Printed analytics report with line charts beside a pen, keyboard and phone on a wooden desk, illustrating AI Overview traffic attribution checks

On October 2, 2026, Search Engine Roundtable reported that links inside Google AI Overviews and AI Mode were carrying a new URL parameter, ?st_source=ai_overview, after SEO Tom Pool shared screenshots on LinkedIn. The story was later corrected: the parameter came from a Chrome extension rewriting links in the browser, not from Google, so there is still no Google-supplied tag that identifies AI Overview clicks.

The correction matters more than the sighting. Here’s why that parameter can’t measure what it appears to, what Google actually reports, and how a small Michigan team can read AI search traffic honestly before holiday reporting.

What was the st_source=ai_overview parameter, and did Google add it?

No, Google did not add it. Per Search Engine Roundtable’s update, Pool traced it to a Chrome extension modifying URLs on the results page, which he suspected was collecting the data to resell. Barry Schwartz had already tried to replicate the parameter across several browsers and couldn’t.

The sighting looked convincing because the tag appeared on links in the AI Overview text and in its source list. The industry wants AI Overview attribution badly enough that a browser add-on briefly passed for a Google feature.

How does a browser extension add tracking parameters to Google’s links?

An extension with permission to read and change pages on google.com injects a content script that rewrites the href of links after the page loads. Google sends a clean URL; the extension edits it in that one person’s browser; the click lands on your site carrying the extra parameter.

Once you understand that mechanism, you can predict how the data behaves in any case. Three consequences follow:

  • It only counts people who installed that extension. The share of your traffic carrying the tag tells you how many visitors run that add-on. It says nothing about how much of your traffic comes from AI Overviews.
  • It doesn’t change channel attribution. GA4 builds source and medium from utm_ parameters and the referrer. An unknown parameter like st_source only shows up in the Page location dimension, so those sessions still count as google / organic.
  • It skews toward personal devices. Many Michigan manufacturers and health systems use managed Chrome policies that block unapproved extensions. A buyer researching on a work laptop at a Sterling Heights Tier 2 supplier won’t appear in that sample at all, but the same person browsing on their phone at home might.

So a team that built an “AI Overview” segment on st_source would see a small number, decide AI search barely sends traffic, and cut content work that is actually earning citations. A consumer-heavy site might see the opposite and over-invest.

What does Google actually report about AI Overview traffic today?

Impressions, by page, and nothing on clicks. Google introduced the Search Generative AI performance report in Search Console on June 3, 2026. It counts how often your URLs appeared in AI Overviews, AI Mode and Discover’s generative features. You can break it down by page, country, device and date.

It launched for a subset of UK sites, tied to the Competition and Markets Authority’s conduct requirements. Search Engine Roundtable reported on June 23, 2026 that access was spreading to the United States, India and Switzerland, and John Mueller said Google was rolling it out to sites gradually. If your Michigan property doesn’t have it yet, that’s expected, not an error.

What’s missing is the important part: no clicks, no click-through rate, no query data. Clicks from AI features still roll into the standard Web performance report, mixed in with everything else.

Why won’t Google just tag AI Overview clicks?

Because the metric that’s missing would make AI Overviews look like a traffic cost, and Google’s recent product changes push the other way. That’s our reading, not a Google statement — but the evidence lines up.

On September 22, 2026, Search Engine Roundtable reported that some links at the bottom of AI Overviews now send users into AI Mode rather than to websites, a change first flagged by Gagan Ghotra. A product that is routing some clicks to itself has little reason to publish a clean per-feature click-through rate. That rate would show publishers exactly how many clicks they lose when they’re cited.

Our view: waiting for an official parameter is the wrong plan. Assume impressions-only reporting will last through Q4, and build attribution you control.

Why do Michigan B2B and multi-location businesses feel this most?

Their buying paths are long or offline, so the AI citation and the sale end up far apart. Take an engineer at a Warren tooling shop who reads an AI Overview citing your material-spec page in March, then sends an RFQ in May. Last-click reporting credits direct traffic or email. The citation that started it all doesn’t show up anywhere.

