AI Overviews vs AI Mode Are Two Different Engines (Only ~13.7% Citation Overlap)

Here is a fact that breaks a lot of SEO dashboards: the page Google cites in its AI Overview is usually not the page it cites in AI Mode.

They look like the same Google. They sit on the same results page. They often even reach the same conclusion. But internally they pull from almost completely different lists of sources.

In 2026, three separate studies put real numbers on this. The cleanest one comes from Ahrefs, who studied around 540,000 query pairs and found that only about 13.7% of cited URLs overlap between the two surfaces.

So if you are measuring only AI Overviews — or only AI Mode — you are flying half blind. This guide explains why the two diverge, shows you the data, and gives you working Python to measure both at once and compute your own overlap.

Two surfaces, two source pools

AI Overviews and AI Mode are two different products built on Google's AI search stack, and they behave like two different engines.

AI Overview (AIO) is the short AI summary that appears at the top of a normal results page. It answers fast and links to a handful of sources.

AI Mode is the full conversational experience — a longer, multi-paragraph answer that you can follow up on. It reads more like a research assistant than a snippet.

The temptation is to think of them as a "short version" and a "long version" of the same answer. The 2026 data says that is wrong. They agree on what to say far more than they agree on who to cite.

The 2026 data: ~13.7% and friends

Multiple independent studies in late 2025 and 2026 converged on the same headline: the citation overlap between the two surfaces is small.

The flagship study is from Ahrefs (published December 2025). They analyzed roughly 730,000 response pairs for meaning and around 540,000 query pairs for citations. Their finding:

• 86% average semantic similarity — 9 times out of 10, the two surfaces agreed on the actual answer.
• Only ~13.7% of cited URLs overlapped. Even limiting to each response's top-3 citations, overlap only rose to ~16.3%.
• AI Mode mentioned 3.3 entities on average versus 1.3 for AI Overviews — roughly 2.5x more.

Two corroborating studies back this up. SE Ranking independently measured ~10.7% URL overlap and ~16% domain overlap. And Victorious (1,540 queries) found that roughly 77% of unique cited domains appeared in only one surface — and that not a single query returned identical citations across both.

The exact percentage moves with methodology — Victorious counted overlap more loosely and landed higher — but every study points the same way: the two surfaces mostly cite different sources.

Why they cite different pages

The most likely reason is how AI Mode gathers sources in the first place: a heavier query fan-out.

Query fan-out means the engine takes your one question and quietly expands it into many related sub-queries, runs them in parallel, and merges passages from all of them. AI Mode does this more aggressively than AIO, so it samples a wider and more varied slice of the web.

That shows up in the numbers. AI Mode cites about 9.2 domains per query versus 7.7 for AI Overviews, and its answers are several times longer. More sub-queries plus more room for citations equals a different — and broader — source set.

We wrote a full breakdown of this mechanism in Google AI Mode query fan-out explained, and the citation-selection logic itself in how AI search selects citations. The short version: AIO answers a question; AI Mode researches a topic. Different jobs pull different evidence.

Citation absence differs too. Ahrefs found AI Overviews give no external citation about 11% of the time, while AI Mode does so only 3% of the time. AI Mode is the more reliable citation surface for brands that qualify — but only if you are tracking it.

Why measuring one is not enough

If a page cited in AIO is usually absent from AI Mode, then a single-surface report gives you a false picture of your AI visibility.

Say you win an AI Overview citation for "best project management software." You celebrate. But that same query in AI Mode might cite a completely different set of pages — and you might be nowhere in it. Your "win" covers maybe half the surface area.

The reverse hurts too. You might quietly dominate AI Mode while assuming you have no AI visibility at all, because your AIO-only dashboard shows zeros.

This is why the modern approach is dual-surface measurement. It is the same idea behind a proper AI Overview exposure audit and a share-of-voice tracker, just extended to cover both Google AI surfaces instead of one.

How to measure both surfaces with one API

The practical fix has three steps: pull AIO sources, capture AI Mode citations, then compute the overlap per keyword.

Serpent API gives you AI Overview sources directly. A full Google search response includes an aiOverview object with the answer text and a sources array of {url, title} objects. No proxy pool or headless browser to manage — the API handles access for you.

Here is the first call. Use the X-API-Key header (the old query-param style is deprecated):

import requests

API_KEY = "sk_live_your_key"
BASE = "https://api.apiserpent.com/api/search"

def aio_sources(query, country="us"):
    """Return the list of domains AI Overview cites for a query."""
    r = requests.get(
        BASE,
        headers={"X-API-Key": API_KEY},
        params={"q": query, "engine": "google", "country": country},
        timeout=60,
    )
    r.raise_for_status()
    data = r.json()
    aio = (data.get("results") or {}).get("aiOverview") or {}
    return [s["url"] for s in aio.get("sources", [])]

print(aio_sources("best project management software"))

That covers the AIO side. We walk through every field of this response in extract Google AI Overviews with an API.

