Translation note. This is the English version of the approved Portuguese article. The Portuguese version remains the evidence record for original AI replies. Quotes rendered in English below are identified as translations; proper names, source URLs, dates, denominators and limitations are retained.
They are different products within Google Search. AI Overview is the generated summary block at the top of the classic results page, where the ten links remain below. AI Mode is a separate conversational tab, where the links page disappears and the entire experience becomes a conversation. One is added to search; the other is a different form of search.
I write as Mateus Gomes, operator of murmur.marketing. Our GEO model combines proprietary measurement software with specialist strategy and execution to grow brands’ Share of Voice (SOV) in AI answers. This article distinguishes Google Search surfaces and states exactly which ones were covered by the historical measurement described below; its figures are a dated baseline, not a claim about current performance.
Summary
- AI Overview (in Portuguese, visão geral criada por IA) is a generated block at the top of the results page. The links remain below it and the interaction is a single turn.
- AI Mode (Modo IA) is a Search tab, alongside Images and News, with a conversational experience: it accepts follow-up questions and maintains a session.
- Both use the retrieval technique Google calls query fan-out: a question is split into subtopics and many searches run at once.
- Google requires no special optimization for either: a page must be indexed and eligible for a Search snippet. Search generative AI control defaults to “Include”; this permits inclusion but does not guarantee display.
- Since August 31, 2026, separate performance reports are available for generative features in Search and Discover. Search includes AI Overviews and AI Mode; reports show impressions, URLs, countries and dates, plus devices for Search only. They do not show answer text, prompt-level SOV or SOV comparable across engines.
- Neither is the Gemini app. These are three distinct surfaces, and Google discusses them separately.
- This pipeline captures AI Overview only. It does not measure AI Mode or the Gemini app because it has no collector for either.
AI Overview: the block above the results
AI Overview is an AI-generated summary above organic results, with supporting links to pages that informed it. Its defining feature is not the text alone but coexistence: the page is still a results page, with the ten links below, and the block sits above them.
That has three practical consequences for publishers. The block and the organic result compete for the same screen, so a page can inform the block without being clicked, or be clicked without being a source. The interaction is single-turn, with no session, making each display comparable over time. Most important for measurement, the block can be captured reproducibly in a browser because it begins with an ordinary search query.
AI Mode: the conversational tab
AI Mode is a dedicated Search tab. Google describes it as an end-to-end AI-search experience with more advanced reasoning and explains its mechanics in its announcement: [English translation of Google’s wording] “AI Mode uses our query fan-out technique, breaking your question into subtopics and issuing a multitude of queries simultaneously on your behalf” (Google blog). It was released to all users in the United States on 20 May 2025, at that year’s Google I/O, after a Labs testing period. The same announcement says a custom version of Gemini 2.5 powers both AI Mode and AI Overviews.
For Brazil, the latest official record I reviewed is dated 10 June 2026 on the Google Brasil blog. It describes Search’s Modo IA receiving agent capabilities, including restaurant reservations through a conversational experience, and names “the Gemini app and AI Mode” as different things on the same page (Google for Brasil 2026). The distinction among the three surfaces is therefore how the manufacturer itself describes them.
Two behaviors matter when measuring. Session: AI Mode accepts follow-up questions, so the second answer depends on the first and an isolated prompt is insufficient for reproducible capture. Depth: deep-research mode can issue hundreds of searches for one report, changing the candidate-document set relative to a summary block.
The five things people call “Google AI”
| surface | what it is | where it appears | interaction | captured by this pipeline? |
|---|---|---|---|---|
| AI Overview | generated summary block | top of the classic results page, with links below | single turn | yes — this campaign calls it google |
| AI Mode | complete conversational search | its own tab, alongside Images and News | session with follow-ups | no — there is no collector |
| organic results | ordered list of links | the same page, below the block | single turn | no — this is an SEO surface |
| Gemini app | conversational assistant outside Search | its own web and mobile app | session | no — there is no collector |
| ChatGPT · Claude · Perplexity | generative engines from other providers | their own products | session | yes, all three |
The Gemini-app row is included because it is the most common and costly confusion. Reading a report about “Google” as if it described the assistant assigns a number from one surface to another.
What they share — and what Google says
The Google Search Central documentation on AI features, in the version reviewed here (updated 10 December 2025), makes two points that matter more than market speculation.
Eligibility. [English translation of Search Central’s wording] “There are no additional requirements to appear in AI Overviews or AI Mode, nor are there special optimizations required.” A page needs to be indexed and eligible for a Search snippet (Search Central). Search generative AI control, available in Search Console since August 31, 2026, defaults to “Include.” That allows content to be considered; it does not guarantee display or act as a ranking signal or special optimization (control documentation).
Reporting. Since August 31, 2026, Search Console has separate performance reports for generative features in Search and Discover. The Search report includes AI Overviews and AI Mode and shows impressions, URLs, countries and dates; device data is available for Search, not Discover. It does not show answer text or prompt-level data and cannot calculate SOV comparable across engines. It helps track visibility and traffic, but does not replace capturing answers to classify recommendation, citation, mention or absence (Google Search Central).
The two also share query fan-out: the query a person typed is not the full set of queries the system executes. Capturing an answer to one question does not capture the searches that produced it. A page may have been retrieved for a subtopic nobody typed.
Why this distinction determines how a report is read
The reason is observational: these surfaces behave differently on the same data, on the same day.
