How to Appear in Gemini and Get Your Brand Cited

The word «Gemini» also covers two different things, and separating them early saves a lot of wasted effort. Gemini is the Google app where someone writes to a model directly, and Gemini is also the family of models that writes the AI answers inside Google Search. Both surfaces cite pages, but they do not decide who gets cited through the same controls.

For a company the practical consequence is that no switch turns presence in this engine on. What exists is a set of Google Search requirements that are either met or not met, and pages that either answer a specific question or do not.

The most expensive confusion in the industry is treating the three surfaces as one and blocking a control while believing something else is being protected. Google spells it out in its AI features documentation: the access control for the AI answers inside Search is the robots.txt rules for Googlebot, and to limit what is shown from a page there are nosnippet, data-nosnippet, max-snippet and noindex.

Google crawler documentation, last updated 14 July 2026, settles it in one line: «Google-Extended does not impact a site’s inclusion in Google Search». The same sentence adds that it is not used as a ranking signal in Google Search either.

Blocking it is a legitimate decision and it has a concrete cost: the company stops being able to ground an answer inside the Gemini app. What blocking it does not do is protect rankings, because rankings do not depend on that control.

The controls that do affect what Google shows from a page inside its AI answers, according to its AI features documentation:

• The robots.txt rules for Googlebot, which decide whether Google may download the page at all. If Googlebot cannot get in, there is no indexing and no citation.
• nosnippet, which stops Google showing a text snippet from that page.
• data-nosnippet, which does the same for one specific part of the HTML and leaves the rest available.
• max-snippet, which limits how many characters Google may show from the page.
• noindex, which takes the whole page out of the index and, with it, out of any answer.

What is not needed is written down just as plainly. That documentation says there is no need to create new machine readable files, AI text files or markup to appear in these features, and that there is no special schema.org structured data to add either.

The consequence for whoever writes the page is direct: a page no longer competes against one query, it competes against the set of sub-questions Google breaks that query into. A page that answers only one of them enters a small part of the answer; a page that answers several separately can enter several.

How a page is written for an engine that splits the question into several:

• One heading per sub-question, phrased the way someone searching would phrase it rather than as a creative title.
• The answer inside the first lines of each section, ahead of any context.
• Every section understandable lifted out of its place: subject named, and no «this process» or «the tool».

The useful reading of that split is not «go and post on Reddit», but that a brand shows up through two routes: its own site and sites it does not control.

The figure carries two limits worth writing down. It measures US queries across all topics, so the split inside one sector and one language can look very different. And it measures AI Overviews inside Google Search, not the Gemini app, for which no equivalent public measurement exists.

Kulturo went from 517.0 daily impressions in its first 28 days with data to 773.1 in the last 28, 49.6% more, according to the generative AI features report in Kulturo’s Google Search Console. That rise includes two record days, 31 August and 1 September 2026; without them the last-28-day average drops to 555.6 and the rise is 7.5%.

HeyTalent grew more steadily: 27,674 impressions in the last 28 days against 9,597 in the first 28, which is 2.9 times as many, according to the generative AI features report in HeyTalent’s Google Search Console. Its daily average climbs month on month, from 293.2 in June to 1,205.5 over the first fourteen days of September 2026.

The third case looks more like a company starting out. A site specialising in rigid inflatable boats, implementing DoGeo’s plan on its own, gathered 895 impressions between 27 July and 14 September 2026, with 839 in the last 28 days against 116 in the first 28, which is 7.2 times as many.

All three figures carry the same limit and it belongs in writing: these are impressions, not clicks and not citations, the report is in beta, and in the third case the volumes are small. They say Google is showing those pages inside its AI answers, not how many people reached those sites that way.

What is not measured well today, worth knowing before building a monthly report:

• How many people reach a site from the Gemini app. That needs web analytics configured to separate that origin, and on DoGeo projects it is not configured yet.
• Whether the brand gets named in answers nobody at the company prompted. That only shows up through a fixed set of questions, repeated at the same interval.
• How often each assistant’s crawler passes through the site. That is read in server or firewall logs, and they have to have been kept from before the question was asked.