Google AI Overviews, AI Mode and Gemini: plan separate measurements

Google AI Overviews, Google AI Mode, Gemini (app) and Gemini (API) are four separate things to measure, even though all four come from Google. Each reaches you through a different collection route, reports different evidence, and fails in a different way: an AI Overview may simply not appear, AI Mode reports no answer features we can count, and the two Gemini engines can give different answers to the same prompt. Plan one measurement per surface, keep their numbers apart, and read each against its own denominator.

In short

• Treat four Google-owned surfaces as four engines: Google AI Overviews, Google AI Mode, Gemini (app) and Gemini (API). Do not average them.
• Google AI Overviews and Google AI Mode can return no AI answer at all. That outcome is a fact about Google, not a failure to appear, and it needs its own rate.
• Gemini (app) and Gemini (API) are different collection routes and can answer the same prompt differently.
• Each surface exposes different evidence: organic results, searches, answer features, citations. A blank is not a zero.
• Everything here describes what a surface returned to a collection request. It does not show what a particular person saw, or why Google produced the answer.

What is the difference between AI Overviews, AI Mode and Gemini?

They are different surfaces with different jobs. An AI Overview is the AI-written answer block on Google's results page for a search. AI Mode is a separate mode of Google Search where the whole result is an AI-written answer that does not sit above a list of organic results. Gemini is Google's assistant, and it is measured two ways: Gemini (app) is the answer gemini.google.com shows a person who is not signed in, and Gemini (API) is Google's Gemini model answering through its API.

The naming is easy to blur, so the rest of this post uses the exact engine names. DiscoveredBy's engines reference treats all four as separate engines, and the glossary defines AI Overviews, AI Mode and Gemini (app) if you need a shared vocabulary for a team.

Enlarge image

How is each one collected, and why does it matter?

Collection route decides what the data can and cannot tell you. Gemini (API) is queried through Google's own API. The other three are collected by DataForSEO, a third-party data provider, which fetches the answer a person would see and returns it. In the product these are labelled "Licensed data".

That difference has practical consequences:

• Google AI Overviews: DataForSEO fetches Google's results page for the prompt, with the AI Overview and the page's organic results.
• Google AI Mode: DataForSEO fetches Google AI Mode's answer, which has no organic results.
• Gemini (app): DataForSEO puts the prompt to gemini.google.com in a session that is not signed in.
• Gemini (API): a call to Google's Gemini API with Google Search grounding.

Because Google does not say which model wrote an Overview or an AI Mode answer, the model recorded for those runs names the surface instead. For Gemini (app), the docs record that in every one of the checks made on 2026-09-28 the data provider reported a model that is not the one Gemini (API) calls. Do not assume the two Gemini engines are the same system with different wrappers.

What can a Google engine return instead of an answer?

Google AI Overviews and Google AI Mode can return no AI answer at all. When the request succeeds but Google shows no Overview, no AI Mode answer, or an Overview made only of product cards, ads or videos, DiscoveredBy stores the run as no AI answer shown, with no text and no citations.

This matters because of what happens next. A run with no AI answer is left out of every answer-based number: visibility, mentions, share of voice, citation rate and answer counts. It is reported as the shown rate instead: the share of runs on which Google showed an AI answer. In the docs' own wording, an example reads "AI Overview shown on 12 of 20 runs (60%)". See the glossary entry for shown rate.

Two things follow for your plan:

• A low citation rate on Google AI Overviews can mean you are rarely cited, or that Google rarely shows an Overview for your prompts. Read the shown rate first.
• If every Google run in a window showed no AI answer, the engine has no value for visibility metrics in that window. It is not 0%.

Only the two Google engines have this outcome. On Gemini (app), when the answer cannot be read the run is a failed run, retried automatically that day, and a run still failed after the last retry stays failed for that day. A failure is a collection problem; "no AI answer shown" is a Google outcome. Keep them in separate columns of your log.

Enlarge image

What evidence does each surface give you?

