---
title: "AI Mode vs AI Overviews: how different are the sources?"
description: "Same 400 searches: AI Mode and AI Overviews shared 13.9% of cited URLs, and AI Mode cited 16.3% of top-10 pages to the AI Overview’s 29.7%."
canonical: "https://underneath.agency/research/ai-mode-vs-ai-overviews-study"
published: 2026-09-26
updated: 2026-10-08
publisher: "Underneath (https://underneath.agency/agent)"
entity: "https://underneath.agency/.well-known/entity.json"
---
Research · AI search

# AI Mode vs AI Overviews: how different are the sources?

Google now answers searches with AI in two places: the AI Overview at the top of the normal results page, and AI Mode, a separate conversational tab. We asked both the same 400 US searches on the same day, 26 September 2026, and compared what each one cited. This version treats the two as separate search surfaces rather than one “Google AI search”. It splits visibility into stages (does the surface answer, does it cite, does a ranking page get cited, which source comes first, does a source cited on one surface also appear on the other) and reports how far each difference holds search by search.

## The short version

1. The two surfaces are different environments. Where both cited sources (236 of 400 searches), they shared a mean 13.9% of cited URLs (Jaccard 0.139, 95% interval 0.111 to 0.163) and 24.2% of cited domains excluding Google’s own. 30.5% of these searches had no cited URL in common, and they led with the same source in 20.9%.
2. Ranking on page one carries over to AI Overviews far better than to AI Mode. Of the 2,118 URLs in the organic top 10 on searches that showed both surfaces, the AI Overview cited 29.7% and AI Mode 16.3%. For positions 1 to 3 the figures were 42.3% and 26.0%.
3. AI Overviews cite more forums, social and video: 18.8% of their citations on the matched searches against 11.2% for AI Mode, and the AI Overview had the higher share in 56.4% of searches against 20.3% the other way.
4. Correction to version 1.0: most of AI Mode’s heavy use of Google’s own pages comes from searches that show no AI Overview at all. On those 147 searches, 69.3% of AI Mode’s citations were Google pages. On the matched searches the Google-owned share was 20.4% for AI Mode and 14.8% for AI Overviews, a gap of 1.4 times rather than 2.7, with wide, overlapping intervals.
5. AI Mode answered all 400 searches and cited a source in 94.0%; an AI Overview appeared on 63.2% (95% interval 52.2% to 74.2%). Counting every search where AI Mode cited something, the AI Overview cited at least one of the same URLs in 43.6%.

## Research questions

| Question | Answered here? |
|---|---|
| RQ1. How often does each surface answer and cite, counting every search? | Yes |
| RQ2. How much do the cited sources overlap, and does that depend on the measure? | Yes, with sensitivity checks |
| RQ3. Given that a page ranks in the top 10, how likely is each surface to cite it? | Yes |
| RQ4. Do the two surfaces lead with the same source? | Yes |
| RQ5. Does the gap depend on industry, intent or popularity? | Yes, with a clustered model |
| RQ6. Is the gap stable across runs, dates, wordings and follow-up turns? | No: one run per surface and search |

## Visibility in stages

Being “visible in Google’s AI” is not one number. We measure five stages separately for each surface. Three further stages are named here because they matter, but this study does not measure them.

| Stage | Question | Measured? |
|---|---|---|
| Appearance | Does the surface answer the search? | Yes |
| Citation | Does the answer cite any source? | Yes |
| Ranking to citation | Is a page that ranks in the top 10 cited? | Yes |
| Prominence | Is the source the first one cited? | Yes |
| Transfer | Is a source cited on one surface also cited on the other? | Yes |
| Attribution | Does the cited page support the sentence it is attached to? | No |
| Influence | Did the source shape the answer? | No |
| User outcome | Did anyone click, trust or act on it? | No |

## What we measured

We took 400 keywords from the 800-keyword US sample of our [AI Overview frequency study](https://underneath.agency/research/ai-overviews-frequency-study), 50 per industry (20 of the highest-volume and 30 of the randomly drawn keywords), spread over 32 seed topics. For each one we recorded the regular Google results page (with any AI Overview) and the AI Mode answer, both from Google.com, United States, English, desktop, on 26 September 2026. From each AI answer we collected every cited link in reading order, normalized the URLs and compared the two sets.

