TL;DR
- Google AI Mode answers in prose and lists its sources, reached from the AI Mode tab or by adding &udm=50.
- Each turn of the conversation runs its own retrieval.
- On one live buying query it cited 26 domains against the AI Overview's 15; a follow-up left only 10 and added 12 new.
- 18 of the 26 never appeared on the classic results page for the same query.
Google AI Mode is a conversational search surface that answers your question with a written response and a panel of the sources it used, instead of a page of ten blue links. It sits behind its own tab in Google Search. Below is what it returns for a normal buying question, and what changes when you ask a second one.
What Google AI Mode is
AI Mode is Google’s chat-style search surface. You type a question, Gemini writes an answer from live web results, and a panel on the right lists the pages it pulled from. You can keep asking questions in the same thread, and the answer builds on what came before.
The mechanic underneath is what Google calls query fan-out. Google describes it as breaking your question into subtopics and issuing many queries at once. One question from you becomes a spread of searches, and the answer is assembled from all of them. Google Search Central confirms both AI Overviews and AI Mode may use that technique.
That single detail explains most of the behaviour in this article. Your page competes across a spread of related queries you never see, all generated on the fly from the one question a person typed.
Where to find AI Mode
Run any search on Google and look at the row of tabs under the search box. AI Mode sits first, to the left of All. Clicking it re-runs your query as a conversation. You can also skip the click by appending &udm=50 to a Google search URL.
The screenshot above is the ordinary results page for best crm for small business. Note that two AI surfaces are already in play: the AI Overview at the top, and the AI Mode tab offering a different treatment of the same query. They are separate systems with separate source lists, which is why we ran a side-by-side of what each surface cites and found the overlap smaller than most people assume.
Google’s own help page covers availability and account requirements in its AI Mode support article. Availability still varies by country, account, and device.
What an AI Mode answer looks like
Three things are worth noticing in that answer.
- It commits to specific brands. The response names Bigin by Zoho, Pipedrive, HubSpot, Less Annoying CRM, and Salesforce Starter, each with a short reason. There is no list of ten links for the reader to sort through.
- Each claim carries its own citation. Every recommendation has a small chip naming the source behind it, so attribution lands per sentence rather than per answer.
- Reddit carries real weight. Community threads sit next to vendor documentation and trade press, and the answer quotes what users said. That matches the pattern we wrote up in the case for treating Reddit as a citation channel.
The count in the panel was 26 sites. The AI Overview for the identical query cited 15. The wider read is the point: AI Mode fans out further, so more pages get a chance to be quoted, and each individual page carries less of the answer.
Where the citations live
Clicking Show all opens the full source list. This panel is the most useful part of AI Mode for anyone doing marketing, because it tells you exactly which pages shaped the answer.
The mix in that panel is the lesson. A Reddit thread, a Salesforce guide, a U.S. Chamber of Commerce listicle, a Zoho product page, a Slack post, and a YouTube review are treated as peers. Google reads all of them and quotes whichever one answers the sub-question in front of it, so a forum thread and a vendor doc can both feed the same paragraph.
That is why format work pays here. Our analysis of which sentences Google actually lifts found the quoted passage is usually a short, self-contained, factual statement. If your page buries its answer in a long paragraph, the engine skips it and quotes the competitor who wrote it plainly. The same logic drives our recommendations on structured data that AI systems can parse.
One caveat on measurement: AI Mode citations do not behave like rankings, and the deep links have changed shape before. When Google dropped text fragment anchors from AI Mode links, it quietly broke a lot of tracking setups.
Follow-up questions reset the source list
This is the part that separates AI Mode from an AI Overview, and it is the part with the biggest consequences for anyone trying to be cited.
I stayed in the same thread and asked a natural second question: which of those is cheapest for a 3 person team. The answer came back with per-plan pricing for all five products. The source panel came back with 11 sites.
The second turn ran its own retrieval from scratch. Comparing the two source lists domain by domain:
- 26 domains cited in turn one.
- 22 domains cited in turn two.
- 10 domains survived both turns.
- 12 domains were brand new in turn two, including Capterra, G2, TrustRadius, and Tech.co, none of which were cited in the first answer.
AI Mode reruns are not identical. Loading the same query twice gave different wording and a different lead product, so these counts describe one session, captured on 6 August 2026 from a logged-out browser with hl=en&gl=us. What held across every run was the shape: wide fan-out, a visible source panel, and a citation set that turns over between turns.
Read that from a brand’s point of view. Losing the opening answer does not put you out of the conversation. A comparison page that says nothing useful about the headline question can still win the pricing follow-up. We went deeper on this in our piece on designing content for the second and third question, and it is the single most under-served opportunity in AI search right now.
It also means single-turn tracking understates your exposure. If your tool only checks the first answer for a prompt, it is measuring one slice of a conversation that usually runs longer. Our rundown of the AI search metrics worth reporting and our method for measuring citation share across engines both assume multi-turn coverage for this reason.
How AI Mode compares with the classic results page
The classic results page for the same query surfaced 18 external domains. AI Mode cited 26. Of those 26, 18 never appeared on the classic page at all.
Names in that group included folk.app, OnePageCRM, Insightly, Databox, and a long tail of smaller vendor and blog domains. None of them ranked in the visible organic results for the query. All of them helped write the answer.
This is the practical case for treating AI visibility as its own measurement problem. A rank tracker pointed at this keyword would have reported nothing about 18 of the 26 sources Google actually read. If you want the traffic side of that picture, our guide to separating AI referral traffic inside GA4 covers the reporting setup, and planning for visits that never happen covers what to do when the answer satisfies the reader completely.
What this changes for your content
Nothing here calls for a new discipline. It sharpens the priorities inside the one you have. Our primer on generative engine optimization has the wider framing, and the AI Mode playbook has the tactical checklist. The four moves this test supports directly:
- Answer sub-questions on their own terms. Fan-out means Google runs searches you were never shown. Pricing, team size, integrations, and switching costs each deserve a passage that stands alone.
- Write the quotable sentence. Short, factual, self-contained. A price, a limit, or a comparison stated flatly is what gets lifted into the answer.
- Cover the follow-up. The second question is cheaper to win than the first and the field is thinner. Comparison and pricing content earned twelve fresh citations in one turn here.
- Be present where the engine already trusts. Reddit, YouTube, and trade associations appeared throughout both answers, which is the pattern behind the platforms AI engines cite disproportionately.
If you want to check your own footing before changing anything, our readiness checklist walks the technical and content prerequisites, and the AI Overviews companion guide covers the surface that still drives more impressions today.
Turning AI Mode off
There is no global switch that removes AI Mode from Google Search. It is a tab, so the simplest answer is to use the All tab instead, or to bookmark searches with &udm=14, which loads a plain link-only results page.
Publishers asking a different question, whether their own pages can be excluded from these surfaces, run into the Search Console preview controls. The trade-offs there are steeper than the setting implies.