The evidence differs, so the questions you can ask differ. Google AI Overviews is the richest: besides the Overview and its cited sources, DiscoveredBy keeps the first page of organic results for the same search, even on runs where no Overview appeared. That lets you put your organic position next to whether the Overview cited you. The Explorer measures this as first-page organic rate and organic position.

Google AI Mode has no organic results, so neither of those metrics has a value for it. Its answer features are also blank: every AI Mode answer checked arrived as plain answer parts, with images and product cards drawn inside the answer text, which is not read.

For citations, Google engines report only the sources their answer cites, never a list of pages searched. So on the Retrieved vs cited page their channel reads "Cited sources only" and is not compared, and their fan-out (the related searches an engine runs before answering) reads "not applicable". Gemini (app) is the same: cited sources only, no fan-out. Gemini (API) reports the pages it grounded its answer on, and it records the searches its response reports, when it reports any.

One more difference for Gemini (app): Google product links and Google Maps place links appear among its sources but are not counted as citations. They are counted as shopping and local businesses among the answer features instead.

What prompt, location and language reach each engine?

The four surfaces do not receive your prompt in the same way, and that limits which variants you can compare across them.

Sources for the table are the engines reference and the prompts guide. The consequence is concrete: if you track a persona-based prompt ("as a procurement lead, which tools..."), it will run on the chat engines but is skipped, not altered, on the three DataForSEO Google engines, because Google receives only the search text. Compare like with like by choosing prompts and variants that all four can run. For city targets, see local AI visibility for how to compare responsibly.

A location is information passed on, not a person's actual location. The engines decide how much weight to give it, and a signed-in person on their own device can see something different.

The measurement matrix

Use this as your plan. Copy it, and fill one row per engine you track. The middle columns come from the product's documented behaviour; the last three are yours to decide.

MEASUREMENT MATRIX: Google surfaces
Project: ______   Prompt set: ______   Window: ______

Engine                | Route            | Can return no answer? | Organic results kept? | Fan-out       | Citations compare to retrieved?
Google AI Overviews   | Licensed data    | Yes (shown rate)      | Yes, first page       | Not applicable| No (cited sources only)
Google AI Mode        | Licensed data    | Yes (shown rate)      | No                    | Not applicable| No (cited sources only)
Gemini (app)          | Licensed data    | No (failed run only)  | No                    | Not applicable| No (cited sources only)
Gemini (API)          | API              | No                    | No                    | Recorded when reported | Not compared (grounding only)

For each engine, record:
1. Question this engine answers for us: ______
2. Primary metric and its denominator: ______
3. Metric we will NOT read on this engine (and why): ______
4. Variants that will not run here (persona, language, length): ______
5. Comparison we will make only within this engine: ______

Some cells need a second reading. "Not applicable" for fan-out means the surface does not report the searches, not that it did not search. "No" in the last column means there is no list of searched pages to compare citations against. For Gemini (API) it reads "Not compared": the product does not turn its grounding links into an attribution rate. And a blank answer feature is never a 0%: AI Mode's image or shopping rate is simply not measured.

Which metric belongs to which engine?

Read each engine with the metric that fits how it works, and never combine the denominators.

These metrics and their denominators are defined in Metrics defined, and you can chart any of them with the Explorer. If you are unsure how a mention differs from a citation, AI visibility, mentions and citations explained sets the definitions out.

Worked example: Quillstone

Quillstone sells document-review software to legal teams. It tracks one prompt, "best document review software for legal teams", in one country, and looks at 20 daily runs per engine (one per day, so 20 days). Its competitors are Brieflane and Clausewise. The made-up results:

The arithmetic shows why the surfaces stay apart. On Google AI Overviews, 5 citations across 12 shown Overviews is 42% (5 of 12); dividing by all 20 runs would give 25% and would blame Quillstone for Google's decision not to show an Overview on 8 runs. Naming is 6 of 12 (50%). On AI Mode, the shown rate is 80%, naming is 8 of 16 (50%) and citing is 6 of 16 (38%, rounded). On Gemini (app), naming is 9 of 20 (45%) and citation is 4 of 20 (20%). Gemini (API) names Quillstone in 11 of 20 answers (55%) and cites its domain in 7 of 20 (35%). Treat that with care: the Retrieved vs cited screen does not compare Gemini (API), because its grounding links credit nearly all the pages it grounded on whenever it sends any, so a rate there would measure whether the links were sent, not which pages it chose.

