can travel farther than the visible warning

Google added a toggle in Gemini and Flow to remove visible watermarks while retaining invisible SynthID markers and C2PA metadata. That decision preserves machine-readable signals while allowing a user to see no immediate mark on the media. Researchers have also demonstrated removing watermarks from AI images and adding them to human-made ones. C2PA adoption faces a separate gap: cameras do not consistently support its metadata. A credential is useful evidence when present and intact, not a guarantee that every unmarked file is human-made or every marked one is authentic.

An operator assembling an answer has to make decisions at several boundaries. It can retain an image’s available origin record; check a license against the intended use; mark a placement as sponsored; and attach citations to the factual claims the answer actually makes. Each check belongs to a different object or decision. Signing the photograph does not sign the generated sentence beside it, and retrieving a source does not authorize the photograph.

Getty can negotiate access faster than it can settle the market

Getty has tried other ways to defend and enlarge its position. It withdrew its primary UK copyright-infringement claims against Stability AI. Getty has not ended its fight over image rights; it still has other UK claims and a separate US case to pursue.

In the coverage record, the UK Competition and Markets Authority raised concerns about Getty’s proposed $3.7 billion Shutterstock merger in October 2025. Getty terminated the deal in July 2026 after the regulator conditioned approval on Shutterstock selling its editorial business. Getty’s TikTok and OpenAI agreements offer a different route: negotiate permission where images enter a creative tool or a discovery product, without first combining two catalogs. The public terms do not establish how much value those integrations will return to Getty.

Getty can license the image before an assistant displays it. The operator still has to mark the paid recommendation beside it.