Case study: Gumlet turned ChatGPT mentions into 20% of inbound revenue. Read it →
AI Overview Links Now Open AI Mode: Why Citation ≠ Click for B2B SaaS
TL;DR
- On September 21–22, 2026, Search Engine Roundtable reported Google testing anchor links inside AI Overviews that open AI Mode instead of publisher pages (first spotted by Gagan Ghotra on X).
- That UI test lands next to a preregistered field experiment by Wang, Gleason, Bart, Wilson, and Metaxa (UPenn / Northeastern): forcing searches into AI Mode cut external click-through by 18.8 percentage points vs current Search (N=1,100). Covered by Search Engine Journal on Sep 21.
- Authors declare no competing interests; NSF-funded. Compliance caveat: the No-AI condition degraded mid-study after Google changed Overview HTML, so the stronger clean estimate is the AI Mode assignment effect.
- Operator read for B2B SaaS: an Overview citation is not a visit, and a visit is not a demo. Measure impressions, outbound clicks, branded search, and CRM self-report as separate layers (see DerivateX on the GSC generative AI report and zero-click pipeline strategy).
- Soft commercial next step stays the same: start with a free AI visibility audit if you need a prompt-level baseline of where Google still sends people off-domain.
Google can cite you in an AI Overview and still keep the next click inside Google.
That is the product change Search Engine Roundtable documented this week, and it matches what a large field experiment already measured about AI Mode: more time on Google, fewer outbound clicks.
For B2B SaaS teams that treat “we appeared in the Overview” as a growth win, the gap between visibility and pipeline just got easier to see.
What Google is testing in AI Overviews
Barry Schwartz at Search Engine Roundtable (Sep 22, 2026) described normal-looking links at the bottom of AI Overviews that open follow-up questions in AI Mode instead of sending the user to a website.
The pattern was first shown publicly by Gagan Ghotra, who framed it as another push from classic results into Google’s conversational surface.
What matters for operators:
- The control still looks like a link.
- The destination is Google’s answer experience, not your URL.
- Citation chips and “Show more” links already sit in a leaky click path; this test makes the leak intentional for follow-ups.
If your content marketing KPI is “AI Overview appearances,” this test asks a harder question: appearances of what, and with what outbound behavior attached?
What the field experiment measured
The draft paper AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence (Wang et al., 2026) ran a preregistered browser-extension experiment on real Google Search use.
Core design:
- N=1,100 participants with tracked searches over seven treatment days (1,444 enrolled; 956 finished the post-survey).
- Three arms: No AI (hide Overviews / redirect AI Mode), Current Search, Forced AI Mode.
- Primary behavioral outcome: external click-through rate from Google.
Headline causal results for AI Mode assignment vs Current Search:
| Outcome | Estimate |
|---|---|
| External click-through rate | −18.8 pp (95% CI −22.2 to −15.3) |
| Search sessions per day | −0.92 sessions |
| Minutes per session | +0.43 minutes |
| Users who also searched Bing / DuckDuckGo / Yahoo | +11.2 pp |
| Trust in information on Google | Down on a 7-point scale |
In plain language: people spent longer inside Google’s AI search surface, clicked out less, ran fewer sessions, and liked the experience less. That is the opposite of “AI Mode sends more open-web traffic.”
Search Engine Journal’s Sep 21 write-up correctly stresses the conflict with Google’s public claims about delight and referral growth.
What the study does not prove
Be precise with buyers and boards:
- The 18.8 pp figure is an experiment effect under forced AI Mode, not a universal forecast for every SaaS site tomorrow.
- The No-AI arm’s +8.8 pp click lift is a local average treatment effect after Overview-hiding compliance fell when Google changed markup mid-study.
- The sample skewed younger, highly educated, and US-based.
- Baseline voluntary AI Mode use was tiny (about 0.6% of searches), so the AI Mode arm is a full-adoption stress test, not organic uptake.
Those limits do not erase the directional story. They do stop you from pasting “−18.8% traffic” onto a board slide.
Why this matters for B2B SaaS pipeline
1. Overview citation and site visit are different products
Google’s Search Console generative AI report now shows impressions for AI Overviews and AI Mode worldwide. It still does not give AI-only clicks or the queries that triggered the answer.
If Overview follow-ups open AI Mode instead of your page, impression growth can rise while referral sessions stay flat. That is not a reporting bug. It is the product.
