If you market an email tool, assume your buyers will see Google's AI Overview before your ranking. It appeared on all 40 US searches in this test, covering how-to questions, best-of lists and brand comparisons. The source it cited most often was YouTube, in 31 of the 40 Overviews. YouTube ranked among the regular results on only 4 of the 40 pages. Of the domains cited in an Overview, only 62 of 283 also appeared in the regular results on the same page.
What the Overviews cited
I ran 40 email-marketing searches through automatic/google-search-results-scraper on November 18, 2026 UTC, in English with the country set to us. Each search read one results page. The Actor reports the Overview's status and its cited links on that page. All 40 rows came back present. The 40 Overviews held 308 cited links from 139 distinct domains. One Overview, for "best email service provider for developers", had no cited links.
| Domain | Overviews citing it | Cited links | Pages where it ranked in regular results |
|---|---|---|---|
| youtube.com | 31 | 49 | 4 |
| reddit.com | 24 | 24 | 16 |
| zapier.com | 13 | 13 | 3 |
| mailchimp.com | 11 | 14 | 4 |
| campaignmonitor.com | 7 | 7 | 4 |
Domains are hostnames without www., so help.klaviyo.com counts separately from klaviyo.com. Regular results are the page's organic list, with roughly 7 to 10 links per page. Some Overviews cited several YouTube videos, giving 49 YouTube links across 31 Overviews. Reddit was closer to its regular ranking. It ranked among the regular results on 16 pages and was cited in 24 Overviews.
Does the type of search matter?
I wrote 14 how-to questions, 13 best-of or tool searches and 13 searches that name a brand. The full list is in the snapshot.
| Search type | Searches | With an Overview | Citing YouTube | Citing Reddit |
|---|---|---|---|---|
| Question or how-to | 14 | 14 | 12 | 8 |
| Best-of or tool search | 13 | 13 | 9 | 8 |
| Brand, pricing or comparison | 13 | 13 | 10 | 8 |
Questions leaned most on video. Brand searches still sent readers toward the brands. In all 12 brand searches that named a vendor, the Overview cited at least one named vendor's own site. "Mailchimp pricing", "sendgrid vs mailgun" and "brevo review" were among them. I counted twilio.com as SendGrid's site. I left out "substack alternatives" because it asks for rivals.
For an email product, I'd put a short video and an honest Reddit answer on my content list next to the blog post. On these pages, they were the two most-cited sources. Keep your own pricing and comparison pages current, too, because Overviews for brand searches cited them.
Limits
This is a single collection from one country, in English, without a signed-in Google account. AI Overviews change between runs and between users, so repeat the searches before relying on one count. I chose the 40 searches. They indicate patterns for this niche and do not measure all of Google. The regular results include a YouTube link only when Google ranks it among the organic links. A separate row of videos elsewhere on the page might not be counted. Nine organic links could not be resolved, and the script skips them. The first run reached its 300-second limit after 18 searches. I split the remaining 22 into two more runs with the same country and language. Every search has exactly one results page in the snapshot.
Reproduce the tables
Download the captured analysis fields as overviews.json. The snapshot preserves the returned order. It keeps the cited and organic URLs and omits answer text. Save the code as overviews.py and run python3 overviews.py overviews.json.
