How Google Gemini Chooses What to Show in Its Answers

Grounding is the mechanism

Gemini does not always search. For general questions it answers from training. For questions that look time-sensitive, commercial, or specific to a named entity, it tends to run a search in the background, read a set of results, and write an answer with those results as support. Google calls this grounding, and in the consumer app the supporting links show up as chips or a "sources" expander under the response.

The practical consequence is simple. If you do not appear in the top results for the query Gemini runs, you are unlikely to be in the answer. Traditional organic ranking is not optional here. It is the entry ticket.

The Knowledge Graph does the entity work

For questions about a company, product, or person, Gemini tends to lean on Google's Knowledge Graph for the basic facts: what the entity is, where it is based, what it makes, who runs it. If your Knowledge Graph entry is thin, wrong, or missing, Gemini is working from a weak foundation and tends to hedge or confuse you with a similarly named company.

The inputs to the Knowledge Graph are things you can influence: Organization schema on your site, a consistent name and description across your profiles, a Wikidata entry where you qualify for one, and a Google Business Profile for local businesses. Every one of these should say the same thing. Entity consistency is boring work and it is the part most companies skip.

YouTube counts more here than elsewhere

Because Google owns YouTube, Gemini tends to surface video results more readily than the other engines do. In our testing, how-to and comparison questions often return a YouTube video as one of the grounding sources. A ten-minute walkthrough with a clear title and a transcript that states facts plainly can be a Gemini source even when your blog post on the same topic is not.

This does not mean every company needs a channel. It means that if your category has buyers who search "how do I" or "X vs Y," a small set of plain, accurate videos is a cheap way into one of Gemini's preferred source types.

Structure the page for extraction

Once Gemini has a page in front of it, the same rules apply as with any engine. It tends to use the first clear statement of the answer. It tends to prefer specific claims over superlatives. It tends to ignore content that only renders with JavaScript it cannot execute in the grounding step. Lists, tables, and short definitional paragraphs are easy for it to pull. Long narrative intros are not.

Schema helps at the margin. FAQ and Product markup give the grounding step a clean set of facts to work from, and they are the same facts that feed the Knowledge Graph.

What we see in audits

At AEO GEO Labs we measure share of answer across the major engines separately, because they behave differently. Gemini is usually the engine most correlated with classic Google rankings and the one where entity errors are most visible. A company with a strong blog but a messy set of profiles tends to do fine in Perplexity and poorly in Gemini. The fix is almost never more content. It is aligning the facts everywhere Google reads them and making sure the key pages rank for the query Gemini will run.

What to do this week

Search your company name and your core category in Gemini and read the sources it attaches. Then search the same things in Google and compare. If the Gemini sources are pages you do not control, check whether your own page ranks for that query at all. If it does not, that is the gap to close first. If it does rank but is not used, the page probably buries the answer. AEO GEO Labs has a free schema markup generator and an on-page checker that cover the two most common fixes.