Some AI Overview links carry a ?st_source=ai_overview parameter

Links inside AI Overviews and AI Mode have been seen carrying ?st_source=ai_overview on the end of the destination URL, Search Engine Roundtable reported on October 2. Tom Pool, who posted the screenshots to LinkedIn, saw the parameter on links cited inside the answer and on the links in the source list underneath, and asked whether it was the start of proper attribution reporting. Barry Schwartz, Search Engine Roundtable’s executive editor, tried to reproduce it in several browsers and could not.

A parameter on the landing URL would separate AI clicks from ordinary ones, which nothing in Search Console does today. AI Overview clicks arrive as plain Google organic referrals, indistinguishable from a click on a blue link, so the only honest answer to “how much traffic did the AI answer send” has been a guess. Google has done the tagging work before on links out of Gemini, which moved those visits out of direct traffic.

Schwartz then added an update that changes what the parameter might be. An anonymous source told him this is a black hat tactic, with people forcing Google to index parameterized URLs at scale to measure exact clicks from AI Overviews. The two explanations put the parameter in very different places. If Google appends it when someone clicks, the parameterized URL has no reason to be in the index, and the parameter is a referral artifact that only affects reporting. If the parameterized URL is instead what Google indexed and cited, the duplicate already exists in the index, and cleaning it up is the site’s job.

Neither reading can be settled from outside Google right now. The parameter is undocumented, unconfirmed, and not reproducible by the editor who wrote it up. The safer assumption is that the exact string and the conditions that trigger it can change without notice, so it is worth counting this week and not worth wiring into a dashboard.

What to do

• Grep server logs and CDN logs for st_source=ai_overview, and for a bare st_source= in case the value differs. A few days of data is the point here, not a single spot check, since Schwartz could not reproduce the behaviour on demand.
• In GA4, this is not a utm_ parameter, so it will not populate source or medium. Report on landing page or page path containing st_source instead.
• Run URL Inspection on one parameterized version of a page that gets cited in AI Overviews, and check the Pages report for parameterized URLs that have been indexed. Confirm the page’s canonical tag points at the clean URL, so a parameterized copy consolidates instead of competing.
• Do not start appending a parameter like this yourself to buy the measurement. The tactic described to Schwartz works by getting Google to crawl and index extra copies of pages that already rank, which is a cost paid in duplicates for a number that Google may yet hand over for free if the test rolls out.