Why Does Gemini Show Redirect Links Instead of the Real Destination URL?

What does a grounded response contain?

A grounded response from the generateContent API includes a groundingMetadata object. Google's documentation describes these parts:

Firebase's example shows the pattern plainly: the uri begins with https://vertexaisearch.cloud.google.com and the title is aljazeera.com. Some chunks point to Google itself, such as a google.com/search link for weather or local information.

Why does Google use redirects?

Google's documentation does not say. Two things are clear from how the feature works:

• The redirect keeps Google in the path. Every click on a source passes through a Google URL before reaching the publisher.
• Google sets display rules for grounded results. Developers must show the Google Search Suggestions returned with a grounded response, exactly as provided, and for image results must give a direct, single-click path to the source page.

Developers raised the issue on Google's forum and the Python SDK's GitHub in October 2025, asking for a field with the original URL for logging, deduplication and display. Google's documentation for the newer Interactions API shows citations as url_citation annotations whose example URLs point straight to the publisher's page, so the behaviour may differ between APIs. Check what your API version returns.

How do you get the real URL?

Within the limits of Google's grounding terms:

1. Use title or domain to label each source for the user who asked the question.
2. Let the link go through the redirect when the user clicks. Do not resolve, cache or log the URLs for other purposes.
3. Use an API that returns publisher URLs if you need them in the response, such as the Interactions API's url_citation annotations, and check its terms.

Google's Grounding with Google Search terms say developers will not cache, analyse or learn from grounded results, and that collecting Links by automated means for another purpose breaks the terms. So grounded API responses cannot be used to build a citation tracker.

Do not show raw redirect strings to users as if they were addresses. Show the domain or page title as the link text and let the link go through the redirect.

Does the Gemini app show redirect links too?

The redirect behaviour described here applies to developers' API responses. In the consumer Gemini app, people see source cards and links, and clicks arrive at your site from Gemini. Analytics tools such as Matomo and Plausible now group visits from Gemini with other AI assistants based on the referrer. If you see traffic you cannot attribute, check both gemini.google.com and Google redirect hosts in your referrer reports.

What does this mean for publishers?

If you want to know whether Gemini cites you:

• Tracking tools should not be built on the grounding API. Google's terms forbid collecting or analysing Links from grounded responses, so ask a vendor how it gathers Gemini data. Checks in the consumer Gemini app, and Search Console's generative AI report for Google Search, are the routes to use.
• Page-level checks happen in the app. Whether Gemini cites your category page or a thin filtered URL is best seen by asking the questions in the Gemini app itself, as covered in checking whether Gemini cites thin faceted URLs.
• Grounding respects Google-Extended. Google says Grounding with Google Search on Vertex AI does not use pages from sites that disallow Google-Extended in robots.txt. Blocking it keeps your pages out of these grounded API answers, though not out of Google Search.
• Search Console will not show these. Google's Generative AI performance report covers AI Overviews and AI Mode in Search, not answers generated through the Gemini API. Other ways to see Gemini citations are covered in seeing which pages Gemini cites without Search Console.

For the basics of being retrievable, see how to make your site crawlable and indexable by Gemini.

Can developers control which sites are used?

Partly. On Vertex AI, the Google Search tool accepts an exclude_domains list, so a developer can stop named domains from being used for grounding, for example to keep a competitor's site or a low-quality aggregator out of an internal tool's answers. The documentation describes excluding domains, not requiring them. This is a control for developers' own applications. It does not change what ordinary Gemini users see.

How should developers display sources?

Google's grounding documentation sets the rules for how grounded answers are shown, so build the source list with them in mind:

• Show the Search Suggestions chip returned in searchEntryPoint wherever the grounded answer appears, without restyling it.
• Use groundingSupports to attach each source to the sentence it supports, so users can see which claim comes from where.
• Label links with the site or page name, taken from title or domain, rather than the redirect string.
• Keep one click to the source. Avoid intermediate screens between a user's click and the publisher's page.

We have not tested how long the redirect tokens stay valid. This page is based on Google's grounding documentation for the Gemini API, Vertex AI and Firebase, and on developer reports on Google's forum and GitHub.

What might this look like in practice?

Consider an illustrative agency, Skellnell, that built an internal tool asking Gemini with Search grounding 200 client questions each week to track citations. Its first reports listed vertexaisearch.cloud.google.com as the top cited site. When the developer looked into resolving the links, they found that Google's grounding terms forbid collecting or analysing grounded links, so the agency retired the tool. It now checks a fixed set of client questions in the Gemini app each month and uses Search Console's generative AI report for AI Overviews and AI Mode. Those checks showed that one client was cited mostly through an old press release rather than its product pages, which gave the content team a clear task.

What do people ask about Gemini redirect links?

Why is every Gemini source URL on vertexaisearch.cloud.google.com?

Google's grounding API returns redirect links. The real domain is in the title field, and in Vertex AI's examples, a domain field.

How do I get the original URL?

Let the user's click follow the redirect. Google's terms forbid resolving and caching grounded links for other purposes; if you need publisher URLs in the response, check whether the Interactions API's url_citation annotations and their terms fit your use.

Is the title always the domain?

Usually, but not always. Some chunks, such as Google weather or local results, have descriptive titles.

Does blocking Google-Extended remove my site from Gemini API grounding?

Google says Vertex AI grounding does not use pages from sites that disallow Google-Extended.