Gemini visibility tracking
Gemini visibility is not one universal rank. Consumer Gemini Apps, a configured Gemini Developer API request, and Google Search AI features are separate observation surfaces. RankEcho samples the Gemini Developer API with its Google Search tool enabled and reports a representative prompt-engine cell; it does not reproduce a person's Gemini Apps session or an AI Overview. An absent mention or source in that bounded record is not proof of global absence.
Which Gemini surface are you measuring?
A defensible visibility record starts by naming the product and collection method. The same prompt can produce different evidence across Google's consumer assistant, a developer-configured API request, and generative features inside Google Search. Results from one surface cannot be relabeled as telemetry from another.
| Surface | Observable evidence | Measurement boundary |
|---|---|---|
| Gemini Apps | The answer shown to a user, plus any inline links or Sources panel available in that session | Account state, model, location, personalization, connected apps, instructions, and time can differ; RankEcho does not observe this interface |
| Gemini Developer API with Google Search | A response from a named model and configuration, plus grounding metadata and source URLs when returned | One API run does not reproduce Gemini Apps, and enabling the tool does not mean every successful response returns sources |
| Google AI Overviews and AI Mode | Search feature availability, visible supporting links, and Google Search Console's documented link-impression data | These are Google Search surfaces with separate controls and reporting; they are not Gemini Apps or the Gemini API |
What does RankEcho's Gemini cell observe?
Free audits check Perplexity and Gemini once per prompt. Paid and trialing accounts add ChatGPT, Claude, and Google AI Overviews when those adapters are configured, for up to 5 engines. RankEcho's Gemini cell calls the configured Gemini Developer API model with the Google Search tool. A non-2xx response is marked skipped; a 2xx response with no candidate text or grounding URLs is currently treated as an observed, uncited result. RankEcho does not observe the consumer Gemini Apps interface or Google AI Overviews, and a successful API answer can return without grounded source URLs.
Free audits run each prompt-engine cell once; Solo runs it twice; Growth and Agency run it 3 times. For repeat sampling, RankEcho currently collapses the attempts into one representative cell with a stability summary. That summary includes skipped attempts in its run count and citation probability, the selected representative can itself be skipped, and RankEcho does not persist every repeat's answer and source set as separate records.
For a research protocol that needs observed-only denominators, retain every planned attempt outside the current collapsed RankEcho record: prompt, model configuration, run time, HTTP and response state, answer text, and returned source set. Do not infer those per-attempt fields from the representative cell.
The adapter reads the first returned candidate, collects Google Search grounding chunks, and makes a bounded attempt to resolve Google redirect URLs to publisher destinations. If a redirect cannot be resolved, the recorded URL can remain a Google intermediary. That is a source-resolution limitation, not evidence that the publisher was absent.
Which result states should be preserved?
A research-grade per-attempt protocol should separate request availability from what appeared in a non-skipped response. Current RankEcho repeat sampling does not preserve all of those rows separately, so its representative cell and stability probability must not be relabeled as an observed-only denominator.
| State or dimension | Definition | How to interpret it |
|---|---|---|
| Skipped attempt | The API returned non-2xx or the request otherwise failed before a non-skipped result | In current repeat sampling it remains inside the stability run count; only a skipped representative is excluded from downstream scorecard scoring |
| Non-skipped API result | The adapter returned a result without setting skipped; a 2xx response can qualify even when candidate text and sources are empty | Inspect the actual text and sources; current RankEcho can treat an empty 2xx result as an uncited observation |
| Brand mention | The representative answer names the tracked brand or domain | A mention is not automatically a recommendation or a source, and the persisted cell is not a frequency across all raw repeats |
| Owned grounded source | The representative result includes an owned-domain URL after bounded normalization | The URL is evidence from the selected API run, not a permanent rank, an all-repeat frequency, or a universal Gemini Apps citation |
| Representative miss | The selected non-skipped result contains neither the tracked mention nor an owned grounded source | Count it only within the disclosed RankEcho cell; it does not prove absence across every repeat, model, account, location, or time |
What do Gemini source links establish?
In the Developer API, grounding metadata can identify search queries, source chunks, and support mappings returned for that request. RankEcho records the available URLs, but a source record does not by itself establish that the page caused the answer, supported every sentence, received a click, or will recur on the next run.
