How to Rank in Gemini: What You Can Control and Test

You get named or cited in Gemini's answers rather than ranked, and answers vary between runs. "Gemini" also covers three surfaces with different rules, so decide which one your buyers use before you change anything.

Because AI Overviews and AI Mode run on Google Search, our guide on how to rank in Google AI Overviews covers that surface in more depth. The rest of this page is about the Gemini app, where testing matters most.

Google says none of these are required for AI Overviews and AI Mode. Google does not document schema or AI text files for the Gemini app, and we have not tested them there. Google-Extended affects training and grounding, not Search ranking.

For AI Overviews and AI Mode, Google states that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," and that you do not need "new machine readable files, AI text files, or markup" (Google Search Central). The same page says existing SEO fundamentals remain worthwhile.

Google-Extended is the control people confuse with ranking. It is a token in robots.txt, the plain-text file at the root of a domain that tells crawlers which paths they may fetch. It is not a crawler. Google's documentation says it lets publishers manage whether crawled content may be used to train future Gemini models and for grounding in Gemini Apps and Vertex AI (Google's cloud AI platform). It has no user agent string of its own, and it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search" (Google crawler documentation).

So blocking it will not remove you from Search, and allowing it will not lift you in Search. Our recommendation is to leave it unrestricted unless you have a licensing or confidentiality reason to opt out, because a block opts your content out of grounding in Gemini Apps. Whether a block, or a page update, changes your mentions is something a test round can measure.

Reddit and YouTube hold the largest shares among Gemini's most-cited domains. Ahrefs Brand Radar is vendor data: over 3 million US queries on all topics, September 2026. "Mention share" is a domain's citations as a percentage of the combined citations of the top 50 domains, so the long tail of other domains is excluded.

Edmunds, Healthline, Kelley Blue Book (KBB), PCMag, Target, eBay, CNET and Trustpilot follow at roughly 1% to 2% each. Because the data covers all topics, your category's mix will differ. Our recommendation is to build your own version: run your prompts, tally the domains cited, and see which of them you appear on.

Count a mention in the answer text as well as a link. Gemini Apps Help says the app can miss a source it used, so a missing link does not prove your page played no part.

Run a robots.txt check: confirm that no rule blocks Googlebot from your whole site and that Google-Extended is not set to Disallow: /. Pass means neither block exists. For your own pages, passing is a necessary condition for Google to read them by crawling; it does not show that crawlable pages get cited.

We fetched robots.txt for the 20 domains in the Ahrefs list on October 9, 2026. Eighteen returned a readable file; Best Buy did not respond and Edmunds returned an access-denied page. Running the script's logic on the 18 files, 17 pass and one fails. Only Trustpilot names Google-Extended, with narrow path rules such as /error and /evaluate/, so it is not blocked sitewide.

The one failure is Reddit, the most-cited domain: its file closes the site to every crawler it does not name, and Googlebot is not named. The reason is a licensing route. Google announced on February 22, 2024 that it has access to Reddit's Data API, so Google's use of Reddit content does not depend on crawling. You do not have that route, which is why your own file matters. Some sites, Reddit included, return an access-denied page to some automated requests, so the file you retrieve can differ from ours.

The script needs bash and curl (macOS, Linux, or WSL, the Linux layer on Windows) and any version of awk. We tested it on sample files with GNU awk, mawk and BusyBox awk.

#!/usr/bin/env bash
# usage: ./gemini-check.sh example.com
f=$(mktemp); trap 'rm -f "$f"' EXIT
code=$(curl -sL -m 15 -A "Mozilla/5.0" -o "$f" -w "%{http_code}" "https://$1/robots.txt")
[ "$code" = 200 ] || { echo "REVIEW: robots.txt returned HTTP $code"; exit 1; }
tr -d '\r' < "$f" | tr 'A-Z' 'a-z' | awk '
function val(s) { sub(/^[^:]*:[ \t]*/, "", s); sub(/[ \t]*(#.*)?$/, "", s); return s }
/^[ \t]*(#|$)/ { next }
/^[ \t]*user-agent:/ {
  if (!inua) n = 0
  inua = 1; a[++n] = val($0)
  if (a[n] == "googlebot") hasgb = 1
  if (a[n] == "google-extended") hasge = 1
  next
}
{ inua = 0 }
/^[ \t]*disallow:/ && val($0) == "/" {
  for (i = 1; i <= n; i++) {
    if (a[i] == "googlebot") gb = 1
    if (a[i] == "*") st = 1
    if (a[i] == "google-extended") ge = 1
  }
}
END {
  if (hasgb) { if (gb) print "FAIL: the Googlebot group contains Disallow: /"; else print "PASS: the Googlebot group does not block the whole site" }
  else { if (st) print "FAIL: the * group contains Disallow: / and no Googlebot group overrides it"; else print "PASS: no sitewide block applies to Googlebot" }
  if (ge) print "REVIEW: Google-Extended is disallowed sitewide (opts out of Gemini training and grounding)"
  else print "PASS: Google-Extended is not blocked sitewide"
}'

• PASS on Googlebot: if your file has a Googlebot group, it does not contain Disallow: /; if it has none, the catch-all * group does not contain Disallow: /. A group named only Googlebot-Image or Googlebot-News does not count as a Googlebot group.
• FAIL: the group that applies to Googlebot contains Disallow: /. Fix this first, because Google cannot crawl those pages.
• REVIEW on Google-Extended: it is disallowed sitewide, which opts your content out of Gemini training and grounding. That is a choice, not an error; confirm it is the one you meant.
• REVIEW on the HTTP code: the file did not return a normal response. Open it in a browser and read it.

This is a first-pass check. Rules that combine Allow and Disallow for specific paths need a manual read.

Run 20 buyer prompts five times each, log every answer, and compare weekly rounds. One answer proves little because a brand can appear in one run and vanish in the next. The steps:

Here is the noise calculation, our model rather than an official figure. You compare two rounds, so the margin that matters is the margin of the difference: 1.96 × √(p₁(1−p₁)/100 + p₂(1−p₂)/100). With p₁ = p₂ = 0.23, that is 1.96 × √(0.001771 + 0.001771), about 11.7 points. We treat week-to-week swings under about 12 points as noise. The model assumes independent answers, which understates the real spread because the same 20 prompts repeat, so confirm a larger swing in a second round before you act.

For a longer method, see our guide to tracking AI citations over time.

Fix any crawl failure first, then work on the sources, then the pages. This is our recommended order, not a Google requirement.

We run your prompts against Gemini every day and record whether your brand is named, which sources are cited and which competitors appear, automating the repeated runs and the logging described above.

As of October 2026, Gemini tracking is on our Growth plan at $699 per month and our Enterprise plan at $1,499 per month. Our Launch plan at $299 per month does not include Gemini. Growth tracks 100 prompts daily across five engines, which is 500 answers a day (100 × 5), so five days give you five answers per prompt per engine, matching the five runs in the manual test. Enterprise tracks 200 prompts daily across seven engines, which is 1,400 answers a day (200 × 7). Current plans are on our pricing page.

From those answers we report citation sources, share of voice and sentiment. We also build content briefs from the citation gaps we find. If you are comparing tools first, use our checklist for evaluating any AI visibility tool.

If you want a baseline first, our free AI visibility audit shows how AI engines describe your brand in about 30 seconds. To see your own Gemini results over time, Book a demo.