Gemini Spark sorted PDFs but returned no Amazon listings in one test

A hands-on evaluation published October 9, 2026, reported mixed results for Google’s Gemini Spark: it summarized online sentiment and organized a folder of PDFs, but returned no Amazon listings for an Amazon-only search and could not interact with an already-open Google Chrome window. These were outcomes from one reported session, not a standardized performance benchmark.

How Google describes Gemini Spark

Google presents Spark as an AI agent for multi-step work. Its product description separates work into Tasks (assigned jobs), reusable Skills (saved instructions) and Schedules (work triggered by a time or condition). Google says Spark can run in the background, including when a phone or laptop is off, while users direct and supervise its work. It is designed to ask before major actions.

Google lists Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube and Google Maps as connectable apps. Those connections are off by default, and users choose which to enable. Google says access is for Google AI Pro or Google AI Ultra subscribers over 18 in select countries, as well as select business users; availability and compatibility vary. The account used in the reported test was described as a Google AI Pro plan costing $20 per month.

The feature list describes what Google intends Spark to do. The reported tasks offer a narrower look at how it performed in one session.

Amazon search returned four results, none from Amazon

In a same-prompt product-search task, Spark returned four results and described them as Amazon listings with Prime delivery, but none came from Amazon. The comparison system, ChatGPT Work, returned 11 Amazon product candidates; the tester selected nine for review.

These timings and outputs describe the reported run only. The shorter elapsed time did not make Spark’s results useful for the specific request: the results missed the retailer the task required.

Sentiment analysis offered a useful summary

For another task, Spark summarized reactions to the Claude Opus 5-to-5.5 update using material from Reddit and Substack. The summary was useful, but it did not address the requested X and Facebook sources. That is the outcome of this particular assignment, not a general finding about Spark’s access to those platforms.

PDF organization worked, with a broad folder warning

Spark renamed and organized a folder containing 507 PDFs into 13 subfolders in the reported session. The folder-access dialog warned that allowing access could include sharing the folder’s contents with third parties.

The session’s interface also displayed a notice that backups of connected-folder files would be deleted after a task follow-up or after 24 hours. The warning described a possibility; it did not say that the folder’s contents had actually been shared.

Spark could not interact with the open Chrome window

Asked to inspect an already-open Google Chrome window and list its active extensions, Spark replied that it could not directly access or interact with that window. That response came from the same reported test and applies to that request and build.