Every search stays in the results. A search with no AI Overview, or an answer with no citation, is recorded as an outcome rather than dropped. Keywords that share a seed topic (“pizza”, “car insurance”) are not independent, so every 95% interval on this page comes from resampling the 32 seed topics, not individual keywords.

## Findings

### Every search, by outcome

| Outcome (400 searches) | Searches | Share |
|---|---|---|
| Both shown and both cited | 236 | 59.0% |
| AI Mode only (no AI Overview shown) | 147 | 36.8% |
| Both shown, only the AI Overview cited | 13 | 3.2% |
| Both shown, only AI Mode cited | 4 | 1.0% |

AI Mode answered every search and cited at least one source in 94.0% of them. The AI Overview appeared on 63.2% and cited a source in 62.2%. Of the 376 searches where AI Mode cited something, the AI Overview cited at least one of the same URLs in 43.6% (95% interval 33.2% to 54.8%) and one of the same non-Google domains in 47.6%.

### The two answers rarely cite the same pages

| Measure (236 searches where both cited) | Result | 95% interval |
|---|---|---|
| Mean URL overlap (Jaccard) | 0.139 | 0.111 to 0.163 |
| Searches with no cited URL in common | 30.5% | 24.4% to 38.1% |
| Mean domain overlap, excluding google.com | 0.242 | 0.192 to 0.281 |
| Share of AI Mode’s URLs also in the AI Overview | 28.4% | 23.9% to 32.6% |
| Share of the AI Overview’s URLs also in AI Mode | 22.2% | 17.0% to 27.0% |

The median URL overlap was 0.1, and 23.2% of searches had no cited domain in common.

The finding does not depend on how overlap is measured. The AI Overview cites more sources (median 8 against 4.5 on these searches), which pulls Jaccard down. The overlap coefficient, which divides by the smaller set instead, is 0.332: even then, two-thirds of the shorter list is not in the longer one. Dropping Google’s own URLs gives 0.138, counting google.com in the domain comparison gives 0.272, and comparing only the first three sources each surface cites gives 0.164.

### They lead with different sources

On the 235 searches where both surfaces cited a source inside the answer, we compared the first source each one cited.

| First source | Result | 95% interval |
|---|---|---|
| Same URL | 20.9% | 14.1% to 26.6% |
| Same domain | 34.0% | 24.1% to 44.5% |
| Overview’s first, cited by AI Mode | 41.7% | 32.7% to 49.6% |
| AI Mode’s first, cited by Overview | 42.6% | 34.2% to 49.8% |

A page that is the lead source on one surface is missing entirely from the other surface’s answer more often than not.

### Ranking pages are cited far less in AI Mode

Organic rankings are the one stage before citation we can observe. On the 253 searches that showed both surfaces, we followed each of the 2,118 URLs in the organic top 10.

| Top-10 URLs cited | AI Mode | AI Overviews |
|---|---|---|
| All top-10 URLs | 16.3% | 29.7% |
| Positions 1 to 3 | 26.0% | 42.3% |
| Positions 4 to 10 | 11.0% | 22.7% |

Intervals for all top-10 URLs: AI Mode 12.1% to 21.2%, AI Overviews 26.9% to 32.7%. The intervals do not overlap.

Of those 2,118 ranking URLs, 10.0% were cited by both surfaces, 19.7% only by the AI Overview, 6.3% only by AI Mode and 63.9% by neither. Seen from the citation side, 19.9% of AI Mode’s citations on the matched searches were top-10 pages, against 29.7% for AI Overviews, and the AI Overview had the higher top-10 share in 50.4% of searches against 30.1% the other way.

