Gemini SEO: what the app, Google Search and the API each tell you

Three things called Gemini

Google uses the Gemini name for its family of AI models and for the products built on them. For a local business, three of those products matter. Each draws on its own mix of sources, answers to its own controls and comes with its own reporting, or none at all.

Google Search belongs on this list. Google's Search blog says Gemini 3 became the default model for AI Overviews in January 2026, and Gemini 3.5 Flash the default for AI Mode in May 2026. But those features live inside Search, with Search's rules and Search's reporting.

One more split before you start. People who search for Gemini SEO mean one of two things: getting a business named and linked in Gemini's answers, or using Gemini as a helper for SEO tasks such as drafting page outlines or grouping keywords. This guide covers the first. If you use Gemini as a writing or research helper, check its output against your own records and Google's documentation before anything goes live.

Reading a Gemini app answer

Treat every app answer as one observation, never as a ranking. The app is the surface people screenshot, and it's the hardest one to measure. Google's Gemini Apps Privacy Hub says some Gemini responses are grounded on Search results. Some, not all. Gemini Apps Help adds that not every response includes sources, and warns that Gemini can hallucinate and present inaccurate information as factual.

The app also knows who's asking. The Privacy Hub says Gemini collects location from the general area of the device, the IP address or the Home and Work addresses in the Google Account, and Gemini Apps Help says it can use precise device location with permission. Gemini can also personalise answers by referencing past chats, which needs Keep Activity on and the separate Memory setting switched on. For place questions, Gemini Apps Help says it uses public information from Google Maps for addresses, descriptions, websites, ratings and opening hours. That places feature is still rolling out gradually, and the same page says it's only available to users who are 18 or over, signed in with a personal Google Account, asking in English and with Keep Activity on. Work and school accounts don't meet that condition.

Say a plumbing business in Caloundra asks the app for a blocked drain plumber from the shop counter. The app sees a Caloundra location, may draw on the owner's own chat history, and may or may not have run a search. A renter in Nambour asking the same question starts from different inputs. Neither answer tells you where the business stands for anyone else.

So log each check with enough detail that someone else could repeat it. Run it in a dedicated personal Google Account used only for these checks, with Keep Activity on, since place answers need it, and Memory off, so past chats don't colour the answer. Repeat each prompt in a fresh chat before you record it:

• The date, time, device and which account you were signed in to
• The suburb the device was in, and whether precise location was on
• Whether Keep Activity and Memory were on
• The exact prompt, typed the same way each time
• Whether a Sources button appeared, and which sites it listed
• Whether your business was named, and every fact stated about it
• What the same prompt returned in a fresh chat

Reading Google Search AI features

AI Overviews and AI Mode are the Gemini-powered surface you can measure with Google's own data, because they're part of Google Search. Google Search Central's page on AI features says a page needs to be indexed and eligible to show in Search with a snippet to appear as a supporting link, and that there are no additional technical requirements. The same page says the existing SEO fundamentals still apply, and lists checking that your Business Profile information is up to date among them.

Measurement runs through Search Console. Google counts traffic from these features in the Performance report under the Web search type, so clicks from AI features sit inside the overall Web totals and aren't shown on their own. The generative AI performance report breaks out impressions from AI Overviews and AI Mode, and Search Console Help says it was rolled out to all websites as of 31 August 2026, though a site needs enough impressions in those features before data appears. It can be broken down by page, country, date and device. The guide to interpreting Google AI impressions covers how to read those numbers.

A Gemini app screenshot says nothing reliable about your AI Overview visibility, and the reverse holds too. The work that gets pages indexed and shown with a snippet is already the Search half of Gemini SEO. For how these features behave on local searches, see AI Overviews and AI Mode for local businesses.

If you suspect AI features are costing you clicks, test it on your own pages rather than trusting an industry average. Work through these checks:

• Pick the five to ten pages that bring most of your enquiries, such as your main service pages
• In the Performance report (Web search type), compare the same months a year apart for those pages, so seasonal swings don't read as an AI effect
• Note whether impressions held while clicks fell, or whether both fell, which points to lost visibility rather than lost clicks
• Open the generative AI performance report for the same pages and dates to see whether AI feature impressions appeared or grew
• Line the results up against your call and form enquiry counts for the same periods, because enquiries are what pay the bills
• Write down anything else that changed: a redesign, new pages, a Business Profile edit, a paid campaign starting or stopping

Reading a Gemini API or tool result

Ask how a figure was produced before you act on it. The Gemini API is how developers build Gemini into their own apps, chatbots and tools. If a report says it checked Gemini, it may have queried the API rather than the app, and API answers depend on settings the developer picked.

Google's API documentation describes grounding with Google Search as a tool the developer switches on. It connects the model to real-time web content so it can cite sources beyond its knowledge cutoff. Even with that tool on, the docs say the model decides whether a search would improve the answer. Grounding with Google Maps is off by default, and the docs say near me questions use map coordinates the developer sends, so check whether a local test sent any. The URL context tool reads pages named in the prompt, so pasting your own URL only shows the page is readable. It says nothing about whether Gemini would find it unprompted.