2. Zero-click was already the default; AI Mode deepens it
DerivateX’s zero-click search strategy already treats citation and recommendation as pipeline inputs that often skip the session. This week’s UI test is a concrete mechanism for that skip: Google answers, then invites another answer inside Google.
3. Recommendation share still matters more than vanity chips
If the buyer finishes the journey in AI Mode, the brands named in the answer still shape shortlists even when no chip is clicked. Keep recommendation share vs citation share as separate lines. Do not let a rising Overview impression chart replace naming tests on ChatGPT, Perplexity, Claude, and Gemini.
Operator playbook this week
Layer 1: Separate the three Google signals
- GSC Gen AI impressions by URL (visibility only).
- Outbound clicks / sessions on pages that usually appear in Overviews (leaky referral).
- Branded search + CRM “how did you hear” with AI options (decision residue).
If Layer 1 rises and Layer 2 does not, you are winning the box, not the visit. That can still be fine if Layer 3 moves.
Layer 2: Make the leave-Google click worth it
When a user does leave AI Mode or an Overview for your URL, the first screen has to deliver what the summary withheld:
- Pricing bands, packaging limits, SSO / security constraints, integration truth.
- A comparison table AI cannot fully compress without losing buying criteria.
- A demo path that matches the prompt class (PLG trial vs sales-led audit).
Thin “we got cited” landing pages will lose the rare click this test leaves on the table.
Layer 3: Prompt-panel hygiene across engines
Run the same 10–20 buyer prompts weekly across Google AI Mode / Overviews, ChatGPT, Perplexity, and Claude. Log named vs cited, sources used, and whether Google kept the follow-up in-product.
That panel is how you notice Overview→AI Mode leakage before analytics invents a false organic dip story.
Worked example: mid-market RevOps SaaS
Priority prompts: “best RevOps platform for Series B,” “HubSpot RevOps alternatives with Salesforce sync,” “what is pipeline hygiene software.”
Weak read of this week’s news: panic-rewrite every blog title because “AI Mode stole clicks.”
Better read:
- Confirm which money URLs show Gen AI impressions in Search Console.
- Compare those impression trends to sessions and demo starts from the same URLs.
- On commercial prompts, check whether Google names you while citing G2 / listicles / docs, then whether follow-up chips stay in AI Mode.
- Invest first in accurate third-party facts and extractable owned constraints, not another Overview-optimized fluff paragraph.
That sequence turns a scary SER screenshot into a measurement and content queue.
How DerivateX thinks about Overview→AI Mode leakage
- Treat Google generative features as a visibility surface first, a referral channel second.
- Pair GSC Gen AI impressions with prompt-panel naming and CRM attribution; never with impressions alone.
- Build pages that win the rare exit click: decision facts, constraints, and proof the model summary cannot finish.
- Keep off-Google engines in the same weekly ritual. Google’s UI can change overnight; buyer shortlists still form across tools.
If you want a baseline that separates Overview impressions from naming and referral reality, start with a free AI visibility audit. Engagement options stay on pricing. For the measurement stack this UI test stresses, see the GSC generative AI report guide and zero-click pipeline strategy.
FAQ
What changed in AI Overviews this week?
Google is testing anchor-style links inside AI Overviews that open AI Mode follow-ups instead of publisher pages, per Search Engine Roundtable coverage of Gagan Ghotra’s examples (Sep 21–22, 2026).
Does that mean Overview citations no longer matter?
No. Citations still signal extraction and can shape answers even when the next click stays on Google. They are a weak proxy for sessions. Measure both.
What does the Wang et al. study add?
Causal evidence that forced AI Mode reduced external click-through by 18.8 percentage points vs current Search in a 1,100-person field experiment, while also lowering trust and satisfaction scores.
Should we opt out of generative AI in Search Console?
Only if you have a deliberate distribution reason. Opting out removes AI Overviews / AI Mode exposure and the impressions that come with it. Most B2B SaaS teams should stay in, then instrument the leaky click path.
How should we report this to a CMO?
Three lines: Gen AI impressions (visibility), outbound sessions from those URLs (referral), branded + CRM AI discovery (decision). Do not collapse them into “AI traffic.”
What should we change this week?
Audit Overview-prone URLs for decision-grade first screens, wire Gen AI impressions next to sessions, and re-run commercial prompt panels to see whether Google keeps follow-ups in AI Mode.