Multi-location service businesses have a different version of the same problem. An HVAC company covering Livonia, Southfield and Rochester Hills gets most AI-influenced demand as phone calls from a Business Profile, never as a tagged website session. Our earlier piece on AI Overviews appearing on 93% of brand searches showed how far that coverage now reaches. When nearly every brand query triggers an AI answer, “we can’t measure it” turns into “we can’t measure most of our search demand.”

How should a small team measure AI search traffic this week?

Compare Google’s impression data with your own click and lead data, page by page, and add one human question to your forms. None of the steps below need paid tools.

  1. Search GA4 for the parameter. In Explore, build a free-form table with the Page location dimension filtered to “contains st_source.” Any hits mean some visitors run a URL-rewriting extension. Note the count, then exclude it from any AI reporting.
  2. Raise GA4 data retention. Under Admin, Data retention, change event data from the default 2 months to 14 months. Explorations can’t look back further than this setting, and you’ll need year-over-year comparisons by spring.
  3. Export the Generative AI report if you have it. Pull page-level AI impressions weekly into a shared sheet. If you don’t have access yet, log the date you checked and look again monthly.
  4. Put AI impressions next to organic clicks. For each page, add Search Console Web clicks for the same week. Pages where AI impressions rise while clicks stay flat or fall are “cited, not clicked” pages.
  5. Give cited pages a reason to click. An AI answer can summarize a spec. It can’t run your fitment lookup, show live inventory at your Dearborn branch, or give a quote. Add the interactive or local element above the fold.
  6. Add a self-reported source field. Add “Google AI answer,” “ChatGPT or another AI tool,” and “Google search results” as separate options to your contact form and your call-intake script. In B2B, this field often captures what analytics can’t.
  7. Run a clean-browser citation check monthly. Pick ten money queries. Search them in a fresh Chrome profile with no extensions, and record which of your URLs are cited.
  8. Confirm self-referencing canonicals. Extension-tagged URLs can get copied into emails and forums. A canonical tag on every page keeps those parameter versions from competing with the clean URL.

Does parameter noise from extensions hurt rankings?

Not directly, as long as your canonicals are sound. The parameter only exists in individual users’ browsers, so Googlebot never sees it on your internal links. The real risk is to your reporting. Session counts split across parameter versions in page reports, and caching layers that key on full query strings can serve slower, uncached pages to those visitors.

What dates should be on the calendar?

  • September 22, 2026 — AI Overview links routing to AI Mode reported. Use this as a marker when you compare clicks before and after.
  • October 2, 2026 — st_source=ai_overview sighting, then corrected to a Chrome extension.
  • Around October 8, 2026 — expected end of the September 2026 spam update’s roughly two-week rollout (phase two hit September 30). Keep spam swings separate from AI-click trends; our spam update diagnostic covers how to tell them apart.
  • October 31, 2026 — first full month of the impressions-versus-clicks sheet. Set your baseline here.
  • November 27, 2026 — Black Friday. Retail and service promotions need at least four weeks of baseline before this date for any lift reading to mean something.

Frequently asked questions

Can I track AI Overview clicks with UTM parameters?

No. You don’t control the links Google shows in AI Overviews, so you can’t add UTMs to them. Any UTM on those links would come from Google or from something in the visitor’s browser. Google hasn’t added one. What you can track is the outcome: page-level AI impressions in Search Console, compared with organic clicks and leads for the same pages.

Is the Search Console Generative AI report available in the US?

It’s rolling out gradually. Google introduced it on June 3, 2026 for a subset of UK sites. Search Engine Roundtable reported on June 23 that US, Indian and Swiss sites were starting to get access, and Google said it was rolling out to sites one by one. If your property doesn’t show it yet, check again monthly.

Should I block visitors who arrive with st_source in the URL?

No. They’re real people who happen to run a browser extension, and blocking or redirecting them costs you leads. Just keep the parameter out of your reporting logic. Exclude it from any AI-traffic segment, make sure canonical tags point to the clean URL, and don’t use the parameter’s share of traffic as a stand-in for AI search volume.

How do I tell if AI Overviews are costing my site clicks?

Compare the same pages over time. When a page’s AI impressions in Search Console climb while its Web clicks stay flat or fall, and rankings for its main queries hold, the AI answer is probably satisfying searchers before they click. Leave out weeks affected by the spam update so a demotion doesn’t look like AI cannibalization.

Category
Tags

Comments are closed