A tool that also claims to show Gemini crawling your site deserves a hard question. Google's crawler documentation says Google-Extended has no separate user agent: crawling is done by Google's existing crawlers, and the token only works as a robots.txt control. Server logs can't separate visits made for Gemini from visits made for Search.

Before you trust an API-based figure, ask for these details. The guide on how to compare AI visibility reports goes further on judging a tracker's method.

• The model name and the date of each run
• Which tools were on: Google Search, Google Maps, URL context or none
• The location sent with local prompts, if any
• How many times each prompt ran, since one run is one sample
• The raw responses, including every cited source

What Gemini SEO adds to normal SEO

Most of Gemini SEO is ordinary SEO done well, plus a few checks that are easy to skip. Google Search Central says you don't need new machine-readable files, AI text files or special markup to appear in AI Overviews or AI Mode. The table shows what carries over and what each surface adds.

Write pages that answer the question a customer asked. Put the direct answer in the first sentence under each heading, use headings that match how people phrase the question, and keep each passage readable on its own, with the suburb, the job and what happens next named in it. Use a table where a reader is comparing options. That helps a reader skimming on a phone and gives Search a clean passage to show as a snippet. Google doesn't promise any of it will be quoted.

Make trust visible on the page. Google's guidance on helpful, people-first content describes E-E-A-T (experience, expertise, authoritativeness and trust), says trust matters most, and asks whether it's self-evident who wrote a page and whether it shows first-hand expertise. For a local service business that means a named author or tradesperson, the licence number where your trade is licensed, real photos of real jobs, specifics only someone who did the work would know, and the date of the last real update.

Look after the details that live off your website. Business Profile Help says businesses with complete and accurate information are more likely to show in local results, that more reviews and positive ratings can help local ranking, and that there's no way to request or pay for a better local ranking. The Gemini app's place answers draw on that same public Google Maps information, ratings included. Google doesn't publish how reviews or directory listings feed AI features, so treat them as groundwork for local search and accurate answers. The guide to getting more Google reviews covers the asking process.

Controls and common mistakes

A setting for one surface can leave the others untouched. This table matches common goals to what Google's own documentation points to.

Pick a snippet control when a page should keep showing in Search but a passage shouldn't be quoted, and pick noindex only when the whole page should leave Search.

Check the Google-Extended row first. Google's crawler documentation says the token controls whether crawled content can train future Gemini models and ground answers in the Gemini app and Google's Vertex AI services, and that it doesn't affect inclusion in Google Search or act as a ranking signal. Blocking it won't remove you from AI Overviews. It does ask Google to keep your pages out of grounded Gemini app answers. If that rule sits in your robots.txt, make sure someone chose it on purpose.

These are the mistakes that most often waste a business owner's time and money:

• Blocking Google-Extended without a deliberate decision, or expecting the block to remove you from AI Overviews
• Rewriting pages because of one screenshot from one phone in one suburb
• Trusting an API-based visibility score without its model, tools, location and raw responses
• Adding an llms.txt file or so-called AI schema and expecting a change in Google Search
• Changing the date on a page without changing the content, which Google's people-first content guidance lists as a warning sign
• Paying anyone who promises your business a place in Gemini answers
• Leaving old hours or a disconnected phone number on the Business Profile while polishing the website

Worked example: a monthly Gemini check for a Caloundra plumber

Here's the method taken end to end for a hypothetical plumbing business in Caloundra. The business, prompts and results are illustrative only.

Step one is choosing prompts from real demand, not guesses. The owner reads back through a month of calls and enquiry forms and the queries in Search Console, then writes eight prompts that cover the main services, the suburbs served and the situations customers mention. For example: "emergency plumber open now in Caloundra", "who can fix a blocked drain in a rental in Golden Beach", "hot water system replacement Pelican Waters" and "plumber for a bathroom renovation in Currimundi". The prompts stay fixed from month to month, so a change in the answer means something.

Step two is running and logging. The owner runs each prompt in the Gemini app from the same two places each month, the workshop and a phone at a regular job site, then fills in one row per prompt. The illustrative entry below shows the level of detail that makes a log useful.

Step three is tracing the wrong fact. The Saturday hours in that entry came from the Business Profile, which still showed the old roster. The owner corrects the profile, notes that Google reviews edits before they go live, and checks the website footer and two directory listings for the same detail. The guide to business listing accuracy covers the directory side.

Step four is fixing the page the answer leaned on. The blocked drains page opened with "We do all drains, big or small, call us today." The rewrite leads with the answer: "We clear blocked drains in Caloundra, Golden Beach and Pelican Waters, including tree root and grease blockages. A licensed plumber inspects the line with a camera, clears it, and explains the cause before you pay." The page also gains the plumber's name, the licence number, and photos from recent jobs.

Then the owner works through the same short routine every month:

1. Open robots.txt and confirm whether Google-Extended is disallowed, and whether that's intended
2. Spot check the main service pages for noindex or nosnippet tags added by mistake
3. Compare Business Profile hours, phone, address and services with the website, using a Business Profile audit checklist
4. Review Search Console's Web performance and generative AI reports for the pages that bring enquiries
5. Run the fixed Gemini app prompts from the same places and log what changed
6. File any API or tool results separately, with their settings attached
7. When a fact is wrong or out of date, fix the source with a real change and record the date; act on a pattern across several months, never on one odd answer