Google’s AI Mode and OpenAI’s ChatGPT were both spotted this week running interface experiments that move source citations from inline placement to the bottom of a generated answer. Sachin Patel documented the AI Mode test on X, writing that “Google appears to be testing a change to citation cards in AI Mode. When opening a free listing product, the citation cards move to the bottom of the page.” Brodie Clark, posting on his SERP Alert account, flagged a parallel shift in ChatGPT: “ChatGPT seems to be experimenting with how it’s linking out to sources from its output. The same goes for the organic shopping experience, which now has a new display and product panel that appears after clicking.”
Neither company has announced the change. Both practitioners describe live, in-the-wild experiments, not confirmed rollouts, and Search Engine Roundtable flagged the two sightings landing in the same week as worth reading together.
That timing is the actual story. One engine testing a citation redesign is a UI tweak. Two engines, built by competing companies with no shared codebase, arriving at the same placement decision within days of each other suggests a considered design direction rather than an isolated experiment. When independent teams converge on an interface choice this specific, it is worth treating as a signal about where answer engines think citations belong, not a coincidence to shrug off.
The open question is which way this cuts for publishers, and nobody selling GEO (generative engine optimization, the practice of optimizing for LLM-driven answers) advice actually has the click data to answer it. Two reads are equally defensible right now.
The optimistic read: a citation placed at the end of the answer meets the reader exactly when they are ready to act. Someone who has finished reading a full AI-generated response and still wants the source is a higher-intent visitor than someone who clicks an inline link mid-answer, out of curiosity, before finishing the thought. Bottom placement could concentrate clicks among readers who were already going to convert.
The pessimistic read: most readers do not finish long generated answers. A citation sitting below several paragraphs of AI Mode or ChatGPT text is a citation most sessions never reach. Inline links interrupt the reading flow, but they are visible to every reader who gets that far in the answer, not just the ones who scroll to the end.
Both mechanisms are plausible and mutually exclusive as descriptions of the same interface change. Nothing published so far, not by Google, not by OpenAI, not by either practitioner who spotted the tests, includes click-through data comparing the two layouts. Any claim beyond that is speculation dressed as analysis.
What publishers can do without waiting for either company to publish results is instrument their own analytics now. If AI Mode or ChatGPT ship bottom-anchored citations broadly, the referral-traffic signature from those engines should move, up or down, inside a site’s own Search Console and analytics data before either company says anything official. Segmenting AI-referral traffic as its own channel, separate from organic and direct, is the only way to catch that shift in real time rather than reconstruct it after the fact from a quarter of declining or rising sessions.
The practical move for the next ninety days is not to guess which read is correct. It is to build the segment that will tell you the answer the moment either platform rolls the change out past the test group.
Per Search Engine Roundtable’s July 22, 2026 report by Barry Schwartz, citing sightings from Sachin Patel (AI Mode) and Brodie Clark of SERP Alert (ChatGPT).
