Does Google AI Mode's two-response test change citations?

What is Google's AI Mode two-response test?

Google is testing a version of AI Mode that shows two responses to the same query and asks the searcher to pick the one they prefer. The prompt reads roughly “You're giving feedback on a new version of AI Mode. Which response do you prefer?”, and the two answers sit side by side. The point is feedback. Google wants real searchers with real queries to tell it which style of answer works better. This isn't the first sighting, either. A similar side-by-side test showed up in the wild in late 2025, and the format is back this October. Most people will never see it, because only some users are shown the test. If you depend on search traffic, though, it gives you a rare look at how AI answers get shaped.

If the side-by-side format looks familiar, that's because it's the standard way AI labs evaluate models. Public leaderboards rank chatbots by showing people two anonymous answers and asking them to vote. Google's own developer tools have offered a compare mode for testing Gemini models against each other for a while now. What's new is seeing the format inside a consumer search product, on live queries, where the answers link to real businesses. That raises the stakes. In a lab, a preference vote shapes tone, length and accuracy. In Search, each of the two answers also comes with its own list of cited sources. That's the part anyone with a website should care about if they want to be one of the links an AI answer recommends.

This is where it touches your business. AI Mode citations have quietly become some of the most valuable organic real estate on Google, because a cited link sits inside the answer the searcher actually reads. By showing two answers at once, Google is openly admitting what SEOs already suspected: a query has no single fixed answer, and no single fixed list of sources either. Every response is generated fresh, and each one can reach for slightly different pages. So the real question for a clinic, a law firm or an online store isn't which phrasing Google prefers this month. It's whether your content is strong enough to show up whichever version the searcher sees. The rest of this article covers how to make that happen.

Can the two AI Mode answers cite different websites?

Yes. Two AI Mode responses to the same query can cite different websites, because each answer is generated on its own, and the sources it uses depend on how that response was put together. AI Mode doesn't look up a stored answer. It splits a question into related sub-searches, which Google calls query fan-out, collects candidate pages, writes a response and attaches links to the passages that support it. Change the model version, the instructions behind the prompt or just the randomness in generation, and both the sub-searches and the supporting passages move. Ask ordinary AI Mode the same question twice and you'll often see different links. The two-response test just puts both outcomes on screen at the same time.

That doesn't make citations a lottery. Research comparing AI Mode with AI Overviews found that the two products usually agreed on what to say, even when they often cited different pages to say it. So the substance tends to stay the same while the sources shift around. The sites that keep showing up across versions tend to have a few things in common. They answer the question plainly near the top of the page. They cover the related subtopics a fan-out search would explore. And the facts they state match what other trustworthy sources say. Sites that appear once and then vanish usually got lucky with one well-matched sentence. A citation in one answer is a fluke. Citations across many answers are a pattern, and you can build that pattern on purpose.

What about the vote itself? It's tempting to think the winning answer gives its cited sites a permanent boost, but nothing Google has said supports that. The vote is feedback on the response as a whole: how clear it is, how useful, how well structured. Those signals shape how future versions of AI Mode write answers. They don't feed a scoreboard of websites. Indirectly, the vote still matters a lot. If searchers keep preferring answers that are specific, well sourced and easy to act on, the model will be tuned to write more of them. To write them, it needs pages that supply that kind of material. You don't win the vote by gaming one phrasing. You win it by being what good answers are built from.

How do you get cited in both AI Mode answers?

You get cited in different versions of an AI Mode answer by making your pages easy to quote however the response is worded. That means leading with direct answers, keeping your business facts identical everywhere and covering your topic deeply enough that any sub-search finds you. Start with answer-first passages. Under each heading, write one or two sentences that fully answer the heading's question, then add the detail. AI systems pull out passages, not whole pages, and a passage that stands on its own is much easier to cite than a conclusion buried in paragraph six. Say an accountant's page on quarterly tax deadlines opens with the actual dates and who they apply to. That opening works just as well in a short answer, a long one or a bulleted list.