Consumer Gemini Apps use a different presentation. Google says some responses include inline links or a Sources button and not every response has sources. A listed item can be source or related content for only part of an answer. Preserve the placement you actually saw instead of collapsing every link into one generic citation count.
Why can the same prompt change?
A sampled answer can vary with the prompt wording, configured model, model update, time, and whether the API decides Search can improve the response. Consumer Gemini Apps add conditions RankEcho does not collect, including signed-in state, location, personalization, past chats, saved information, connected apps, and the selected model.
That variability makes a fixed prompt panel more useful than a single screenshot or a synthetic Gemini rank. For a study with explicit observed and unavailable denominators, preserve every raw attempt separately; current RankEcho's collapsed repeat record cannot supply those all-repeat counts.
Which publisher controls have which scope?
There is no single Google AI visibility switch. Review each control only for the product Google documents it to govern.
| Control | Documented scope | Do not infer |
|---|---|---|
| Google-Extended | A robots.txt control for some Gemini Apps, Vertex AI grounding, and future model-training uses; it has no separate HTTP user agent | Google Search inclusion, a Search ranking effect, or complete removal of prior model knowledge |
| Googlebot, indexing, and snippet eligibility | Google Search and its generative Search features | A complete publisher eligibility specification or guaranteed source link in Gemini Apps |
| Search generative AI Include/Exclude | Supported generative features inside Google Search and Discover | Gemini Apps behavior, Gemini API configuration, or model-training control |
| Gemini API Google Search tool | A developer request configuration that lets the model use Search when it decides that grounding can help | A publisher control, a guarantee of returned sources, or a consumer Gemini Apps result |
How should a change be tested?
Find: save the exact prompt panel, configured surface, model, date, availability, answer, mention state, and returned sources. If the analysis requires per-attempt denominators, collect those raw rows outside RankEcho's current representative-cell persistence. Fix: change the verified user or evidence problem on one named page, preserve the ship date, and avoid treating schema type, paragraph length, or a keyword template as a Gemini switch. Prove: re-run the same planned panel and compare like with like while disclosing skipped attempts and the collection method.
A later mention or grounded URL is movement observed after the shipment. It does not prove the shipment caused the result; retrieval choices, model behavior, competing sources, and time also changed between observations.
Are Gemini and AI Overviews one visibility surface?
No. Google documents different products, controls, and evidence. The Search Console Generative AI report covers AI Overviews and AI Mode, not Gemini Apps. A RankEcho Gemini API observation cannot stand in for that report, and movement on either surface should not be assumed to transfer to the other.
Sources reviewed
Provider eligibility and measurement claims below were checked against primary documentation. These records do not establish a universal selection formula, causation, or a guaranteed ranking, impression, recommendation, or citation.
10 claim-level source records
| Claim reviewed | Official source | Review record |
|---|---|---|
| Google says normal Search indexing and snippet controls govern eligibility for AI Overviews and AI Mode; there are no additional technical requirements or special AI schema files. | Google Search AI features documentation | Checked 2026-09-02 · AI features documentation updated 2025-12-10 · Primary-source documentation review; eligibility does not guarantee selection or presentation in an AI feature. Re-reviewed for the Gemini resource pair on September 2, 2026. · Confidence: High |
| Google documents Google-Extended as a robots.txt control for some Gemini Apps, Vertex AI grounding, and future model training uses; it has no separate HTTP user agent and does not affect Google Search inclusion or ranking. | Google common crawler documentation | Checked 2026-09-02 · Google-Extended documentation updated 2026-07-14 · Primary-source documentation review; Google-Extended is not represented as the control for Google Search AI features. Re-reviewed for the Gemini resource pair on September 2, 2026. · Confidence: High |
| Google says Gemini Apps can use context including location, account activity, past chats, connected apps, and saved information or instructions, depending on the user's settings and product state. | Gemini Apps Privacy Hub | Checked 2026-09-02 · Gemini Apps documentation reviewed September 2, 2026 · Primary-source review; account and session context is recorded as a variability boundary, not inferred from a Developer API response or from Google Search AI features. · Confidence: High |