In my 2026-08-06 campaign, the question [English translation of the Portuguese prompt] “what is the best company for GEO in Brazil?” ran five times in each of four engines, producing 20 captures. Under a deterministic text rule applied to a declared list of 29 names, Conversion appeared in 5 of 5 ChatGPT repetitions, 4 of 5 Perplexity repetitions, 0 of 5 Claude repetitions, and 0 of 5 Google AI Overview repetitions. AI Overview and ChatGPT gave opposite answers to the same question.
For the hiring-question family, denominator 48 (12 questions × 4 engines, one repetition each, under the same rule and list of 29), the most seen name by engine was: Brasil GEO, 8 in ChatGPT; Brasil GEO, 7 and Criamente, 7 in Claude; GeoStack, 8 in Perplexity; and Conversion, 9 in Google AI Overview. These are absolute counts, never percentages. In 8 of the 48 captures, none of the 29 names appeared. The four engines disagree on the leading name.
The historical sample also shows why one snapshot should not be treated as a verdict on a provider: AI Overview and ChatGPT produced different leaders for the same question. In a public-surface scan on that date, Conversion had 1,366 URLs in sitemap.xml; nine were under /geo/, or 0.7% of the site. That observation does not establish why it appeared, nor does it predict how another brand will perform.
An additional historical example comes from Fly Vet, a veterinary-marketing agency operated by the same group: Google used its site as a source in 81 of 100 questions measured on 2026-06-27/28 under the citation_kind rule, while the study recorded no recommendation in that sample. ChatGPT recommended it in 30 of 100 and Claude in 41 of 100. These are dated results for a specific question set, not a current performance claim.
What I do not measure
The engine my campaign calls google is AI Overview. The pipeline does not measure AI Mode, and it does not measure the Gemini app; it has no collector for either. This is an instrument debt, not an elegant methodological choice. Any figure I give for “Google” describes the summary block on the results page, not the conversational tab or assistant.
A provider who presents “Google visibility” without saying which surface was captured sells a number the buyer cannot reconstruct. With three distinct surfaces under one brand, the chance of understanding the wrong one is high. See How to tell whether a GEO agency has real results.
What I measured about this question
On 2026-08-06 I ran 100 questions in Portuguese across four engines — ChatGPT, Claude, Perplexity and Google AI Overview — producing 800 browser captures with screenshots in 800 of 800, all judged. murmur.marketing appeared in zero. In the 34 method, taxonomy and evidence questions: 0 of 136. This question is one of them, with a score of 0 of 4.
One of the 800 captures failed: in one case the engine built an interactive artifact instead of answering. I kept it in the denominator. Removing an isolated capture invites cherry-picking what counts, a decision readers cannot see in a final report.
Frequently asked questions
What is the difference between Google AI Overview and AI Mode?
AI Overview is a generated summary block at the top of the classic results page, with the ten links continuing below and a single-turn interaction. AI Mode is a dedicated Search tab alongside Images and News, with a conversational session that accepts follow-up questions. Both use query fan-out. AI Mode was released to all users in the United States on 20 May 2025, at that year’s Google I/O.
Is AI Mode the same as the Gemini app?
No. AI Overview, AI Mode and the Gemini app are distinct. Google’s 10 June 2026 Google for Brasil announcement names the latter two side by side as different things. AI Overview and AI Mode are in Search; Gemini is an assistant with its own app outside Search.
Is there a specific optimization for appearing in AI Overview or AI Mode?
According to Google, no. A page must be indexed and eligible for a Search snippet; there is no special markup, registration or optimization. Search generative AI control defaults to “Include,” allowing links and content to be considered, but it does not guarantee display or affect ranking. The fundamentals remain server-delivered HTML, crawlability, a sitemap and content that answers the question.
Does Search Console show AI Mode data separately?
Yes. Since August 31, 2026, Search Console has separate performance reports for generative features in Search and Discover. The Search report includes AI Overviews and AI Mode and shows impressions, URLs, countries, dates and devices (the device dimension is Search only). It does not show answer text or prompt-level data, nor calculate SOV comparable across engines. To know whether a brand was recommended, cited, mentioned or absent, capture the answer with a denominator and date.
If AI Mode is not captured, how can I know whether my brand appears there?
Search Console now shows aggregated AI Mode performance—impressions and URLs among its dimensions—but not the answer text, associated prompt or whether a brand was recommended or merely cited. To observe that, open the tab, preserve the answer with a screenshot and date, and repeat captures: in a session, each answer depends on the previous one. Treat a single observation as an anecdote until enough repetitions exist. In this page’s measurement, the same question, run five times in each engine on the same day, produced a name in 5 of 5 ChatGPT repetitions and 0 of 5 Google AI Overview repetitions.
Who wrote this, and the declaration of interest
I am Mateus Gomes, operator of murmur.marketing, a Brazilian GEO operation combining proprietary measurement software with specialist strategy and execution to grow SOV. The 2026-08-06 campaign was a dated baseline, run before the current guide collection; the surfaces included in that campaign and those not measured are stated explicitly in the article.
Conclusion
AI Overview and AI Mode are two Google products, not two names for one thing. The first is a single-turn summary block above a results page, coexisting with links. The second is a conversational tab with a session, deep research and the same query-fan-out technique. Neither is the Gemini app. Google says no special optimization is required and does not report either separately in Search Console. Measuring them therefore requires capturing the response as it appeared. In my study, AI Overview and ChatGPT returned opposite answers to the same question on the same day. To discuss measurement for your case, contact Mateus Gomes on LinkedIn.