### Source types on the same searches

On the 236 searches where both cited, AI Mode made 1,728 citations and the AI Overviews 2,013. “Other websites” are brands, publishers and blogs.

| Source type | AI Mode | AI Overviews |
|---|---|---|
| Other websites | 62.8% | 59.0% |
| Google’s own pages | 20.4% | 14.8% |
| Forums, social, video | 11.2% | 18.8% |
| Reviews and directories | 1.9% | 2.6% |
| Government | 1.4% | 0.9% |
| Education | 1.2% | 1.5% |
| Marketplaces, retailers | 0.6% | 1.4% |
| Wikipedia | 0.5% | 0.8% |

The forum, social and video gap holds search by search: the AI Overview had the higher share in 56.4% of searches, AI Mode in 20.3%, and they tied in 23.3%. The Google-owned gap is smaller and less consistent. In 72.5% of searches the two had the same Google share (usually none), AI Mode was higher in 17.8% and the AI Overview in 9.7%. The pooled shares have wide intervals (AI Mode 8.7% to 36.5%, AI Overviews 7.3% to 24.1%) because Google pages cluster in a few local and shopping topics.

AI Mode also spreads its citations more widely. Counting each domain once per answer, its effective number of domains was 93.5 against 58.8 for AI Overviews, even though it drew on fewer distinct domains (512 against 602).

### Where the Google-owned citations come from

On the 147 searches with no AI Overview, AI Mode made 1,155 citations, and 69.3% of them were Google pages, mostly the “searchviewer” listing pages for local businesses and products. Those searches are concentrated in franchises (20.4% of them), healthcare, home services and legal (17.7% each). On the same searches only 7.8% of AI Mode’s cited URLs ranked in the top 10. Version 1.0 compared AI Mode across all 400 searches with AI Overviews on the searches that showed one, so it mixed a surface difference with this difference in search mix.

### The gap depends on the industry, not the intent

| Industry | Searches where both cited | Mean URL overlap |
|---|---|---|
| Hospitality and travel | 33 | 0.221 |
| Financial services and insurance | 44 | 0.184 |
| B2B software and technology | 46 | 0.144 |
| Healthcare and dental | 23 | 0.136 |
| Franchises and multi-location brands | 18 | 0.114 |
| Legal and professional services | 21 | 0.099 |
| Retail and ecommerce | 30 | 0.075 |
| Home and local services | 21 | 0.061 |

In a model of per-search URL overlap with industry, search intent and sample stratum together, and standard errors clustered by seed topic, industry mattered (p < 0.001) but intent (p = 0.624) and highest-volume versus random keywords (p = 0.494) did not. Overlap was highest in travel (95% interval 0.187 to 0.269) and lowest in home and local services (0.039 to 0.094). The model explains little of the variation (R² 0.138): most of the difference is search by search.

## How this compares with other studies

| Source | Sample and date | Figure |
|---|---|---|
| This study | 236 US search pairs, September 2026 | Mean URL overlap 13.9%; domain overlap 24.2% |
| Ahrefs | 540,000 query pairs, US, September 2025 | 13.7% of cited URLs matched |
| SE Ranking | 10,000 US keywords, August 2025 | URL overlap 10.7%; domain overlap 16% |
| SE Ranking | Same | AI Mode against organic top 10: 14% of URLs, 21.9% of domains |
| seoClarity | 1,000 transactional US queries, October 2025 | 19% of AI Mode citations from the top 20 rankings |

A year on, the two products still diverge about as much as Ahrefs and SE Ranking measured in 2025: our URL overlap is within a few points of both, and our domain overlap is higher than SE Ranking’s. Each study defines overlap slightly differently, so small gaps are not meaningful. Our share of AI Mode’s cited URLs that rank in the top 10 across all 400 searches (15.1%) is close to SE Ranking’s 14%.

Sources: [Ahrefs](https://ahrefs.com/blog/ai-overviews-vs-ai-mode/); [SE Ranking](https://seranking.com/blog/ai-mode-research/); [seoClarity](https://www.seoclarity.net/ai-mode-rankings-overlap).

