PrismCrawl
← Back to blog

Published

Tracking AI Overview and AI Mode Citations When the Answers Keep Changing

Classic rank tracking has one convenient property: a page's position usually holds from one day to the next. AI answers don't work that way. The same question can cite different sources an hour later, and Google's two AI surfaces, AI Overviews and AI Mode, rarely cite the same pages.

If you report AI visibility from a single check, you are reporting noise. This guide covers what the research says about citation stability and how to measure it in a way that holds up.

What the research shows

Ahrefs has published the most useful public data on this.

AI Overview citations used to follow rankings. In a July 2025 study of about one million keywords and 1.9 million cited links, Ahrefs found that 76.1% of pages cited in AI Overviews ranked in Google's top 10 for the same query. Ranking first still didn't guarantee a citation. A companion study found pages ranking first were cited about half the time.

Citations churn quickly. In an October 2026 study, Ahrefs asked five AI tools the same questions every day for 20 days and followed about 37,500 citations. About 60% of citations in Google AI Mode were gone by day two, and about 50% in AI Overviews. About 75% of the citations that dropped came back at some point within the 20 days.

AI Mode and AI Overviews disagree. A December 2025 Ahrefs study of September 2025 US data compared the two surfaces for the same queries. Their cited URLs overlapped only 13.7% of the time, even though the answers were 86% similar in meaning. AI Overviews also had no citations in 11% of responses, against 3% for AI Mode.

Many cited pages don't rank at all. The October 2026 study used a different measure: whether a cited page ranked in Google for the same question at all. Around 75% of pages cited by AI Mode didn't, and nearly half of AI Overview citations didn't either. Because the measures differ, this isn't directly comparable with the 76.1% figure, but both point the same way: organic rank is only part of the picture.

Taken together, the studies show that a single check tells you very little, AI Mode and AI Overviews need separate tracking, and organic rank alone doesn't predict AI visibility.

Measure how often you're cited

The fix is to treat citations like a sample. Instead of asking "are we cited today?", ask "how often are we cited across many checks?"

  1. Check each question several times. Spread checks across the day or across days.
  2. Record every cited domain and URL for each check, along with the location, device, and time.
  3. Report citation share: the percentage of checks in which your domain was cited.
  4. Track domains and URLs separately. A domain can stay visible while the cited page changes.
  5. Keep AI Overviews and AI Mode apart. They are different products with different sources.

A citation share of 40% over 20 checks means much more than "cited" or "not cited" on a single day.

Collecting AI Overview and AI Mode citations

With PrismCrawl, AI Overviews come back in a normal Google Search response when Google shows one. AI Mode uses the same endpoint with udm: 50. In both cases the answer and its source links are in the ai_summary item of serp_features:

import os
from collections import Counter
from urllib.parse import urlparse

import requests

API = "https://api.prismcrawl.com/v1/google/search"
HEADERS = {"x-api-key": os.environ["PRISMCRAWL_API_KEY"]}


def cited_domains(query: str, ai_mode: bool = False) -> set[str]:
    body = {"query": query, "gl": "us", "hl": "en"}
    if ai_mode:
        body["udm"] = 50
    content = requests.post(API, json=body, headers=HEADERS, timeout=120).json()["data"]["content"]
    domains = set()
    for feature in content.get("serp_features") or []:
        if feature.get("type") == "ai_summary":
            for link in feature.get("links") or []:
                host = (urlparse(link.get("url", "")).hostname or "").removeprefix("www.")
                if host:
                    domains.add(host)
    return domains


def citation_share(query: str, checks: int = 10, ai_mode: bool = False) -> Counter:
    counts = Counter()
    for _ in range(checks):
        counts.update(cited_domains(query, ai_mode))
    return Counter({d: n / checks for d, n in counts.items()})

In production, spread the checks over time instead of running them back to back, and store each response so you can recompute shares later. Our TypeScript tracker for AI Overview citations shows a complete storage setup, and the brand visibility guide covers reporting.

Pitfalls that skew AI visibility data

  • Missing answers. AI Overviews often load after the main page through a separate request. A data source that doesn't wait for it records no AI Overview. See Google's JavaScript requirement.
  • Layout changes. On November 2, 2026, Google moved AI Overviews for many brand queries to the right column on desktop. A parser that only looks in the main column misses them.
  • Search Console blending. Search Engine Roundtable reports that all links in an AI Overview share one position in Search Console, so it can't show where in the answer you were cited.
  • Bad underlying SERPs. A poisoned results page can carry a changed AI Overview too. Validate the page before trusting its citations.
  • Location. AI answers are localized like other results. Compare checks from the same location only.
  • Sponsored cards. Google now places ads inside AI Overviews and AI Mode. Count a sponsored card as an ad, never as a citation.

What to report

A useful AI visibility report shows, for each tracked question:

  • Citation share in AI Overviews and in AI Mode, separately
  • How often an AI Overview appeared at all
  • The top cited competing domains
  • The trend in citation share over weeks

Daily citation movements are mostly noise. Weekly shares over many checks are where real changes show up.

Frequently asked questions

Why do AI Overview citations change every day?

AI answers are generated for each request and draw on sources that vary over time. In an October 2026 Ahrefs study that followed about 37,500 citations for 20 days, about half of AI Overview citations and about 60% of AI Mode citations were gone by day two, though most returned at some point.

Do AI Mode and AI Overviews cite the same sources?

Rarely. An Ahrefs study of September 2025 US data found the cited URLs overlapped only 13.7% of the time, even though the answers were 86% similar in meaning. Track the two surfaces separately.

Does ranking in the top 10 get you cited in AI Overviews?

It helps. In July 2025, Ahrefs found that 76.1% of pages cited in AI Overviews ranked in the top 10 for the same query. A separate October 2026 Ahrefs study, using a different measure, found that nearly half of AI Overview citations and around 75% of AI Mode citations didn't rank in Google for the same question at all.

How should I measure AI search visibility?

Check each question several times over days, record every cited domain, and report citation share: the percentage of checks in which your domain was cited. Keep AI Overviews and AI Mode separate and compare checks from the same location.

How do I get AI Mode results from an API?

With PrismCrawl, send a Google Search request with udm set to 50. The generated answer and its citations are in the ai_summary item of serp_features, the same place AI Overview data appears in a normal search.

Related posts