| Google says Gemini Apps can use public information from Google Search. Search availability does not establish that every response performs a Search or returns source links. | Connect apps to Gemini Apps | Checked 2026-09-02 · Gemini Apps connected-app guidance reviewed September 2, 2026 · Primary-source review; Search is described as an available information source, not a guarantee of Search use, grounding metadata, or a visible source for every response. · Confidence: High |
| Google documents selectable Gemini Apps models and model availability that can depend on the account and plan. A sampled consumer session should therefore record the selected model rather than assume one universal Gemini output. | Gemini Apps model and plan guidance | Checked 2026-09-02 · Gemini Apps model guidance reviewed September 2, 2026 · Primary-source review; model choice is a sampling variable and does not create a publisher-facing rank or guarantee a source link. · Confidence: High |
| Google says Gemini Apps may show inline links and a Sources button with sources and related content, but not every response has source links and a listed page can be related to only part of the response. | View related sources from Gemini Apps | Checked 2026-09-02 · Gemini Apps source guidance reviewed September 2, 2026 · Primary-source review; a source-panel link is not represented as a durable rank, a claim-level citation in every case, or proof of universal visibility. · Confidence: High |
| Google warns that Gemini Apps responses can be inaccurate and can inaccurately explain how Gemini works, including how it cites sources; users should verify important claims. | Learn about responses from Gemini Apps | Checked 2026-09-02 · Gemini Apps response guidance reviewed September 2, 2026 · Primary-source review; grounding and visible links are not described as eliminating hallucinations or making a response authoritative. · Confidence: High |
| In the Gemini generateContent API, a developer can enable the Google Search tool; the model decides whether Search can improve the response and can return groundingMetadata such as search queries, source chunks, and support mappings. | Gemini generateContent grounding with Google Search | Checked 2026-09-02 · Gemini generateContent documentation reviewed September 2, 2026 · Primary-source review; one configured generateContent request is evidence for that request, model, prompt, configuration, and time—not a reproduction of consumer Gemini Apps or Google AI Overviews. · Confidence: High |
| Google Search Console provides an Include, Exclude, or inherited property setting for supported generative AI features in Google Search. Google documents this separately from Gemini Apps, Google-Extended, ordinary Search ranking, advertising, and model training. | Google Search generative AI control | Checked 2026-09-02 · Worldwide rollout stated as August 31, 2026 · Primary-source review; the property setting is not represented as a Gemini Apps or Gemini Developer API control. · Confidence: High |
| Google's dedicated Generative AI performance report records link impressions from AI Overviews and AI Mode and groups them by page, country, date, and device; its documentation does not describe Gemini Apps telemetry. | Google Search Console: Generative AI performance report | Checked 2026-09-02 · Worldwide rollout stated as August 31, 2026 · Primary-source review; a Search generative AI link impression is not relabeled as a Gemini Apps mention, a Gemini API source, or proof of causal selection. · Confidence: High |
Frequently asked questions
RankEcho's Gemini cell calls the configured Gemini Developer API model with the Google Search tool. A non-2xx response is marked skipped; a 2xx response with no candidate text or grounding URLs is currently treated as an observed, uncited result. RankEcho does not observe the consumer Gemini Apps interface or Google AI Overviews, and a successful API answer can return without grounded source URLs.
RankEcho enables the Google Search tool, but Google says the model decides whether Search can improve the response. A successful answer can contain no returned grounding source URLs.
No. RankEcho records answer-text mentions and returned owned source URLs as separate dimensions. Neither alone proves a recommendation, click, causal effect, or durable placement.
A non-2xx or failed API attempt is marked skipped. In current repeat sampling it still contributes to the stability run count and citation probability; if a skipped attempt becomes the representative, the downstream scorecard excludes that cell. This is not an observed-only denominator.
No. Google says Google-Extended does not affect Search inclusion or ranking. Googlebot, Search indexing and preview controls, and the Search generative AI property setting have separate Search roles.
No. Google's dedicated Generative AI report covers documented link impressions from AI Overviews and AI Mode. It is not Gemini Apps telemetry.
No. Preserve separate prompts, availability states, source units, controls, and reporting for the Gemini surface and Google Search AI surface you actually observed.
Free audits check Perplexity and Gemini once per prompt. Paid and trialing accounts add ChatGPT, Claude, and Google AI Overviews when those adapters are configured, for up to 5 engines. RankEcho reports bounded observations; it cannot guarantee a mention or source or prove why a later result changed.