Next, lock down the facts about your business. Your name, address, services, opening hours, published prices, credentials and founding details should read the same on your website, your Google Business Profile, directories and any other profiles you control. When two AI answers are built from different pages, consistency lets both of them describe you correctly and with confidence. Contradictions have the opposite effect. If one page says you serve three cities and another says five, a model may leave you out rather than risk a wrong claim. Add a clear about page, author bios with real qualifications and structured data where it fits. None of this is glamorous, but it's the basic plumbing that turns a business from findable into citable.

Then build depth. Because AI Mode splits a query into related searches, one page rarely covers every angle an answer needs. A group of connected pages does. If you run a physiotherapy clinic, a single page on lower back pain won't carry much weight. A set of pages on causes, exercises, when to see a specialist, recovery times and costs gives the model reasons to come back to you for several parts of the same answer. Link those pages to each other so crawlers can see how they fit together. Our opinion here is firm: twenty thorough, interlinked pages on your core topic will do more than two hundred thin posts chasing every keyword variation. Depth earns repeat citations, and volume without depth earns you nothing.

How can you check if your site appears in AI Mode answers?

The most reliable check is still a manual one. Ask AI Mode the questions your customers really ask, several times and worded different ways, and note which sources get cited. Keep a simple spreadsheet with the query, the date, whether you were cited and which competitors appeared. Run the same queries every couple of weeks. Responses vary, so one test tells you very little, but a month of checks shows whether you're a regular source or an occasional one. Use a private or logged-out browser window to cut down on personalization, and test on mobile as well as desktop. If you happen to get the two-response test, take a screenshot of both answers. Comparing their citation lists is the clearest view you'll get of how much your visibility varies.

Search Console helps, with some limits. Google counts AI Mode clicks and impressions in the regular Performance report for web search and doesn't break them out, so you can't filter for AI Mode alone. What you can do is look for longer, conversational queries showing up in your query data. You can also watch for pages that gain impressions without a matching rise in classic rankings. In GA4, AI Mode visits often look like ordinary Google organic traffic, so trends by landing page tell you more than referral labels do. If you want to know how other assistants treat you, we've written a separate guide on tracking Gemini referrals in GA4. Segmenting by landing page works the same way on every AI surface.

To go further than spot checks, score how citable your pages are before you even test them. Our free GEO score tool checks for the signals AI engines tend to reward, like answer-first structure and clear facts, and shows where a page falls short. Run it on your ten most important pages: the service pages, pricing explainers and FAQs customers read just before they contact you. Keep an eye on Google's interface experiments too, because each one changes where links appear on the screen. Google has also tested an AI Mode side panel next to the regular results, for example. Every layout change like that affects how many clicks a citation actually brings in, even when you're still being cited.

Why can't ads buy you a citation in AI Mode?

Ads can't buy a citation in AI Mode. The model picks its citations from organic content that supports the answer, while ads appear in separate slots labeled as sponsored. Google has been testing ads below AI Mode answers and even text ads inside them, but those ads are marked as sponsored and kept apart from the source links. Paying for one doesn't add your website to the pages the answer is built from. The two-response test makes that distinction more important, not less. Whichever answer a searcher prefers, its cited sources got there through their content, and no bid changes which pages the model trusts. Advertisers can rent space next to the answer. Only publishers can earn a place inside it.

The money is the real argument. Paid clicks stop the day the budget stops, and in competitive niches they tend to cost more every year. A well-built page keeps getting crawled, ranked and cited long after you publish it, and every new page on the same topic makes the others stronger. If you're comparing the two, our breakdown of whether SEO costs less than Google Ads this year walks through a 12-month comparison of cost per lead. We'll be blunt. For most service businesses, the money spent buying one more month of clicks would do more good building pages that AI answers can cite for years. Ads still make sense for launches and seasonal pushes, but you shouldn't build your marketing on them.

Trust points the same way. Google is giving searchers more say over which sources they see, including a test that lets people mark their preferred sources in AI Mode. Add side-by-side voting and the direction is clear: the system is being tuned around what readers find useful. You earn that by publishing regularly on the topics your customers care about. That's the work Artiql automates. It produces answer-first articles with consistent facts, in English, Hebrew and any other language you sell in, so your site keeps earning rankings and AI citations without a content team. Pricing is tailored to each business after a short discovery call, so the best next step is to book a short demo and see what it would look like for your site.

Frequently asked questions