Enlarge image

Add the organic side. Quillstone's domain sits on the first page of organic results on 9 of the 20 Google AI Overviews runs (45%), including runs where no Overview appeared. The team's readings:

• The shown rate is the first finding. Google shows an Overview on 60% of these runs, so 40% of the time there is nothing to be cited in.
• Quillstone is cited in 5 of 12 shown Overviews while ranking on the first page in 9 of 20 runs. That is a hypothesis to test, for example by opening the runs where it ranked but was not cited, not a conclusion about why.
• The two Gemini engines give different naming rates (45% and 55%). The team logs that as a difference between collection routes, not as an improvement or a decline.

What can you not conclude?

These measurements are observations of collected answers. They do not explain Google's choices, and several common conclusions overreach.

• Not a real user's view. Collection is a signed-out request with a country or city passed in. A signed-in person on their own device can see something else.
• Not a cause. A citation or a mention shows what an answer contained. It does not prove why Google wrote it.
• Not one "Google AI" score. Blending the four engines hides the shown rate, the different denominators and the different failure modes.
• Not a Gemini forecast. Gemini (API) is not a proxy for Gemini (app) or for AI Overviews. In the docs' checks the app's answer to a prompt can differ from the API's.
• Not a missing engine when it is absent. An engine only runs if your plan includes it and it is currently available. If one of these is missing from a project, check that before reading it as a result. See plans and limits.
• Comparisons across periods. The product's own comparisons leave out an engine that ran on fewer than 80% of a period's answered days. An Explorer comparison with every engine selected does not: an engine that answered in only one period counts in that period. Choose engines deliberately when you compare.

Frequently asked questions

Are AI Overviews and AI Mode the same thing?

No. An AI Overview is an AI answer block on a results page, with organic results beside it. AI Mode is a separate mode where the whole result is an AI answer, with no organic results. DiscoveredBy measures them as two engines, because their evidence differs.

Is Gemini the same as Google AI Overviews?

No. Gemini is Google's assistant; an AI Overview is an answer inside Google Search. Gemini (app) and Gemini (API) are also separate from each other. A result on one tells you nothing certain about the others.

What does a low shown rate mean?

It means Google showed no AI answer on many of your runs for that engine, for those prompts and that location. It says something about the prompt set, not about your brand. Those runs are excluded from answer-based numbers so they do not lower your visibility unfairly.

Why does my brand appear in Gemini (API) but not in the app?

The two are different engines. Gemini (API) is Google's model called with Google Search grounding; Gemini (app) is what gemini.google.com shows a signed-out visitor, collected by a data provider. Their answers to the same prompt can differ, so report them separately.

Can I track persona-specific prompts on Google engines?

Not directly. Persona variants do not run on Google AI Overviews, Google AI Mode or Gemini (app), because Google receives only the search text and there is nowhere to put an audience. Those variants are left out rather than sent in a changed form. See tracking buyer personas for how to keep persona results comparable.

Do I need to check retrieved pages for these engines?

Not for the three DataForSEO engines. They report only the sources their answer cites, so there is nothing to compare citations against. The retrieved-versus-cited question applies to engines that report searched pages; see pages retrieved but not cited.

Next step

Set up one prompt on all four surfaces, then fill in the matrix above from what you see for that prompt before you add more. DiscoveredBy collects the four Google-owned engines as separate engines, keeps the shown rate and the organic results beside the Overview, and lets you chart each metric per engine in the Explorer. Learn more on the Google AI Overviews, Google AI Mode and Gemini engine pages, then sign in to start measuring.