## Observed, inferred and unknown

**What we observe.** For the same search on the same day, AI Mode and the AI Overview cite mostly different pages, lead with different sources, and turn organic rankings into citations at very different rates. AI Overviews lean more on forums, social and video. AI Mode appears on every search, including the local and shopping searches where no AI Overview is shown, and on those it cites Google’s own listings heavily.

**What we infer.** “Google AI search” is not one retrieval environment. A result measured on one surface should not be assumed to hold on the other, and visibility figures that blend them hide which one a brand is winning.

**What remains unknown.** Whether the gap is stable across repeated runs, days and rewordings (part of it may be ordinary run-to-run variation). Whether AI Mode behaves differently as a follow-up in a conversation. Why each surface picks the sources it does: this study observes citations, not the retrieval behind them. Whether the cited pages support the sentences they are attached to, whether they shaped the answer, and whether anyone clicked.

## What this means

The points below are our interpretation. They follow from the findings but were not tested.

- **Track the two surfaces separately.** Being cited in the AI Overview says little about being cited in AI Mode for the same search. Report each on its own.
- **Ranking well is a better route into AI Overviews than into AI Mode.** A top-3 page was cited by the AI Overview about four times in ten and by AI Mode about one time in four.
- **For searches without an AI Overview, Google’s own listings are the front door.** Where AI Mode is the only AI answer, most of its citations point to Google’s listing pages, so the information Google holds about a business (its Business Profile, product feed and reviews) is likely to matter most there.
- **AI Overviews reward video and community presence more.** The forum, social and video share is higher in AI Overviews on most searches, not just on average.

## Where this sits in GEO research

Research on generative engine optimization began by asking whether changing a page changes how often an AI answer uses it. Later work compared the sources that different AI search systems draw on, and argued that visibility should be measured as separate quantities (retrieval, citation, prominence, attribution, stability) rather than one score, and that search surfaces within one engine can differ as much as engines do. This study measures that last point directly for Google: same engine, same searches, same day, two surfaces. It is observational. It shows where the two surfaces differ, not what would change a page’s visibility on either, and it does not yet measure stability. A second collection of the same searches is planned, and we will add run-to-run figures to this page when it is analyzed.

## Methodology

- **Keywords:** 400 of the 800 US keywords in our AI Overview frequency study, 50 per industry (the first 20 of the 40 highest-volume keywords and the first 30 of the 60 randomly drawn keywords, in sample order), from 32 seed topics.
- **Collection:** DataForSEO SERP API, Google organic (with asynchronous AI Overviews) and Google AI Mode, Google.com, United States, English, desktop, 26 September 2026, one request each.
- **Citations:** every cited link in each AI answer (links attached to each paragraph in reading order, then the reference list), normalized (tracking parameters, fragments, “www.” and trailing slash removed) and de-duplicated. The first source is the first link cited inside the answer body.
- **Overlap:** Jaccard similarity (shared divided by all distinct sources) per search, averaged over the 236 searches where both answers cited at least one source; domain overlap excludes google.com unless stated. Sensitivity checks use the overlap coefficient, overlap without Google URLs and the first three sources.
- **Ranking to citation:** the share of organic top-10 URLs that each surface cites, on the 253 searches that showed both.
- **Uncertainty:** 95% bootstrap intervals resampling the 32 seed topics (2,000 resamples, seed 20260926).
- **Moderators:** OLS of per-search URL overlap on industry, keyword intent (DataForSEO’s label) and stratum, with standard errors clustered by seed topic; joint Wald tests.
- **Update schedule:** a second collection date with our week-over-week volatility study; monthly after that.

## Limitations

- One run per surface and search on one date. Both surfaces vary from run to run (see our [consistency study](https://underneath.agency/research/ai-recommendation-consistency-study)), so part of the difference may be ordinary variation.
- Observational only. The organic top 10 is the only stage before citation we can see; a page can be retrieved without ranking there.
- Citation is not attribution or influence. We did not check whether cited pages support the claims attached to them, or measure clicks.
- AI Mode was queried as a first question, not as a follow-up in a conversation.
- Source types use a fixed domain list; intent is DataForSEO’s keyword label. Small industry groups (18 to 46 searches) have wide intervals.

