Google made Gemini 3.7 Flash selectable inside the AI Mode model menu in Search on 14 August 2026, one day after the model itself was released on 13 August 2026. The option appears under a section labelled "Gemini 3 models" next to the existing "Auto" and "Pro" choices, it is available worldwide but in English only, and it is restricted to people on a Google AI Pro or Google AI Ultra subscription.
Robby Stein, VP of Product for Google Search, announced the rollout on X, describing the model as "better at following instructions + understanding your intent, so you get even more helpful responses." Matt G. Southern reported the rollout for Search Engine Journal the same day. Google has not said whether Gemini 3.7 Flash will become the default model in AI Mode, and it has not said whether the model feeds into the automatic "Auto" routing system that launched in November.
Where the new option appears inside AI Mode
Users reach the model menu by clicking the "+" icon in AI Mode's "Ask anything" bar. Google groups the available models under a heading that reads "Gemini 3 models", and as of 14 August 2026 Gemini 3.7 Flash sits in that group beside Auto and Pro. Nothing else about the entry point changed: same bar, same icon, one extra row in the menu.
That row is the entire change, and it is easy to underrate because of how small it looks. A person on the free tier who opens AI Mode on the same day, in the same country, on the same query does not see the row at all. Two people can run an identical search and receive answers produced by different models without either of them doing anything unusual. Model choice has become part of the state of a search, in the same way that country, device and login status already were.
The timeline: model released 13 August, picker updated 14 August
Gemini 3.7 Flash was released on 13 August 2026. It became selectable inside AI Mode on 14 August 2026. The gap between the two is one day, and Google offered no explanation for the speed of the second step. Search Engine Journal published its report on 14 August, the same day the option appeared.
It is worth being precise about what that timeline does and does not establish. It establishes that a newly released Gemini model reached a Search surface within a day. It does not establish a policy, a pattern or a commitment. The source does not say whether future Gemini releases will follow the same one day path into AI Mode, and nobody at Google is quoted saying so. Anyone building a monitoring routine around model releases should treat the one day interval as a single observation, not a schedule.
Who gets the option, and who does not
Access is limited in two ways at once. The first limit is commercial: only Google AI Pro and Google AI Ultra subscribers see the model picker entry for 3.7 Flash. Free tier users of AI Mode do not get the option, which means the majority of people who encounter AI Mode in ordinary search behaviour are unaffected by this specific change.
The second limit is linguistic. The rollout is global in geography but English only in language. An English-only rollout does not reach Thai-language queries at this stage. That is a plain reading of the availability that Google stated, and it should not be dressed up as a statement about Thailand. Google said nothing about Thailand, nothing about Thai-language AI Mode, and nothing about when or whether other languages will be added.
What Google confirmed and what it left open
The reporting on this rollout is short, and the confirmed facts fit into a small table. The right hand column below separates what has been stated from what has not, because the second group is larger than the first and is where most of the wrong assumptions will come from.
| Detail | Status as reported |
|---|---|
| Gemini 3.7 Flash release date | Confirmed: 13 August 2026 |
| Selectable in the AI Mode model menu | Confirmed: 14 August 2026, under the "Gemini 3 models" section next to Auto and Pro |
| Who can select it | Confirmed: Google AI Pro and Ultra subscribers only, global availability, English only |
| Whether 3.7 Flash becomes the default model | Not stated by Google |
| Whether 3.7 Flash feeds the automatic "Auto" routing that launched in November | Not stated by Google |
Why one extra row in a menu fragments AI visibility measurement
Brand citation tracking in AI Mode has largely been run as if AI Mode were a single surface: you ask a question, you look at the answer, you record whether the brand appears and which page was cited. A model picker breaks that assumption. The answer a user sees, and whether a brand is cited inside it, now depends partly on which model that user picked before asking.
Stein's description of 3.7 Flash points at exactly the behaviour that matters here. He called it better at following instructions and at understanding intent. Those are properties of how a prompt is read and how a response is assembled, which is the same machinery that decides which sources get pulled in and named. Google published no comparison of citation behaviour between Auto, Pro and 3.7 Flash, so there is no measured difference to quote. The point is structural rather than measured: where there was one answer surface to sample, there are now several, and they are selected by the user rather than by the observer.
