Google adds temporary AI Mode threads for searches without past preferences

Temporary threads offer a documented session condition for AI visibility tests. Use a fixed-query comparison and inspect sources separately.

Sonar compares two search sessions with different account contexts.

Google introduced temporary threads in AI Mode on November 17, 2026. The feature lets searchers ask questions without carrying their account’s past preferences into the conversation. Google says temporary threads also stay out of search history and do not influence later personalization. The rollout covers desktop and mobile browsers, across languages supported by AI Mode. Google’s announcement

For an SEO team, the useful addition is a documented session condition to record in an AI visibility test. A recommendation observed in an ordinary conversation and one observed in a temporary thread can now be compared with a clearer account of their context. A temporary thread still has a location, language, prompt, and observation time, so the test record needs those details.

What changes when you open a temporary thread

Google’s instructions place the temporary-thread icon in the upper-right corner of AI Mode. Open it before entering the first question. The announcement says the conversation is detached from the account, excludes past preferences, and disappears when closed. Readers should capture the relevant evidence while the conversation is open.

Those statements describe how the product handles the thread. They don’t provide a fixed answer for everyone asking the same question. A business shortlist can vary by question, available sources, and run conditions. Record the feature as one part of the test environment and keep other known inputs steady.

A sensible first use is to inspect a question that may depend on prior interests. A returning user asking for a walking holiday could get a different response than someone starting a fresh conversation. The difference is worth describing, but its cause requires more than one answer.

Design a comparison around one reader decision

Pick a small decision your content helps a reader make. For example, a travel publisher could compare recommendations for a two-day itinerary in a named city. A software publisher could ask readers to choose a product for a clearly stated task and budget. Use the same complete wording in both conditions.

Prepare a fixed set of ten prompts before running the test. Include factual questions, recommendation questions, and questions requesting sources. Give each prompt an identifier so it can be repeated without silent edits. Keep a copy of the prompt file with the observations.

For the ordinary condition, begin a new conversation and record that it is an ordinary session. For the second condition, open a temporary thread and enter the same prompt. Use the same browser, device, language, and geographic setting. Record the model selection shown by the interface and the time of each run.

Alternate which condition you run first across the prompt set. This reduces the chance that one condition always happens later during a changing news cycle. It cannot eliminate every source of variation, but it makes the comparison easier to inspect.

Capture sources and recommendations separately

Save the answer, its visible source links, and any named organizations or products. A business mention can appear without a direct link, and a source can support an explanation without being recommended. Give those observations different columns.

A compact record needs the prompt identifier, session type, timestamp, country, language, device, visible model, named entities, source URLs, and a screenshot reference. Add a short note for incomplete or failed responses. Keep a missing answer in the log so the sample denominator remains visible.

For an initial check, count how many prompts named the target business in each condition. Then count how many contained a link to its site. Compare the destination pages as well as domains. A homepage link and a link to the relevant service page offer different next steps to the reader.

Interpret a difference without overstating it

Consider a hypothetical ten-prompt test. The ordinary condition names a business on six prompts, while temporary threads name it on four. The observed mention rates are 60% and 40%, a 20-percentage-point difference. That calculation describes those twenty responses.

It does not establish a general personalization effect. Repeat the same prompts on another date, check which questions changed, and report whether links supported the business names. If both conditions vary between repeats, include the variation in the results rather than discarding it.

Also compare recommendation quality. Does the answer honor the budget, geography, and task in the question? Does a cited page actually substantiate the sentence beside it? A stable mention with the wrong location can be less useful than a changing answer that follows the request accurately.

Use the result in editorial work

Repeated omissions are a reason to inspect the question and the sources AI Mode used. Review whether your page answers that task clearly, names its geographic scope, and supplies the evidence a reader needs. Keep any edits tied to the page’s real purpose.

The SERP testing guide provides a broader method for recording conditions. The AI visibility measurement crosswalk explains how appearances, citations, and referrals fit into different evidence layers.

The comparison above is a proposed procedure, and the figures are illustrative. No new account-level experiment is claimed here. Use the feature’s documented session behavior to make future observations more explicit, then preserve enough evidence for another person to review them.

Keep learning

Continue this topic

Community discussion

Discuss: Google adds temporary AI Mode threads for searches without past preferences

Have a question, a useful example, or a different perspective? Join the discussion, share evidence, and help other readers reach a better answer.

0 replies Moderated
No replies yet.

Be the first to ask a focused question, share a practical example, or add useful evidence.

Ask a question or join the discussion

Share evidence, a useful example, or a clear question. Be specific, stay on topic, and challenge ideas without attacking people. First-time replies may be held for moderation.