## What changed in version 1.1

Version 1.0 (26 September 2026) compared the two surfaces on overlap and source mix. Following an external review, version 1.1 (28 September 2026) re-analyzes the same 400 searches with no new queries. It keeps every search as an outcome, splits visibility into stages, adds prominence, ranking-to-citation and transfer measures, adds intervals clustered by seed topic, sensitivity checks and a moderator model.

One headline changed. Version 1.0 reported that 39.9% of AI Mode’s 2,894 citations were Google pages, 2.7 times the 14.6% in AI Overviews, and that AI Overviews cited forums, social and video 2.4 times as often (19.2% against 8.1%). Those figures compared AI Mode on all 400 searches with AI Overviews on the searches that showed one. On the same searches the ratios are 1.4 and 1.7. Version 1.0’s overlap figures are unchanged.

## Data and downloads

- Per-search results for version 1.1 (outcome, overlap, first sources, top-10 citations): [s6_searches_v11.csv](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/s6_searches_v11.csv) and [JSON](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/s6_searches_v11.json)
- Every statistic on this page: [stats.json](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/stats.json)
- Machine-readable methodology: [methodology.json](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/methodology.json)
- Chart: [top-10 URLs cited by each surface](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/ranking-to-citation-ai-mode-vs-overviews.svg)
- Version 1.0 data and statistics: [s6_ai_mode_vs_overview.csv](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/s6_ai_mode_vs_overview.csv) and [stats_v1.0.json](https://underneath.agency/research-data/ai-mode-vs-ai-overviews-study/stats_v1.0.json)

The data is free to reuse with attribution (CC BY 4.0).

To cite: Underneath. (2026). *AI Mode vs AI Overviews: how different are the sources?* (Version 1.1). Underneath Research. https://underneath.agency/research/ai-mode-vs-ai-overviews-study

## Frequently asked questions

### Do Google AI Mode and AI Overviews cite the same sources?

Mostly not. For the same search on the same day, they shared a mean 13.9% of cited URLs, 30.5% of searches had no cited URL in common, and they led with the same source in 20.9%.

### Does ranking on page one get you cited in AI Mode?

Less often than in AI Overviews. Of URLs ranking in the organic top 10, AI Mode cited 16.3% and the AI Overview 29.7% on the same searches; for positions 1 to 3, 26.0% against 42.3%.

### Does AI Mode cite Google’s own pages more than AI Overviews?

Mainly on searches that show no AI Overview, such as local and shopping searches, where 69.3% of AI Mode’s citations were Google pages. On searches where both cited sources, the shares were 20.4% and 14.8%.

### Does every search have an AI Mode answer?

In our sample, yes: AI Mode answered all 400 searches, while an AI Overview appeared on 63.2% of them.

## Related research

- [Do AI Overviews cite the pages that rank?](https://underneath.agency/research/ai-overview-citations-study)
- [How often do AI Overviews appear?](https://underneath.agency/research/ai-overviews-frequency-study)
- [Do ChatGPT, Gemini, Perplexity and Claude agree on brands?](https://underneath.agency/research/ai-assistants-brand-agreement-study)
- [What pages cited by AI Overviews have in common](https://underneath.agency/research/ai-overview-cited-pages-study)

## Related guides

- [How much traffic would AI Mode as Google’s default cost you?](https://underneath.agency/resources/ai-mode-default-traffic-loss)
- [If we rank on Google, will we show up in AI Overviews?](https://underneath.agency/resources/does-ranking-on-google-get-you-into-ai-overviews)
- [Do AI search engines cite the same websites as Google?](https://underneath.agency/resources/do-ai-search-engines-cite-the-same-sites-as-google)
- [How much control do we have when Google changes how search results look?](https://underneath.agency/resources/google-search-design-changes-traffic-risk)
- [Does social media content help your brand show up in AI search?](https://underneath.agency/resources/does-social-media-help-ai-search-visibility)

---

This is the Markdown twin of https://underneath.agency/research/ai-mode-vs-ai-overviews-study. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.