For anyone tracking brand citations, the practical consequence is blunt. A single AI Mode check is no longer representative of AI Mode. A screenshot showing a brand cited, or absent, is now a statement about one model, one language and one account tier at one moment. Reported as "we appear in AI Mode" or "we do not appear in AI Mode", it overstates what was actually observed.
What to record when you check AI Mode from now on
The fix is boring and it is a records problem, not a tooling problem. Every AI Mode observation should carry enough context that a later change can be attributed to something. At minimum, record the following alongside the result:
- The date and time of the check.
- The exact prompt text, copied rather than paraphrased.
- Which model was selected: Auto, Pro or Gemini 3.7 Flash.
- The account tier used for the check, since the picker only exists on Google AI Pro and Ultra.
- The query language and the country the check was run from.
- Whether the brand appeared, and the exact URLs cited in the answer.
Two habits follow from that list. The first is to stop comparing results across models as if they were the same measurement: a citation present under Pro in July and absent under 3.7 Flash in September is not a decline, it is two different observations. The second is to keep a fixed model for trend reporting and, if resources allow, sample the others separately. A stable series under one model is more useful than a mixed series that quietly changed method halfway through.
The honest limits of this story
Three limits are worth repeating because they are the parts most likely to be dropped in a summary. The rollout is English only, so Thai-language AI Mode queries are outside it at this stage. It is restricted to Google AI Pro and Ultra subscribers, so free tier users cannot select the model at all. And Google has stated nothing about whether 3.7 Flash will become the default, nor whether it takes part in the automatic Auto routing introduced in November.
None of those gaps should be filled with a guess. A default switch would change the picture considerably, because it would move 3.7 Flash from an option a minority of paying users choose to the model most AI Mode answers come from. Google has not announced that, and there is no reported timeline for it. The same applies to Auto routing: if 3.7 Flash were part of Auto, users who never touch the picker would already be seeing its output, and if it is not, they are not. The source does not resolve it either way.
What this means for Thai marketers
In immediate practical terms, the footprint of this change in Thailand is small. English only plus paid tiers only means most Thai-language searches, and most free tier users, are untouched by the 14 August update. Nobody needs to rewrite a content plan this week because of it.
The part that does carry over is measurement discipline. Thai teams that track English-language queries, for regional buyers, expat audiences or B2B categories where the research happens in English, are already sampling a surface that now has a user-selected variable in it. Recording the model alongside the result costs one column in a spreadsheet and protects every conclusion drawn from that sheet later. Teams working on visibility in AI answer engines and on AI-driven search optimisation should treat model choice as part of the test conditions from now on, in the same way sampling location is already treated. The underlying content work, the structured, citable pages that get pulled into answers in the first place, sits on the same foundations as ordinary organic search work in Thailand and does not change because a menu gained a row.
FAQ: what marketers are asking about the AI Mode picker
Is Gemini 3.7 Flash available in AI Mode for Thai-language queries?
No, not at this stage: Google described the rollout as global but English only, so Thai-language queries are outside it. Google did not say when or whether other languages will be added, and it made no statement about Thailand specifically.
Do I have to pay to get the model picker option?
Yes, the option is limited to Google AI Pro and Google AI Ultra subscribers. Free tier users of AI Mode do not see Gemini 3.7 Flash in the model menu at all, so a check run from a free account is not comparable with one run from a paid account.
Will Gemini 3.7 Flash become the default model in AI Mode?
Google has not said. The 14 August rollout made it a selectable option under the "Gemini 3 models" heading, and no statement has been made about changing the default or about a timeline for doing so.
Does the model a user picks change whether my brand gets cited?
The answer a user sees now depends partly on which model they picked, so a citation observed under one model cannot be assumed to hold under another. Google published no comparison of citation behaviour between Auto, Pro and 3.7 Flash, so there is no measured size for that difference, only the structural fact that the surface is no longer single.
Do I need to change anything on my website because of this?
No, nothing about this rollout asks for a change to a site. What it changes is how AI Mode results should be recorded: note the model used, the language and the account tier next to every check, so a shift later can be attributed to the right cause.
If your team reports on AI Mode visibility and the reports do not currently say which model produced the answer, that is the one thing worth fixing this month. We are happy to look at how your AI answer tracking is set up and where the gaps are.