import json
import sys
from collections import Counter
from pathlib import Path
from urllib.parse import urlparse
runs = json.loads(Path(sys.argv[1]).read_text())["runs"]
pages = [page for run in runs for page in run["items"]]
order = runs[0]["input"]["queries"] # all 40; runs 2 and 3 finish the rest
groups = {"Question or how-to": order[:14], "Best-of or tool search": order[14:27],
"Brand, pricing or comparison": order[27:]}
def domain(url):
return (urlparse(url).hostname or "").removeprefix("www.") if url else None
by_term = {page["term"]: page for page in pages}
assert sorted(by_term) == sorted(order)
present = [p for p in pages if p["aiOverviewStatus"] == "present"]
print("Searches", len(pages), "with an AI Overview", len(present))
cited_in = Counter()
links = Counter()
for page in present:
domains = [domain(url) for url in page["aiOverviewSources"]]
links.update(domains)
cited_in.update(set(domains))
print("Cited links", sum(links.values()), "distinct domains", len(links))
print("Overviews with no cited links", sum(not p["aiOverviewSources"] for p in present))
def in_organic(page, site):
return site in {domain(url) for url in page["organicUrls"]}
for site, overviews in cited_in.most_common(5):
organic = sum(in_organic(page, site) for page in pages)
print(f"{site}: cited in {overviews} overviews ({links[site]} links), "
f"organic results on {organic} pages")
for name, terms in groups.items():
group = [by_term[t] for t in terms]
print(name, len(group),
"overviews", sum(p["aiOverviewStatus"] == "present" for p in group),
"youtube", sum("youtube.com" in map(domain, p["aiOverviewSources"]) for p in group),
"reddit", sum("reddit.com" in map(domain, p["aiOverviewSources"]) for p in group))
pairs = shared = 0
for page in present:
cited = {domain(url) for url in page["aiOverviewSources"]}
organic = {domain(url) for url in page["organicUrls"] if url}
pairs += len(cited)
shared += len(cited & organic)
print(f"Cited domains also in the same page's organic results: {shared} of {pairs}")
vendors = {"mailchimp": {"mailchimp.com"}, "klaviyo": {"klaviyo.com", "help.klaviyo.com"},
"beehiiv": {"beehiiv.com"}, "substack": {"substack.com"}, "kit": {"kit.com"},
"sendgrid": {"sendgrid.com", "twilio.com"}, "mailgun": {"mailgun.com"},
"brevo": {"brevo.com"}, "mailerlite": {"mailerlite.com"},
"activecampaign": {"activecampaign.com"}, "hubspot": {"hubspot.com"},
"postmark": {"postmarkapp.com"}, "constant contact": {"constantcontact.com"}}
named = own_site = 0
for term in groups["Brand, pricing or comparison"]:
if term.endswith("alternatives"):
continue # names one brand but asks for its rivals
sites = set().union(*(v for k, v in vendors.items() if f" {k} " in f" {term} "))
named += 1
own_site += bool(sites & {domain(u) for u in by_term[term]["aiOverviewSources"]})
print(f"Brand searches citing a named vendor's own site: {own_site} of {named}")
To run a fresh check in your own niche, put your searches in queries and use the saved first input object as the body for POST https://api.apify.com/v2/acts/automatic~google-search-results-scraper/runs?maxTotalChargeUsd=0.15&timeout=600 with an Apify bearer token. When the run finishes, read defaultDatasetId from the run record, then fetch GET https://api.apify.com/v2/datasets/{datasetId}/items?format=json. Each row's aiOverview has status, text and sources. Use the 600-second timeout above for 40 searches, or split them across runs as I did.
On November 18, 2026 UTC, the run records listed customer prices of $0.0015 per results page and $0.005 per GB of memory per run start. At the default 2 GB, that's $0.01 per start. The Actor's README says the page price becomes $0.002 on November 23, 2026. At the November 18 prices, these three runs would have cost a customer about $0.09 for 40 pages. On the publisher account, they used $0.067 of platform usage.
Apify run IDs
| Searches | Run ID | Dataset ID | Result |
|---|---|---|---|
| 1–18 | WxcfF9xAOIuMtUN8c |
uiugKAd9X4qNPoSy3 |
18 pages, stopped at the time limit |
| 19–29 | aTFyGo3gFj0xJvd5s |
W9QPJVBfhOlMcJcbz |
11 pages |
| 30–40 | 5wsKVXR4phhWe6THj |
SDdG5aZ6jvzBTgyJb |
11 pages |
Affiliate disclosure: The Apify Store link carries our referral code. If you sign up for a paid plan through it, ssh.autos may earn a commission at no cost to you.
Originally published at ssh.autos.
Top comments (0)