For the AI Mode side, you capture citations the same way — query AI Mode for your tracked keyword and collect the cited URLs into a parallel list. Our companion guide track Google AI Mode citations in Python covers that capture in detail. The post you are reading now is the part neither of those covers: putting the two lists side by side and quantifying the gap.

Compute your own overlap (Python)

Once you have two lists of cited URLs per keyword, the overlap is simple set math. We compare at the domain level, because that is what brand visibility actually cares about.

from urllib.parse import urlparse

def domains(urls):
    out = set()
    for u in urls:
        host = urlparse(u).netloc.lower().replace("www.", "")
        if host:
            out.add(host)
    return out

def overlap_report(keyword, aio_urls, mode_urls, my_domain):
    a, m = domains(aio_urls), domains(mode_urls)
    shared = a & m
    union = a | m
    pct = round(100 * len(shared) / len(union), 1) if union else 0.0
    return {
        "keyword": keyword,
        "aio_domains": len(a),
        "mode_domains": len(m),
        "overlap_pct": pct,
        "in_aio": my_domain in a,
        "in_mode": my_domain in m,
        # the row that should page you:
        "diverged": (my_domain in a) != (my_domain in m),
    }

# Example: your domain shows in AIO but NOT AI Mode
rep = overlap_report(
    "best project management software",
    aio_urls=aio_sources("best project management software"),
    mode_urls=["https://asana.com/...", "https://en.wikipedia.org/..."],
    my_domain="yoursite.com",
)
print(rep)
if rep["diverged"]:
    print("ALERT: presence differs between AIO and AI Mode for", rep["keyword"])

The key field is diverged. It fires when you appear in exactly one surface — the single most actionable signal in dual-surface tracking. Run this nightly across your keyword set, store each row, and alert on any diverged == True.

Because both surfaces shift run to run, do not trust one snapshot. Track for several days and use a rolling presence rate (e.g. "cited in AI Mode in 5 of the last 7 days"). The same volatility discipline applies to ordinary rankings — see building a rank-drop alerting system.

Reading the results

Here is what a few rows of output look like once you let it run. This is illustrative, but the shape matches what the studies predict — low overlap, frequent single-surface presence.

The "Diverged?" column is where the value is. Row 1 says you have an AIO win you would lose if Google nudged the user into AI Mode. Row 2 says you are invisible in AIO despite a real AI Mode citation — a content or structured-data gap worth fixing. For the latter, structured data to get cited by AI and AI Mode ranking factors are your next reads.

A simple dual-surface playbook

Treat AI Overviews and AI Mode as two separate channels with their own scorecards. Here is the minimal routine.

• Track both, always. Every tracked keyword gets an AIO source list and an AI Mode citation list on the same schedule.
• Compute overlap per keyword. Expect it to be low (~10–20%). A keyword with high overlap is the exception, not the goal.
• Alert on divergence. The diverged rows are your priority queue — they show exactly where one surface is leaking visibility.
• Use rolling windows. Both surfaces are noisy; judge presence over 7 days, not one request.
• Fix to the surface. Missing from AI Mode? Strengthen topical depth and entities (its fan-out rewards breadth). Missing from AIO? Tighten the direct, snippet-style answer.

This sits naturally inside a broader strategy. If you are still mapping how SEO, AEO and GEO fit together, start with our GEO vs AEO vs SEO guide — dual-surface tracking is the measurement layer underneath all of it.

FAQ

Do AI Overviews and AI Mode cite the same sources?

Mostly no. Ahrefs studied roughly 540,000 query pairs and found only about 13.7% of cited URLs overlap between the two surfaces. SE Ranking found a similar 10.7% URL overlap. A page cited in one is usually not cited in the other.

Why do AI Overviews and AI Mode cite different pages?

AI Mode uses a heavier query fan-out: it breaks one question into many sub-queries and pulls from a wider, more varied source pool. It cites about 9.2 domains per query versus 7.7 for AI Overviews, so the two surfaces sample the web differently.

If I rank in AI Overviews, will I also show in AI Mode?

Not reliably. Studies found that around 77% of cited domains appear in only one surface. Visibility in AI Overviews does not guarantee visibility in AI Mode, so you must measure both separately.

How do I track citations in both AI Overviews and AI Mode?

Pull AI Overview sources from the search API response field results.aiOverview.sources, capture AI Mode citations the same way for your tracked queries, then compute the overlap percentage between the two domain sets per keyword and alert when they diverge.

Are AI Mode and AI Overview citations stable run to run?

No. Victorious reported that not a single query produced the exact same citations across both surfaces, and each surface also shifts between runs. Track over several days and use rolling windows rather than a single snapshot.

Does Serpent API return AI Overview sources?

Yes. A full Google search response includes results.aiOverview with the answer text and a sources array of {url, title} objects, so you can parse cited domains directly without managing scraping yourself.